Available online at www.sciencedirect.

com
Finite Elements in Analysis and Design 40 (2004) 645663
www.elsevier.com/locate/nel
Combined adaptive meshing technique and segregated nite
element algorithm for analysis of free and forced convection
heat transfer
Niphon Wansophark, Pramote Dechaumphai

Department of Mechanical Engineering, Faculty of Engineering, Chulalongkorn University, Patumwan,

Bangkok 10330, Thailand
Received 26 October 2002; accepted 28 April 2003
Abstract
A combined algorithm between a segregated nite element method and a monotone streamline upwinding
method for solving two-dimensional viscous incompressible thermal ows is presented. The nite element
equations are derived from a set of coupled nonlinear NavierStokes equations that consists of the conserva-
tion of mass, momentums and energy. The method uses the three-node triangular element with equal-order
interpolations for all the variables of the velocity components, the pressure and the temperature. The eciency
of the developed nite element formulation was evaluated by solving natural convection examples of the ther-
mally driven ows in a square cavity and in concentric cylinders. The formulation was further evaluated
by solving a forced convection example of a ow past a heated cylinder. In addition, an adaptive meshing
technique is also implemented to further increase the solution accuracy, and at the same time, to minimize
the computational time and computer memory requirement.
? 2003 Elsevier B.V. All rights reserved.
Keywords: Adaptive meshing technique; Segregated algorithm; Finite element method; Convection heat transfer
1. Introduction
Convection heat transfer is one of the most challenging problems for computational methods due
to its inherent coupling between the governing equations of the uid motion and the energy equation.
This coupling eect can be seen noticeably at high Rayleigh numbers in free convection problem
and at high Reynolds numbers in forced convection problem. Another main reason which increases

Corresponding author. Tel.: +66-2-218-6621; fax: +66-2-218-6621.

E-mail address: fmepdc@eng.chula.ac.th (P. Dechaumphai).
0168-874X/$ - see front matter ? 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0168-874X(03)00101-X
646 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
the diculty for solving the convection heat transfer problems is the non-linear phenomenon of the
convection terms in both the momentum equations and the energy equation. One successful technique
for solving such couple partial dierential equations is known as the SIMPLER (semi-implicit method
for pressure-linked equations revised) algorithm [1] which computes the primitive variables, i.e.,
the velocity components, the pressure and the temperature, separately. The algorithm, originally
devised for nite dierence method, has been modied and proposed by several researchers for
the nite element method [24]. In computing the convection terms, a widely used technique is
known as streamline upwinding nite element method. One successful method was reported by
Rice and Schnipke [5] using the four-node quadrilateral element. The method evaluates convection
terms directly along the local streamlines. Calculations presented in their papers have shown that the
method is monotonic and introduces articial numerical diusion. The eect of numerical diusion is
to smear the solution in area of high ow eld gradients and hence decreases the solution accuracy.
The accuracy in a numerical solution is an important factor that must be considered especially
for large scale problems. The solution accuracy can be increased by using small elements in the
computational domain, but it will require additional computer time and data storage [6]. To reduce
such diculty, a technique of adaptive meshing [7,8] is incorporated in the computational algorithm.
In this paper, the technique of streamline upwinding nite element method for triangular elements
is developed for computing the convection terms which appear in both the momentum and the energy
equations. Triangular elements are employed in order to combine eectively with the adaptive mesh-
ing technique presented herein. These triangular elements use equal-order interpolation functions for
the velocity components, the pressure and the temperature, to reduce the complexity in deriving the
corresponding nite element equations. A segregated solution algorithm [9,10] is also incorporated
to compute the velocity components, pressure and temperature separately for further improving the
computational eciency. In addition, the adaptive meshing technique is applied to reduce the com-
putational time as well as the computer memory. The technique places small elements in the regions
of large change in the solution gradients to increase solution accuracy, and at the same time, places
larger elements in the other regions to reduce the computational time. The paper starts from briey
describing the set of the partial dierential equations that satisfy the law of conservation of mass,
momentums and energy. Corresponding nite element equations were derived and element matrices
are presented. The computational procedure used in the development of the computer program and
the basic idea behind the adaptive meshing technique is then described. Finally, the nite element
formulation and the computer program have been veried using examples that have prior numerical
solutions, experimental results, and empirical correlations.
2. Theoretical formulation and solution procedure
2.1. Governing equations
The fundamental laws used to solve two-dimensional, steady state, viscous incompressible thermal
ow behavior consist of: (a) the law of conservation of mass which is called the continuity equation,
(b) the law of conservation of momentums, and (c) the law of conservation of energy, as follows,
9u
9x
+
9t
9,
= 0, (1a)
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 647
j
_
u
9u
9x
+ t
9u
9,
_
=
9
9x
+ j
_
9
2
u
9x
2
+
9
2
u
9,
2
_
, (1b)
j
_
u
9t
9x
+ t
9t
9,
_
=
9
9,
+ j
_
9
2
t
9x
2
+
9
2
t
9,
2
_
jq(1 [(1 1
o
)), (1c)
jc
_
u
91
9x
+ t
91
9,
_
= k
_
9
2
1
9x
2
+
9
2
1
9,
2
_
, (1d)
where u and t are the velocity components in the x and , direction, respectively; j is the density,
is the pressure, j is the viscosity, q is the gravitational acceleration constant, [ is the volumetric
coecient of thermal expansion, 1 is the temperature, 1
o
is the reference temperature for which
buoyant force in the ,-direction vanishes, c is the specic heat, and k is the coecient of thermal
conductivity.
2.2. Finite element formulation
The three-node triangular element is used in this study. The element assumes linear interpolation
for the velocity components, the pressure, and the temperature as,
u(x, ,) = N
i
u
i
, (2a)
t(x, ,) = N
i
t
i
, (2b)
(x, ,) = N
i

