Вы находитесь на странице: 1из 20

COMMUNICATIONS IN NUMERICAL METHODS IN ENGINEERING

Commun. Numer. Meth. Engng 2008; 24:1723–1740


Published online 29 October 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/cnm.1062

A fast method for numerical simulation of


casting solidification

Rohallah Tavakoli∗, † and Parviz Davami


Department of Material Science and Engineering, Sharif University of Technology, P.O. Box 11365-9466,
Tehran, Iran

SUMMARY
An efficient numerical method for simulation of casting solidification is presented. The presented method
is based on a stable explicit finite difference solution of the heat equation in conjunction with the
temperature recovery method to incorporate latent heat effect. The computational cost of the presented
method is approximately the same as an explicit method (per time step), while it is free from the time step
limitation due to the stability criterion. A simple domain decomposition method is included to improve
the computational performance of the presented method. The efficiency, stability and accuracy of the
presented method are supported with illustrative examples. Copyright q 2007 John Wiley & Sons, Ltd.

Received 23 April 2007; Revised 10 September 2007; Accepted 20 September 2007

KEY WORDS: casting simulation; phase change; solidification; stable explicit; Stefan problem; temper-
ature recovery

1. INTRODUCTION

The ability of fluid to assume the shape of its container is exploited by casting processes, which
involve melting and pouring liquid metal into a sand or metal mold and allowing it to solidify,
yielding the shape close to that of the desired product. Metal casting continues to be the preferred
process for intricate shapes of any size and weight with varying wall thicknesses and internal
features [1]. Solidification of molten metal after being poured into a mold cavity is an important
phase in the casting process, which greatly influences the product quality and yield. When molten

∗ Correspondence to: Rohallah Tavakoli, Department of Material Science and Engineering, Sharif University of
Technology, P.O. Box 11365-9466, Tehran, Iran.

E-mail: tav@mehr.sharif.edu, rohtav@gmail.com

Copyright q 2007 John Wiley & Sons, Ltd.


1724 R. TAVAKOLI AND P. DAVAMI

metal enters a mold cavity, its heat is transferred through the mold wall. In the case of pure metals
and eutectics (or generally narrow band solidification), the solidification proceeds layer by layer,
starting from the mold wall and proceeding inwards. The moving isothermal interface between
the liquid and the solid region is called the solidification front. As the front solidifies, it contracts
in volume and draws molten metal from the adjacent liquid layer. When the solidification front
reaches the central hot spot, there is no more liquid metal left and shrinkage cavity is formed.
This is avoided by attaching a feeder designed to solidify later than the hot spot. The shrinkage
cavity shifts to the feeder, which is cut off after casting solidification and recycled [1]. Also, the
total volume of feeders should be minimized to improve the yield and productivity.
In recent years, numerical modeling of casting solidification has received increasing attention
because of its potential to improve the productivity of the metal-casting industry by reducing the
cost and time associated with the traditional, experimentally based, design of castings [2].
The casting modeling systems, which are available for foundry users today, are software programs
which accept a user’s design and then analyze the design to predict the likelihood of defects. Once
an analysis has been completed, the user views the results of the analysis, and if an area of potential
defect is found within the casting, then the user needs to make some logical modifications in design
and repeat the simulation until the desired result is obtained. Thus, the traditional trial-and-error
design cycle on the foundry floor has been replaced with trial-and-error on the computer [3]. But
simulation packages are often too tricky to use and need expertness in computer solid modeling as
well as the feeding design principles. Furthermore, the resulted design is not essentially optimal
and its quality is a function of the user expertness and patience.
One of the major problems during a computer-aided casting design cycle is the time taken to
perform the numerical simulation to predict the solidification patterns and hot spots locations. Very
often, this design–simulation–evaluation cycle is repeated many times before a final satisfactory
design is obtained. This iterative cycle can be extremely time consuming because the simulation
time required for the re-evaluation is significantly large [4].
There are generally two approaches for numerical solution of solidification problems, implicit
and explicit schemes. The former has no restriction on its time stepping. However, in each time
step one has to solve a global system of non-linear equations. The solution of such a system of
equations usually takes high computational cost and memory. Also, implementation of implicit
methods is usually a difficult task. The latter has local nature and is easy to implement. However,
it suffers the severely restricted time step from stability requirement. It seems that in spite of
large time increment, the computational efficiency of implicit methods is not essentially better
than explicit methods for solidification modeling. In [5] this issue has been discussed in detail. In
[5] we present an unconditionally stable fully explicit finite difference method for the simulation
of 2D solidification problems on simple geometries. The computational cost of this method is the
same as an explicit method per time step, while it is free from the time step limitation due to the
stability criterion.
In the present study the stable explicit phase change solver is extended to simulate casting
solidification on complicated 3D geometries. The efficiency of the presented method is improved
by applying a simple domain decomposition method, same as Reference [6]. In this approach,
instead of using a single time step throughout the computational domain, the objective is to group
the elements according to the maximum allowable time step size (or reasonable time step size
for stable method) and advance the solution independently within each group. Time accuracy of
the procedure is maintained by appropriate interchange of information across the boundaries of
groups.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
NUMERICAL SIMULATION OF CASTING SOLIDIFICATION 1725

