You are on page 1of 4

68th WFC - World Foundry Congress PP-53

7th - 10th February, 2008, pp. 563-566

Starting Advantage with Simulation-


Based Intelligent Feeder Design
E. Jacob, M. Manoj and R. Sasikumar
Computational Modeling and Simulation Group,
National Institute for Interdisciplinary Science and Technology (CSIR) Trivandrum, India
Email : liz.csir@gmail.com

ABSTRACT was applied to feeder optimization by Guleyupoglu et al 7


where the case of a semicircular plate fed from a single
Casting simulation codes are validation programs that test a frustum shaped feeder was studied. The volume and surface
given rigging design for defects but cannot generate a design.
area were calculated by geometric formulae and the yield
Hence, the initial design of the rigging system depends heavily
function with penalty terms was maximized by the GA. The
on approximate methods and foundry experience. The objective
here is to create a simulation-based initial design by integrating
synergy of CAD and Genetic Algorithms were used for feeder
the complementary strengths of a rigorous heat transfer analysis optimization 8. Using any standard CAD package, the solid
with practically derived rules for feeder design. The casting model of the casting is created, divided manually into feeding
solidification is first simulated without feeders to identify the sections such that the feeding rules of Campbell are satisfied.
hot spots and get the solidification pattern. A novel pattern The surface area and volume of sections, calculated by CAD,
recognition algorithm divides the casting into clusters each are input to the GA that designs feeders with maximum yield.
cluster being a “feeding section” requiring a separate feeder. Construction of an intelligent expert system for riser design
From the volume and modulus of a feeding section, the size using a database system and optimization tools 6 is an
of the feeder is calculated. For every section, a genetic algorithm attempt to combine simulations with optimization. A
finds a class of feeder dimensions such that the yield is
solidification simulation without risers detects possible
maximized. The design can be iteratively refined using further
shrinkage cavities. A database of cast images aids in the
simulations.
choice of feeders by looking at similarity to stored cases.
Keywords: pattern recognition, clustering, feeding section, The optimization program Visual DOT determines dimensions
genetic algorithm of the riser (in this case cylindrical) such that the volume of
the riser is minimized.
INTRODUCTION We have implemented a strategy for integrating solidification
modeling with optimized design of feeders using the
Casting process design using computer simulation of the
technique of genetic algorithms (GA) such that the
phenomenon of solidification has become an integral part of
conventional rules for feeder design are satisfied.
the design pipeline. Casting solidification simulations using
numerical techniques are generally used as validation METHODOLOGY
programs to test a given rigging system 1-3. The initial design
of feeders and gating is done by the foundryman using his An important feature of the proposed method is that the
experience, domain knowledge and thumb rules developed in simulation code tests a design that was generated by the
the late 50s 4. simulation code itself. Computationally, the casting design
process can be divided into two phases:-
The major considerations in casting design are the quality of
the final product and the yield of the casting both of which Phase I: Identification of Hot Spots and Feeding Sections
depend on the rigging system used 5. The conventional rules
of feeder design ensure a casting with minimum shrinkage 1. The casting simulation is run on a discretised 3D model
defects but result in relatively low yield. Improving both of the casting alone, without feeders which will give the
quality and yield lead to opposing constraints. The feeders solidification pattern in the casting in terms of the
should have sufficient feeding capacity so that they freeze solidification time values at every location (cell or
only after the shape has solidified completely. At the same element) of the casting.
time, over-designed feeders with large safety margins 2. A clustering algorithm works on the solidification data
increase the cost and reduce the yield. This calls for a fresh to divide the casting into feeding sections each requiring
look at optimization in the feeder design stage. a separate feeder. The algorithm locates hotspots which
Depending upon the nature of the design variables, a suitable become seeds for the clusters. The underlying logic is
optimization algorithm can be chosen like quadratic that every section follows the pattern of decreasing
programming, response surface modeling, simulated solidification time towards the outer parts of the
annealing or genetic algorithms 6. A Genetic Algorithm (GA) section.
3. The program computes the volumes and exposed surface M required = Modulus required for the feeder
areas of each feeding section M = Actual modulus of the designed
feeder
4. From 3, the modulus of each section (volume/surface feeder
area) is calculated. Mneck = Modulus of the neck
Phase II : Design of Feeders Mconnection = Modulus of the connection in case
of side feeder
This phase is carried out for every feeding section created
in Phase I. A set of optimal feeders is generated by a single run of the
GA. GA runs by fixing or varying one or more feeder
1. Feeder options are selected. The following feeder dimensions will give more possibilities of yield and different
options have been built into the software :- dimensions. The foundryman can select a solution for the
 Feeder types : top, side or premade final design based on practical considerations.
 Feeder shapes : cylindrical, rectangular, elliptical 4. Placement of feeders Automation of the placement of
 Feeders can tapered or straight feeders depends upon where the shrinkage is located
with respect to the casting. Perpendicular lines are
 Feeders can be solid or hollow drawn vertically or horizontally to find where the cast
 Feeders can have a neck and the mould touch. Sometimes, a feeder cannot be
 Side Feeder in cases where a top feeder cannot be installed in the determined position because of
placed on the casting. Premade feeders are non- interference with the cast or another feeder. Therefore,
standard shapes that can be imported as solid it was felt that the foundryman is the best judge for the
models in STL format and checked for efficacy. positioning of the feeder. The hotspot in the feeding
section is highlighted and by observation, the
1 The dimensions of the feeder are input as a coordinates of the point of attachment of the feeder can
range which is generally a design constraint be either located with the mouse or entered manually.
specific to the casting geometry.
RESULTS
2 The GA evolves a population of possible sets
of feeder dimensions that maximize yield A walkthrough of the whole process of feeder design from
subject to conventional design constraints. simulation till optimization has been carried out on a clamp
casting, a part used in an aeroengine. The material is
Vcasting Aluminium alloy 356 (Al356). The solid model of the casting
Maximize : (Fig. 1) is imported as an STL file from any standard CAD
V feeder + Vcasting
package.
subject to ‘Virtual Casting’, an in-house developed code for
solidification simulation is used. Virtual Casting uses the
i) volume constraint V feeder ≥ Vmin Finite Difference Method to solve the heat diffusion equation
numerically. The casting is simulated without feeders. It is
β discretised into brick elements by the mesh generator of
where Vmin = Vcasting ×
ε −β Virtual Casting. After simulation, the solidification time data
is available for every element in the format (X, Y, Z, stime)
ii) modulus constraint M feeder ≥ M required where (X,Y,Z) is the spatial location of the element and stime
is the solidification time.
M required = mfactor × M section