i
, (2c)
1(x, ,) = N
i
1
i
, (2d)
where i = 1, 2, 3; and N
i
is the element interpolation functions.
The basic idea of the solution algorithm presented in this paper is to use the two momentum
equations for solving both of the velocity components, use the continuity equation for solving the
pressure, and use the energy equation for solving the temperature. The nite element equations
corresponding to the momentum, the continuity and the energy equations, are shown in the next
section.
2.2.1. Discretization of momentum equations
The two momentum equations, Eqs. (1b)(1c), are discretized using the conventional Bubnov
Garlerkins method. However, a special treatment for the convection terms is incorporated. These
terms are approximated by a monotone streamline upwinding formulation for using with the triangular
element [11]. In this approach, the convection terms in the form,
u
9[
9x
+ t
9[
9,
, (3)
which are related to the transport variable, [, are rst rewritten in the streamline coordinates as
U
s
9[
9s
, (4)
648 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
where U
s
and 9}9s are the velocity and the gradient along the streamline direction, respectively. For
pure convection, the term in Eq. (4) is constant along the streamline. These terms are evaluated by
a streamline tracing method which keeps track of the direction of the ow within the element.
Using the standard Galerkin approach, each momentum equation is multiplied by weighting func-
tions, N
i
, and then the diusion terms are integrated by parts using the Gauss theorem [12] to yield
the element equations in the form,
[A]{u} = {R
x
} + {R
u
}, (5a)
[A]{t} = {R
,
} + {R
t
} + {R
b
}, (5b)
where the coecient matrix [A] contains the known contributions from the convection and diusion
terms. The load vectors on the right-hand side of Eqs. (5a)(5b) are dened by
{R
x
} =
_
O
{N}
9
9x
dO, (6a)
{R
,
} =
_
O
{N}
9
9,
dO, (6b)
{R
u
} = j
_
I
{N}
_
9u
9x
n
x
+
9u
9,
n
,
_
dI, (6c)
{R
t
} = j
_
I
{N}
_
9t
9x
n
x
+
9t
9,
n
,
_
dI, (6d)
{R
b
} =
_
O
{N}[jq(1 [(1 1
o
))] d O, (6e)
where O is the element area and I is the element boundary. The element equations are assembled
to yield the global equations for the velocity components. Such global equations are then modied
for the specied velocity components along the boundaries prior to solving for the new velocity
components.
2.2.2. Discretization of pressure equation
To derive the discretized pressure equation, the method of weighted residuals is applied to the
continuity equation, Eq. (1a),
_
O
N
i
_
9u
9x
+
9t
9,
_
dO=
_
O
_
9N
i
9x
u +
9N
i
9,
t
_
dO
+
_
I
N
i
(un
x
+ tn
,
) dI = 0, (7)
where the integrations are performed over the element domain O and along the element boundary
I; n
x
and n
,
are the direction cosines of the unit vector normal to element boundary with respect
to x and , direction, respectively. As mentioned earlier, the continuity equation is used for solving
the pressure, but the pressure term does not appear in the continuity equation. For this reason, the
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 649
relation between velocity components and pressure are thus required. Such relations can be derived
from the momentum equations, Eqs. (5a)(5b) as,
A
ii
u
i
=