2. MATHEMATICAL MODELING

From a macroscopic point of view, if the effect of melt flow during solidification is neglected,
solidification is governed by the heat conduction equation as follows:

*
c = ∇ ·(k∇) (1)
*t
where  is the temperature field,  is the density, c is the specific heat and k is the thermal
conductivity. The initial condition is

(t = 0) = 0 (2)

where 0 is the pouring temperature of liquid metal. The natural boundary condition is applied at
the mold–air interface
*
k = h ∞ (∞ −) (3)
*n
where h ∞ is the mold–air convective heat transfer coefficient, ∞ is the ambient temperature and
n is the normal outward unit vector to the interface. The cast–mold interface is modeled by the
interfacial heat transfer coefficient method. In this method the geometry, including the casting and
mold, is separated by an interface boundary surface that is subdivided into a number of segments
that represent different interface conditions. Each boundary segment is identified by two interface
segments; a segment that belongs to the casting region and a corresponding segment associated
with the mold region. These two segments are actually lying on each other and represent the same
boundary. For each boundary segment in the casting, the corresponding boundary segment in the
mold acts as an ambient condition and vice versa. This means that the heat transfer between these
two surfaces is characterized by their temperatures and the local heat transfer coefficient specified
for that segment of the interface:
 
*cast  *mold 
−k =k  = h i (cast −mold )|interface (4)
*n interface *n 
interface

where h i is the local interfacial heat transfer coefficient which is extracted from the material
database. It can be fixed or a function of time and temperature. The concept of interfacial convective
heat transfer can easily be applied to any kind of contact surface in casting configuration. This
might be a surface between a chill and casting or mold and a surface separating two parts of a
mold. In this way, virtually all contact surfaces in the casting system can be conveniently modeled
[7].
There are several methods to incorporate the phase change effect in the heat conduction equation
(for good survey, see [8, 9]). Following [5], the temperature recovery method [10, 11] is used in this
study. This method consists of transforming the latent heat into an equivalent number of degrees
by division of latent heat by the specific heat.
The temperature field is computed in the absence of latent heat based on (1), and after each
time step the temperature of any point that falls below the freezing point is reset to the freezing
point until the solidification is over. The liberated latent heat, Hf , is evaluated as a function of

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
1726 R. TAVAKOLI AND P. DAVAMI

corresponding solid fraction increment,  f s , over a time step, t, as follows [11]:

Hf = cVi  = Hf Vi  f s (5)

where Vi is the volume of the finite difference cell considered,  is the temperature drop from
the freezing point over a time step and Hf is the latent heat of fusion. The current solid fraction is


n
fs = f (6)
l=1

where n is the number of elapsed time steps. This procedure may be repeated until the solidification
is over at f s = 1. This method works only for isothermal solidifications. However, with a simple
modification, a more generalized temperature recovery method can be developed for non-isothermal
phase changes [10, 11].
Consider a cell for which the temperature is in the solidification range. Suppose 1 is the
temperature at the start of time step, 2 is the calculated temperature without considering the effect
of latent heat release and 2 is the actual temperature, then 2 can be obtained from the following
relation:
 2  2  2 * fs
c d = c d+ Hf d (7)
1 1 1 *