iii) neck, if present M section < M neck < M feeder


iv) connection for side feeder
Msection < Mconnection < Mfeeder
Vfeeder = volume of the feeder
Vcasting = volume of the feeding section
Vmin = minimum volume required for the
feeder
β = shrinkage factor
ε = feeding efficiency
Msection = Modulus of the feeding section
mfactor = multiplication factor
Fig. 1 : Solid Model of the Casting

Proceedings of 68th World Foundry Congress 564 7th-10th February 2008, Chennai, India
Fig. 2 : Clustering Algorithm in progress: casting divided into
two sections

The clustering module is evoked. Using the solidification


time data, the clustering algorithm identifies the hotspots
and feeding sections (clumps of elements) are grown around
the hotspots (Fig. 2). The program divides the casting into Fig. 3 : Assembly of feeder system from two feeding sections
two feeding sections. Due to the symmetry of the casting,
the feeding sections are identical and hence the volumes and Table 1 shows the dimensions of the feeder corresponding
exposed surface areas calculated are equal, subject to to maximum yield for different choices of design parameters.
numerical approximation due to meshing. So, only one feeder It can be seen from the five design cases that as efficiency
needs to be designed. The volume and surface area are increases, the minimum volume required decreases. Hence
calculated from phase I and constants are input : the feeder designed by the GA is smaller in volume giving
Vcasting = 139284 mm3, higher values of yield.