)=i
A
i)
u
)
+ [
u
i

_
O
N
i
9
9x
dO, (8a)
A
ii
t
i
=

)=i
A
i)
t
)
+ [
t
i

_
O
N
i
9
9,
dO, (8b)
where [
u
i
and [
t
i
are the surface integral terms and the source term due to buoyancy. By assuming
constant pressure gradient on an element,
u
i
= u
i
K

i
9
9x
, (9a)
t
i
= t
i
K

i
9
9,
, (9b)
where
u
i
=


)=i
A
i)
u
)
+ [
u
i
A
ii
, (10a)
t
i
=


)=i
A
i)
t
)
+ [
t
i
A
ii
, (10b)
K

i
=
_
O
N
i
dO
A
ii
. (10c)
By applying the element velocity interpolation functions, Eqs. (2a)(2b), into the continuity equation,
Eq. (7),

_
O
9N
i
9x
(N
)
u
)
) dO
_
O
9N
i
9,
(N
)
t
)
) dO +
_
I
N
i
(un
x
+ tn
,
) dI = 0 (11)
and introducing the nodal velocities u
)
and t
)
from Eqs. (9a)(9b), then Eq. (11) becomes
_
O
9N
i
9x
(N
)
K

)
)
9
9x
dO +
_
O
9N
i
9,
(N
)
K

)
)
9
9,
dO
=
_
O
9N
i
9x
(N
)
u
)
) dO +
_
O
9N
i
9,
(N
)
t
)
) dO
_
I
N
i
(un
x
+ tn
,
) dI. (12)
Finally, applying the element pressure interpolation functions, Eq. (2c), the above element equations
can be written in matrix form with unknowns of the nodal pressures as
[K
x
+ K
,
]{} = {F
u
} + {F
t
} + {F
b
}, (13)
650 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
where
[K
x
] =
_
O
_
9N
9x
_
(N
)
K