If 1 is greater than the liquidus temperature, liq , 1 in the lower bound of the second integral
in the right-hand side of (7) should be replaced with liq . Also, if 2 is fallen below the solidus
temperature, sol , after solution of (7), 2 in the upper bound of the second integral in the right-hand
side of (7) should be replaced with sol and (7) should be solved again.
The term * f s /* in (7) can be computed from the relationship between the solid fraction
and temperature. There are several models that are generally used for evaluating the relationship
between the solid fraction and temperature in practical solidification analysis, such as lever rule
model (equilibrium solidification model), Scheil model (complete mixing of solute in the liquid
and no mixing in the solid), Brody–Flemings model (complete mixing of solute in the liquid and
some mixing in the solid) and linear model [12]. The latter approach is used in the present study. It
is useful when there is no explicit relationship between the solid fraction and temperature or when
the solid fraction cannot be easily evaluated as a function of temperature, since the phase diagrams
are not known (usually happen for multi-component alloys). In this approach it is assumed that
the latent heat of freezing is distributed linearly over the solidification interval; therefore, the solid
fraction and temperature have a linear relation as follows:

liq −
fs = (8)
liq −sol

Hence

* fs 1
= (9)
*  liq −sol

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
NUMERICAL SIMULATION OF CASTING SOLIDIFICATION 1727

3. NUMERICAL METHOD

An efficient numerical solution of (1) is a key to develop a practical simulation tool. In [13],
Saul’yev has introduced two unconditionally stable fully explicit asymmetric schemes for the
solution of a 1D diffusion equation. In [5] authors adapted Saul’yev’s method to simulate phase
change problems over a simple 2D domain. In this section we extend this method for numerical
solution of practical casting solidification. For this purpose we extend the method to 3D and include
variation of physical properties and effect of cast–mold interface in numerical method. Also, a
domain decomposition strategy is included to increase the computational efficiency.
Suppose that the spatial domain (mold-box) is divided into Imax , Jmax and K max uniform
Cartesian grids with spatial step size x in the x, y and z directions, respectively, and the
temporal domain is divided into N grids with time step t. The grid points (xi , y j , z k , tn ) are
given by (ix, jx, kx, nt). We use i,n j,k to denote the finite difference approximations of
(ix, jx, kx, nt). The finite difference approximation of Equation (1) with Saul’yev’s down-
stream method has the following form:

smat
i,n+1
j,k = n+1 +i,n+1 n+1
j−1,k +i, j,k−1
1+3smat i−1, j,k
  
1
+ −3 i,n j,k +i+1,
n
j,k +n
i, j+1,k +n
i, j,k+1 (10)
smat

where smat = (kmat t)/(mat cmat x 2 ) and the subscript ‘mat’ denotes the material type such as
cast, mold, chill, etc. In (10) it is assumed that the physical properties are temperature invariance.
For the case that the physical properties are temperature dependence, we neglect the variation
of physical properties during each time step and use from their previous updated values. In this
manner Equation (10) has the following form:

  −1
i,n+1
j,k = si, j,k [1+si, j,k (ki−1/2, j,k +ki, j−1/2,k +ki, j,k−1/2 )]
n n n


n+1 n+1 n+1
× ki−1/2,
n
j,k i−1, j,k +ki, j−1/2,k i, j−1,k +ki, j,k−1/2 i, j,k−1
n n


1
+ −(ki+1/2,
n
j,k +ki, j+1/2,k +ki, j,k+1/2 )
n n
i,n j,k
si, j,k

n n n
+ ki+1/2,
n
j,k i+1, j,k +ki, j+1/2,k i, j+1,k +ki, j,k+1/2 i, j,k+1
n n
(11)

where si, j,k = t/(i, j,k ci, j,k x 2 ) and face conductivity value such as ki+1/2,
n
j,k can be computed
with the geometric averaging as follows [14]:
 −1
0.5 0.5
j,k = + n
n
ki+1/2, n (12)
ki, j,k ki+1, j,k

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
1728 R. TAVAKOLI AND P. DAVAMI