Surface Area of the section = 24889 mm2 The code is developed in Visual C.Net and runs on a
Msec tion = 5.596 mm, β = 0.07, mfactor = 1.4, Windows platform. The software is GUI-based, guiding the
Mrequired = 7.834 mm user through forms for input of parameters and display of
output data. OpenGL graphics library is used for visualization.
From the standard shapes, the tapered, solid, rectangular
shape was chosen for design. The dimensions for this shape DISCUSSION
are defined by the parameters: base length, base breadth,
height and taper angle. As the feeder is made to fit into the The software process developed for feeder design ideally
top extrusions of the casting, the base dimensions are fixed aims at ‘getting it right the first time’. This is rarely possible
as 31mm and 42 mm. The design variables for the GA will because the predictions will depend on the simulations.
hence be the height and taper angle. Fig. 3 is a visualization When the parameters of simulation are changed, the
of the feeder system designed for the two feeding sections solidification pattern could change leading to different results
and their assembly corresponding to the second design case in feeder design. The asymmetry in the mesh can also affect
in Table 1. Here the efficiency ε = 0.16, the clustering algorithm. What can be assured is that this
simulation-based approach would give a starting advantage
Modulus of feeder designed Mfeeder = 8.618, Yield = 53.1%. and reduce the number of design and simulate cycles.

Table 1
Design Cases with different values of design variables

Case Height Taper Efficiency Height Taper Modulus Yield


limits mm limits (deg) (%) mm (deg) Feeder mm (%)
1 75-85 3-5 0.14 78.3 4.4 9.05 49.8
2 72-80 1-3.5 0.15 75.4 3.2 8.61 53.1
3 72-80 1-3.5 0.16 72.4 2.46 8.61 55.49
4 70-99 0-4 0.14 87.2 3.24 9.43 48.51
5 65-85 0-4 0.17 74.4 0.43 8.45 58.23

7th-10th February 2008, Chennai, India 565 Proceedings of 68th World Foundry Congress
Feeders can be designed for all standard shapes as the REFERENCES
volume and area can be computed geometrically. The volumes 1 Midea A, Nariman R, Yancey B and Faivre T, Using Computer
and surface areas of complicated casting sections can be Modeling to Optimize Casting Processes, Modern Casting, May
computed because they are a collection of well-defined cells. 2000, pp44-47.
This eliminates the interfacing with CAD programs except for 2 Upadhya G and Paul A J, Solidification Modeling : A
the importing of model files in STL format. phenomenological review, AFS transactions, 102.
3 Stefanescu D M, Computer simulation of shrinkage related
Though data from Virtual Casting has been used, the program
defects in metal castings – a review, International Journal of
can be interfaced with solidification data generated from any Cast Metals Research, Vol. 18, No. 3, 2005, pp129-143.
simulation code. 4 Campbell J, Castings, 2nd ed., Oxford: Butterworth-Heinemann,
It would also be interesting to generate the design using one 2005.
simulation code and testing the design with feeders on 5 Guleyupoglu S, Casting Process Design Guidelines, Modern
another simulation code. casting, 1997, Available at http://www.moderncasting.com/
archive/transactions/97083.pdf.
CONCLUSIONS 6 Lim C H, Cho S H, Lee Y C and Choi J K, Basic construction
of intelligent expert system for riser design using database system
The integration of simulations with an optimization technique and optimization tools, International Journal of Cast Metals
for the design of feeders has been achieved without the use Research, Vol. 18, No. 4, 2005, pp195-201.
of any external optimization software. Even if there is no 7 Guleyupoglu S, Upadhya G, Paul A J, Yu K O and Hill J L, Casting
optimization needed, a given design with absolute values for Riser Design Optimization Using Genetic Algorithms, Proceedings
dimensions can be checked for sufficiency. of the 7th Conference on Modeling of Casting, Welding and
Advanced Solidification Processes, Minerals, Metals and
The attempt to bring together the knowledge from Materials (TMS), 1995, pp389-396.
solidification data, an optimization technique for determining 8 Jacob E, Sasikumar R, Praveen B and Gopalakrishna V, Intelligent
feeder dimensions while satisfying the standard design Design of Feeders for Castings by Augmenting CAD with Genetic
constraints of modulus and minimum volume provides a Algorithms, Journal of Intelligent Manufacturing, 15, 2004,
cutting-edge tool for the design of feeding systems for the pp299-305.
production of cast components.
ACKNOWLEDGEMENTS
This work is financially supported by AR&DB (Aeronautical
Research and Development Agency) of India.

Proceedings of 68th World Foundry Congress 566 7th-10th February 2008, Chennai, India

You might also like