)
)
_
9N
9x
_
dO, (14a)
[K
,
] =
_
O
_
9N
9,
_
(N
)
K

)
)
_
9N
9x
_
dO, (14b)
{F
u
} =
_
O
(N
)
u
)
)
_
9N
9x
_
dO, (14c)
{F
t
} =
_
O
(N
)
t
)
)
_
9N
9,
_
dO, (14d)
{F
b
} =
_
I
{N} (un
x
+ tn
,
) dI. (14e)
The above element pressure equations are assembled to form the global equations, boundary condi-
tions for the specied nodal pressures are imposed prior to solving for the updated nodal pressures.
2.2.3. Discretization of energy equation
The nite element equations corresponding to the energy equation are derived using an approach
which is similar to the momentum equations. The streamline upwinding method is applied to the
convection term in the energy equation, Eq. (1d). The standard Galerkin method is then applied to
yield the element equations which can be written in matrix form as
[K]{1} = {R}, (15)
where the matrix [K] consists of the contributions from the convection and diusion terms and the
load matrix {R} represents the heat ux along the element boundary as
{R} = k
_
I
{N}
_
91
9x
n
x
+
91
9,
n
,
_
dI. (16)
These element equations are again assembled to yield the global temperature equations. Appropriate
boundary conditions are applied prior to solving for the new temperature values.
2.2.4. Computational procedure
The computational procedure is illustrated in Fig. 1. A set of initial nodal velocity components,
pressures, and temperatures are rst assumed. The new nodal temperatures are computed using
Eq. (15). The new nodal velocity components and pressures are then computed using Eqs. (5a)
(5b) and Eq. (13), respectively. The nodal velocity components are then updated using Eqs. (8a)
(8b) with the computed nodal pressures. This process is continued until the specied convergence
criterion is met. Such segregated solution procedure helps reducing the computer storage because
the equations for the velocity components, the pressure, and the temperature are solved separately.
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 651
Fig. 1. Schematic diagram of the computational procedure.
3. Adaptive meshing technique
The idea behind the adaptive meshing technique presented herein is to construct a new mesh based
on the solution obtained from the previous mesh. The new mesh consists of small elements in the
regions with large change in solution gradients and larger elements in the other regions where the
change in solution gradients is small. To determine proper element sizes at dierent locations in
the ow eld, the solid-mechanics concept for determining the principal stresses from a given state
of stresses at a point is employed. Since small elements are needed in the regions where the change
652 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
in the temperature gradients occurs, thus the ow temperature can be used as an indicator for the
determination of proper element sizes.
To determine proper element sizes, the second derivatives of the temperature with respect to the
global coordinates x and , are rst computed,
_

_
9
2
1
9x
2
9
2
1
9x9,
9
2
1
9x9,
9
2
1
9,
2
_

_
. (17)
The principal quantities in the principal directions X and Y where the cross derivatives vanish, are
then determined,
_

_
9
2
1
9X
2
0
0
9
2
1
9Y
2
_

_
. (18)
The magnitude of the larger principal quantity is then selected,
z = max
_

9
2
1
9X
2

9
2
1
9Y
2

_
. (19)
This value is used to compute proper element size, h, at that location from the condition, [13]
h
2
z = constant = h
2
min
z
max
, (20)
where h
min
is the specied minimum element size, and z
max
is the maximum principal quantity for
the entire model.
Based on the condition shown in Eq. (20), proper element sizes are generated according to the
given minimum element size h
min
. Specifying too small h
min
may result in a model with an excessive
number of elements. On the other hand, specifying too large h
min
may result in an inadequate solution
accuracy or excessive analysis and remeshing cycles. These factors must be considered prior to
generating a new mesh.
4. Results
In this section, three example problems are presented. The rst example, thermally driven cavity
ow, is chosen to evaluate the nite element formulation and to validate the developed computer
program. The second and the third examples, thermally driven ow in concentric cylinders and a
ow past heated cylinder, are presented to illustrate the capability of the adaptive meshing technique
for providing analysis solution accuracy of viscous incompressible thermal ows.
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 653
Fig. 2. Finite element model and boundary conditions for thermally driven cavity ow.
Fig. 3. Predicted velocity vectors and streamline contours for thermally driven cavity ow at Ra=10
4
.
4.1. Thermally driven cavity ow
The rst example for evaluating the nite element formulation and validating the developed com-
puter program is the problem of free convection in a square enclosure. The square enclosure, shown
in Fig. 2, is bounded by the two vertical walls with specied temperatures of 20