Better accuracy and error distribution are achieved by alternating the sweep direction according
to the previous work [5]. In this manner when we move from a lower index, e.g. i = 1, to an
upper index, e.g. i = Imax , we use the Saul’yev downstream scheme (in the x direction) and after
a complete line sweep, we alternate sweep direction and move from the upper index to the lower
index and use the Saul’yev upstream scheme (in the x direction). Generally when we sweep from a
lower index to an upper index for each spatial direction, we use the Saul’yev downstream scheme
for that direction and vice versa. Applying the above sweeping strategy to Equation (10) yields
the following form:


smat
i,n+1
j,k = n+1 +i,n+1 n+1
j− j swp,k +i, j,k−k swp
1+3smat i−i swp, j,k
  
1 n n n n
+ −3 i, j,k +i+i swp, j,k +i, j+ j swp,k +i, j,k+k swp (13)
smat

where i swp, j swp and k swp are indicators of the sweep direction. These parameters have a value
of 1 when the sweep direction is from a lower to a higher index, and a value of −1 when the sweep
direction is from a higher to a lower index. Application of this sweeping strategy for Equation
(11) is straightforward.
Since the cast–mold and other interfaces are modeled with some convective boundary conditions
in the current study, (13) should be modified to incorporate such boundaries. For this purpose, we
rewrite the energy equation in the following discretized form:

i,n+1 n
j,k −i, j,k n n n
cVi, j,k = qi+1/2, j,k −qi−1/2, j,k +qi, j+1/2,k
i+1/2, j,k i−1/2, j,k i, j+1/2,k
t
n n n
−qi,i,j−1/2,k
j−1/2,k
+qi,i,j,k+1/2
j,k+1/2
−qi,i,j,k−1/2
j,k−1/2
(14)

where Vi, j,k is the volume of finite difference cell centered at (i, j, k) and q is the face heat flux,
e.g. for face (i +1/2, j, k) we have


n i+1/2, 
n i+1/2,

n i+1/2, i+1, j,k
j,k
−i, j,k
j,k

qi+1/2, j,k = (15)


j,k x

where superscript n  indicates implicitness or explicitness of the face flux which is dependent on
the sweeping direction, e.g. for the x-direction we have


n i±1/2, j,k = n +max(∓i swp, 0) (16)

To include the effects of interfacial boundary conditions in (14), it is sufficient to replace the
corresponding conductive heat flux (related to the interface) with desired convective heat flux.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
NUMERICAL SIMULATION OF CASTING SOLIDIFICATION 1729

As an example for convective face (i +1/2, j, k),


n n n
j,k = h i [(amb )i, j,k −i,i+1/2, ]
i+1/2, j,k i+1/2, j,k j,k
qi+1/2, j,k (17)

n
where (amb )i,i+1/2,
j,k
j,k
is the ambient temperature related to face i +1/2, j, k, which is computed
with averaging between its two adjacent cells as follows:
n n n
(amb )i,i+1/2,
j,k
j,k
= 0.5(i+1, j,k +i, j,k
i+1/2, j,k i+1/2, j,k
) (18)

As discussed in [5], there are several approaches to increase/control the accuracy and stability
of the presented method for brevity of this communication we refer the interested readers to the
above-mentioned reference.

3.1. Time stepping strategy


Although applied numerical approximation is unconditionally stable and selection of time step is
arbitrary, to balance between accuracy and efficiency, we compute the time step from the following
equation:

t = texp (19)

where (user defined) parameter  controls solution accuracy and texp is the allowable time step of
the traditional explicit finite difference method which is computed from the following equations:
 
mat cmat x 2
texp = (20)
6kmat min

for temperature invariant physical properties and




i, j,k ci, j,k x 2
texp = (21)
6ki, j,k
min

Figure 1. Alternating sweeping direction strategy used in the present study: four consecutive time steps
on a 3×3 array of finite difference cells.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
1730 R. TAVAKOLI AND P. DAVAMI

Figure 2. A diagram of the hammer casting illustrating the dimension of the rigging and risering
(all dimensions are in inches) [16].

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
NUMERICAL SIMULATION OF CASTING SOLIDIFICATION 1731

Figure 3. Three views of the hammer casting showing the location of thermocouples
(all dimensions are in inches) [16].

for temperature dependence physical properties. The Fourier number (Fo, ratio of the heat conduc-
tion rate to the rate of thermal energy storage in a solid) is defined as

kt
Fo = (22)
cx 2

The stability bound of the explicit time integration is Fo 16 . Note that when we have convective
boundaries, contribution of Biot number (Bi = h/kx, ratio of the internal thermal resistance of
a solid to the boundary layer thermal resistance) should be included in the stability criterion.
In this manner stability bound of a 1D heat equation with the convective boundary condition is
Fo(1+ Bi) 12 (instead of Fo 12 ). Since in casting simulation Bi  1 (particulary for sand-mold
casting), the effect of Biot number is neglected in the current study.
Although the stability of solution does not impose any limitation for the selection of , it can
be said that by decreasing the phase change range (particularly for iso-thermal phase change), the
smaller value of  should be applied, and by increasing the value of Hf /c,  should be decreased
to achieve reasonable accuracy (as shown in [5] via numerical experiments).

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
1732 R. TAVAKOLI AND P. DAVAMI

Figure 4. Cooling curve measured in the steel hammer casting (thermocouples 1–5) and sand
mold (thermocouples 6 and 7) [16].

3.2. Improving time stepping efficiency


Since in the casting process, particulary for the sand-casting process, we have different regions
(cast, mold, core, etc.) in the casting system and usually we have considerable difference between
physical properties, it is possible to use the domain decomposition concept to increase the efficiency
of computation. In this approach, instead of using a single time step throughout the computational
domain, the objective is to treat time stepping of each region according to its local allowable time
step and advance the solution independently within each medium (same as Reference [6]). Since
in the present study the interfaces between different parts of the casting system are treated as
convective boundary conditions, interchange of information across the boundaries and time accu-
racy of the procedure are automatically maintained. The implementation of the method proceeds
algorithmically as follows:
• Compute the allowable time step of each region from Equation (19). The parameter  in
Equation (19) is selected so that the relation (or inverse of the relation) of allowable time
step of each two regions becomes an even number.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
NUMERICAL SIMULATION OF CASTING SOLIDIFICATION 1733

Table I. Physical properties and initial and boundary conditions used in the
analysis of steel hammer casting in the present study (adapted from [17, 18]).
Casting Sand mold Core

Density (kg/m3 ) 7300 1500 2960


Heat capacity (J/kg ◦ C) 627 1128.6 543.4
Heat conductivity (W/m ◦ C) 33.44 0.7 0.8
Latent heat (J/kg) 2.717×105 — —
Initial temperature (◦ C) 1621 20 20

Tliq (◦ C) 1488
Tsol (◦ C) 1439

Interfacial heat transfer coefficient (W/m2 ◦ C)


Casting–mold 418.0
Casting–core 418.0
Casting–air 75.0
Mold–core 418.0
Mold–air 75.0
Radiation heat transfer was ignored.

Table II. Elapsed CPU time (in minute) of simulation related to steel
hammer casting benchmark in the present study.

Case Fo = 16 Fo = 26 Fo = 46 Fo = 1

Without domain decomposition


Grid 1 2.33 1.16 0.58 0.39
Grid 2 22.78 11.39 5.69 3.80
Grid 3 131.65 65.81 32.89 21.98

With domain decomposition


Grid 1 0.70 0.35 0.18 0.12
Grid 2 6.87 3.43 1.72 1.15
Grid 3 40.96 20.27 10.16 6.76

• Determine the minimum time step size (tmin ) and sort the regions based on their allowable
time steps (ti ), e.g.

Region ti
1 t1 = tmin
2 t2 = 2tmin
3 t3 = 4tmin

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
1734 R. TAVAKOLI AND P. DAVAMI

1800 1800
#1 #1
#2 #2
1600 #3 1600 #3
#4 #4
#5 #5
1400 #6 1400 #6
TEMPERATURE (C)

TEMPERATURE (C)
#7 #7
1200 1200

1000 1000

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a) TIME (MINUTE) (b) TIME (MINUTE)

1800 1800
#1 #1
#2 #2
1600 #3 1600 #3
#4 #4
#5 #5
1400 #6 1400 #6
TEMPERATURE (C)

TEMPERATURE (C)

#7 #7
1200 1200

1000 1000

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(c) TIME (MINUTE) (d) TIME (MINUTE)

Figure 5. Hammer casting benchmark: cooling curve of thermocouple positions P1–P7 that are calculated
on grid 1 with various Fourier numbers. (a) Fo = 1/6; (b) Fo = 2/6; (c) Fo = 4/6 and (d) Fo = 1.

• Advance the solution one global time step, e.g. with three regions the solution is advanced
through 4tmin by the following sequence:

Region Number of time steps


1 2
2 1
1 2
2 1
3 1

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
NUMERICAL SIMULATION OF CASTING SOLIDIFICATION 1735

1800 1800
#1 #1
#2 #2
1600 #3 1600 #3
#4 #4
#5 #5
1400 #6 1400 #6
TEMPERATURE (C)

TEMPERATURE (C)
#7 #7
1200 1200

1000 1000

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a) TIME (MINUTE) (b) TIME (MINUTE)

1800 1800
#1 #1
#2 #2
1600 #3 1600 #3
#4 #4
#5 #5
1400 #6 1400 #6
TEMPERATURE (C)

TEMPERATURE (C)

#7 #7
1200 1200

1000 1000

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(c) TIME (MINUTE) (d) TIME (MINUTE)

Figure 6. Hammer casting benchmark: cooling curve of thermocouple positions P1–P7 that are calculated
on grid 2 with various Fourier numbers. (a) Fo = 1/6; (b) Fo = 2/6; (c) Fo = 4/6 and (d) Fo = 1.

Note that for all (convective) interfacial boundaries the last available value of its ambient tempera-
ture is used during time stepping. Since in the sand-mold casting the allowable time step (consider
the explicit method) of cast and sand differs by more than one order of magnitude, also we
usually have large number of finite difference cells in the sand region, efficiency of simulation is
considerably increased with the application of this domain decomposition method.

4. NUMERICAL EXPERIMENTS

In this section we show the performance of the presented method on some test cases. As the
computing platform we have used a personal computer with an AMD 2.41 GHz CPU and 2 GB

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
1736 R. TAVAKOLI AND P. DAVAMI

1800 1800
#1 #1
#2 #2
1600 #3 1600 #3
#4 #4
#5 #5
1400 #6 1400 #6
TEMPERATURE (C)

TEMPERATURE (C)
#7 #7
1200 1200

1000 1000

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a) TIME (MINUTE) (b) TIME (MINUTE)

1800 1800
#1 #1
#2 #2
1600 #3 1600 #3
#4 #4
#5 #5
1400 #6 1400 #6
TEMPERATURE (C)

TEMPERATURE (C)

#7 #7
1200 1200

1000 1000

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(c) TIME (MINUTE) (d) TIME (MINUTE)

Figure 7. Hammer casting benchmark: cooling curve of thermocouple positions P1–P7 that are calculated
on grid 3 with various Fourier numbers. (a) Fo = 1/6; (b) Fo = 2/6; (c) Fo = 4/6 and (d) Fo = 1.

RAM. The opensource software package CartGen [15] was used for grid generation purpose in
the present study.

4.1. Steel hammer casting


In this example, we consider the solidification of a steel hammer casting used in a 1988 conference
[16] as a benchmark problem to study the accuracy of solidification analysis. All of the simulations
presented at the conference produced approximately similar results. Each predicted the appear-
ance of macro-porosity in the hammer body due to the formation of an isolated hot spot during
solidification. The predicted porosity was also observed in the test casting. The dimensions of the
pattern used for casting are shown in Figure 1. The mold and core were hand-rammed silica and
zircon sand, respectively. The casting was poured in 8740 (modified) steel. Seven thermocouples
were used to measure cooling curves in the casting and mold. All thermocouples were located on

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
NUMERICAL SIMULATION OF CASTING SOLIDIFICATION 1737

1800 1800
#1 #1
#2 #2
1600 #3 1600 #3
#4 #4
#5 #5
1400 #6 1400 #6

TEMPERATURE (C)
TEMPERATURE (C)

#7 #7
1200 1200

1000 1000

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(a) TIME (MINUTE) (b) TIME (MINUTE)

1800 1800
#1 #1
#2 #2
1600 #3 1600 #3
#4 #4
#5 #5
1400 1400
TEMPERATURE (C)
#6 #6
TEMPERATURE (C)

#7 #7
1200 1200

1000 1000

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
(c) TIME (MINUTE) (d) TIME (MINUTE)

Figure 8. Hammer casting benchmark: cooling curve of thermocouple positions P1–P7 that are calculated
on grid 3 with various Fourier numbers in conjunction with the proposed domain decomposition method.
(a) Fo = 1/6; (b) Fo = 2/6; (c) Fo = 4/6 and (d) Fo = 1.

the parting plane. The sensing junctions of the thermocouples were located as shown in Figures
2 and 3. Thermal histories were recorded for 90 min at the indicated locations in the casting and
mold and are shown in Figure 4.
The numerical analysis was carried out with different grid resolutions (three resolutions)
and Fourier numbers (Fo = 16 , 26 , 46 , 1) for 90 min from the initial condition. The total number
of meshes, number of cast meshes and number of mold (sand mold and core) meshes are
(524 392, 43 769, 480 623), (2 090 560, 178 336, 1 912 224) and (6 133 180, 532 489, 5 600 691) for
grid resolutions 1, 2 and 3, respectively. The physical properties and initial and boundary conditions
used in our analysis are presented in Table I.
Figures 5, 6 and 7 show the temperature history of thermocouple positions P1–P7 that are
computed with various Fourier numbers on grids 1, 2 and 3, respectively. The plots show good

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
1738 R. TAVAKOLI AND P. DAVAMI

Figure 9. Three-dimensional configuration of: steel motor-shell (left), cast-iron engine-block (middle) and
cast-iron heat exchanger (right) in the present study.

agreement between computed temperature history and experimental results. It seems that increasing
Fourier number does not have a considerable effect on the computed temperature history. Figure 8
shows the temperature history of thermocouple positions P1–P7 that are computed on grid 3 with
various Fourier numbers in conjunction with the proposed domain decomposition method. In this
experiment the energy equation was solved every 10 time steps in the sand medium. The plot
shows that the application of the proposed domain decomposition method has no sensible effect
on the computed temperature history of cast and sand media. Plate 1 shows the contour plot of
the local solidification time of the cast medium that is computed on grid 2 with various Fourier
numbers with and without application of the proposed domain decomposition method. Table II
shows the elapsed CPU times related to this numerical experiment. It is clear that the simulation
time is considerably reduced by the application of the presented stable explicit phase change solver
in conjunction with the proposed domain decomposition strategy; hence, numerical simulation of
casting solidification on a sufficiently fine grid is performed in a few minutes on a traditional
desktop computer.

4.2. Some industrial-casting parts


In this numerical experiment we show the efficiency of the presented method to simulate solidifi-
cation of some complex industrial-casting parts. For this purpose, we select three complex casting
parts: (1) steel motor shell, (2) cast-iron engine block and (3) cast-iron heat exchanger. The mold is
the silica sand in all cases. Figure 9 shows the 3D configuration of these castings. Grid generation
for these casting parts are performed so that the total number of meshes, number of cast meshes
and number of mold meshes are (524 392, 43 769, 480 623), (2 090 560, 178 336, 1 912 224) and
(41 457 420, 3 337 531, 38 119 889) for cases 1, 2 and 3, respectively. Table III shows the physical
properties and initial condition related to these simulations. Other properties and parameters are
the same as Table I. We have used Fourier number 1 for test cases 1 and 2 and Fourier number
2
6 for test case 3. For domain decomposition purpose we solve the energy equation every 10 time
steps in the sand medium.
Plate 2 shows temperature contour at a selected section of the mentioned casting parts after
complete solidification. The elapsed CPU times related to these simulations were 2.9, 3.01 and
28.7 min for cases 1, 2 and 3, respectively. It is clear that the proposed method is very efficient,
and solidification analysis of the complex casting parts with sufficiently fine meshes is performed
in a few minutes on a traditional desktop computer.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
Plate 1. Hammer casting benchmark: contour plot of local solidification time (in second) of cast medium
that is calculated on grid 2 with Fourier number 16 (up) and 1 (down), with (right) and without (left) the
application of the proposed domain decomposition method.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24(12)
DOI: 10.1002/cnm
Plate 2. Temperature contours after finalization of solidification for mentioned
industrial-casting parts in the present study.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24(12)
DOI: 10.1002/cnm
NUMERICAL SIMULATION OF CASTING SOLIDIFICATION 1739

Table III. Physical properties and initial condition related to the simulation
of industrial casting parts in the present study.
Steel Cast-iron

Density (kg/m3 ) 7440.0 7100.0


Heat capacity (J/kg ◦ C) 627.0 800.0
Heat conductivity (W/m ◦ C) 38.0 40.0
Latent heat (J/kg) 290 000.0 209 200.0
Initial temperature (◦ C) 1600.0 1370.0
Tliq (◦ C) 1510.0 1230.0
Tsol (◦ C) 1460.0 1150.0

5. CONCLUSIONS

Stable explicit finite difference method for the simulation of casting solidification is presented.
Its stability, accuracy and efficiency are shown on several test cases. The evaluated temperature
field for large values of Fourier numbers (large time step) has good agreement with the results of
small Fourier numbers. Combination of the presented method with a simple domain decomposition
strategy improves the performance of computation without sensible loss of accuracy. Success of the
presented method for efficient implementation of the computer-aided feeder design optimization
is supported by the recent works due to Tavakoli and Davami [3, 19, 20].

ACKNOWLEDGEMENTS
The authors would like to thank Mr. Moravvej from the engineering & design division of Razi Metallurgical
Research Center (RMRC) for his collaboration in computer solid modeling in the present study.

REFERENCES
1. Ravi B. Metal Casting: Computer Aided Design and Analysis. Prentice-Hall: India, 2005.
2. Gafur MA, Haque MN, Prabhu KN. Effect of chill thickness and superheat on casting/chill interfacial heat
transfer during solidification of commercially pure aluminium. Journal of Materials Processing Technology 2003;
133(3):257–265.
3. Tavakoli R, Davami P. Optimal riser design in sand casting process with evolutionary topology optimization.
Structural and Multidisciplinary Optimization, 2007, submitted.
4. Pao WK, Ransing RS, Lewis RW. A medial-axes-based interpolation method for solidification simulation. Finite
Elements in Analysis and Design 2004; 40:577–593.
5. Tavakoli R, Davami P. Unconditionally stable fully explicit finite difference solution of solidification problems.
Metallurgical and Materials Transactions B 2007; 38(1):121–142.
6. Löhner R, Morgan K, Zienkiewicz OC. The use of domain splitting with an explicit hyperbolic solver. Computer
Methods in Applied Mechanics and Engineering 1984; 45(1–3):313–329.
7. Manzari MT, Gethin DT, Lewis RW. Optimisation of heat transfer between casting and mould. International
Journal of Cast Metals Research 2000; 13(4):199–206.
8. Huy H, Argyropoulosz SA. Mathematical modelling of solidification and melting: a review. Modelling and
Simulation in Materials Science and Engineering 1996; 4:371–396.
9. Lewis RW, Ravindran K. Finite element simulation of metal casting. International Journal for Numerical Methods
in Engineering 2000; 47:29–59.
10. Hong CP, Umeda T, Kimura Y. Numerical models for casting solidification: Part I. The coupling of the boundary
element and finite difference methods for solidification problems. Metallurgical Transactions 1984; 15B(1):91–99.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm
1740 R. TAVAKOLI AND P. DAVAMI

11. Hong CP. Computer Modelling of Heat and Fluid Flow in Materials Processing. CRC Press: Boca Raton, FL,
2004.
12. Kurz W, Fisher DJ. Fundamentals of Solidification (4th edn). Trans Tech Pub.: Switzerland, 1998.
13. Saul’yev VK. Integration of Equations of Parabolic Type Equation by the Method of Net. Pergamon Press: New
York, 1964.
14. Voller VR. Numerical treatment of rapidly changing and discontinuous conductivities. International Journal of
Heat and Mass Transfer 2001; 44:4553–4556.
15. Tavakoli R. CartGen: robust, efficient and easy to implement uniform/octree/embedded boundary Cartesian grid
generator. International Journal for Numerical Methods in Fluids 2007; DOI: 10.1002/fld.1685.
16. George ES, Elder SP, Abbaschian GJ. Thermal profiles measured during casting of a steel hammer. In Modeling
of Casting and Welding Process IV, Giamei AF, Abbaschian GJ (eds). TMS/AIME: Warrendale, PA, 1988;
775–788.
17. Nakagawa T. Three-dimensional simulation of a hammer casting using the finite element method. In Modeling
of Casting and Welding Process IV, Giamei AF, Abbaschian GJ (eds). TMS/AIME: Warrendale, PA, 1988;
833–838.
18. Zhu JD, Ohnaka I. Three dimensional analysis of a hammer casting by direct finite difference method. In
Modeling of Casting and Welding Process IV, Giamei AF, Abbaschian GJ (eds). TMS/AIME: Warrendale, PA,
1988; 839–844.
19. Tavakoli R, Davami P. Automatic optimal feeder design in steel casting process. Computer Methods in Applied
Mechanics and Engineering, 2007; DOI: 10.1016/j.cma.2007.09.018.
20. Tavakoli R, Davami P. Feeder growth: a new method for automatic optimal feeder design in gravity casting
processes. Structural and Multidisciplinary Optimization, 2007, submitted.

Copyright q 2007 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng 2008; 24:1723–1740
DOI: 10.1002/cnm

Вам также может понравиться