C and 60

C and
insulated along the top and bottom surfaces. The nite element model consisting of 2809 nodes
and 5408 triangles, as shown in Fig. 2, is used in this study. Figs. 3 and 4 show the predicted
velocity vectors, streamline and temperature contours, respectively, for the case with the Rayleigh
number of 10
4
. The gures show relatively smooth velocity vectors of the ow that circulates in
the counterclockwise direction and the smooth temperature distribution. The same analysis is re-
peated but with the Rayleigh number of 10
5
. Dierent ow patterns of the velocity vectors, stream-
line and temperature contours are obtained as shown in Figs. 5 and 6. The ow velocity pattern
654 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
Fig. 4. Predicted temperature contours for thermally driven cavity ow at Ra=10
4
.
Fig. 5. Predicted velocity vectors and streamline contours for thermally driven cavity ow at Ra=10
5
.
Fig. 6. Predicted temperature contours for thermally driven cavity ow at Ra=10
5
.
indicates two regions of circulation, both in counterclockwise direction. The predicted velocity pro-
les and the temperature distributions along the mid-height of the square enclosure for both ow
cases are compared with the results from Sai et al. [14] as shown in Figs. 7 and 8, respectively.
The gures show good agreement of the solutions for both the ow cases. In addition, Table 1
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 655
Fig. 7. Comparative velocity proles along the mid-height of the square enclosure.
Fig. 8. Comparative temperature distributions along the mid-height of the square enclosure.
Table 1
Comparison of the predicted average Nusselt numbers for thermally driven cavity ow
Ra Average Nusselt number along the hot wall (% dierence from Ref. [15])
Benchmark solution [15] Finite element [14] Present work
10
4
2.238 2.289 (2.28%) 2.256 (0.80%)
10
5
4.509 4.688 (4.00%) 4.349 (3.55%)
compares the predicted average Nusselt numbers at the hot wall, Nu
x=L
, with those presented
in Refs. [14,15]. The table shows good agreement of the average Nusselt numbers for both the
ow cases.
656 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
Fig. 9. Thermally driven ow in concentric cylinders.
4.2. Thermally driven ow in concentric cylinders
To demonstrate the capability of the combined adaptive meshing technique and the nite element
method, the problem of thermally driven ow in concentric cylinders is selected. The uid is freely
convected in the annular space between long, horizontal concentric cylinders due to high temperature
on the inner cylinder and lower temperature on the outer cylinder. The schematic of the problem is
shown in Fig. 9. This problem was previously studied by experiment [16] for which the result can
be used for comparison. The comparison of the results presented herein are at Rayleigh number of
4.7 10
4
and Prandtl number of 0.706 with a ratio of gap width to inner-cylinder diameter (L/D)
of 0.8.
Due to symmetry of the ow solution, only the right half of the concentric cylinders can be
analyzed. The adaptive meshing technique starts from creating a relatively uniform mesh as shown
in Fig. 10. This initial mesh consists of 1045 nodes and 1942 triangles. The gure also shows the
predicted temperature contours on the left half.
The numerical solution obtained from the initial mesh is then used to construct the second adaptive
mesh as described in Section 3. The second adaptive mesh and the predicted temperature contours
are shown in Fig. 11. The gure shows smaller elements are generated in some regions near the
cylinder surfaces where large change in temperature gradients occurs.
The entire procedure is repeated again to generate the third adaptive mesh for predicting the
temperature contours as shown in Fig. 12. Fig. 13 shows the comparisons of the equivalent con-
ductivities between the initial and the third adaptive meshes at the inner and outer cylinder sur-
faces with the experimental results. The equivalent conductivity is dened as the actual heat ux
divided by the heat ux that would occur by pure conduction in the absence of uid motion.
The gure shows the adaptive mesh provides good solution accuracy compared to the experi-
mental results because small elements are generated automatically in the regions of complex ow
behavior.
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 657
Fig. 10. First nite element mesh and the predicted temperature contours.
Fig. 11. Second nite element mesh and the predicted temperature contours.
It should be noted that if a uniform mesh is used to capture the same ow behavior with equivalent
solution accuracy as obtained from the third adaptive mesh, the number of nodes will increase
approximately from 3200 to 17,000. This implies that the computer memory requirement will increase
approximately ve times and the computational time may increase by the factor of ten.
4.3. Flow past heated cylinder
To further demonstrate the capability of the adaptive meshing technique on forced convection
analysis, the problem of viscous thermal ow past a heated cylinder as shown in Fig. 14 is selected.
The uid temperature of 20

C at the Reynolds number of 40 enters through the left boundary of

a computational domain and ows past the circular cylinder that has higher temperature of 40

C.
In the computational process, the average Nusselt number, Nu, along the cylinders surface was
calculated. This average Nusselt number was compared to an empirical correlation of Kang et al.
658 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
Fig. 12. Third nite element mesh and the predicted temperature contours.
Fig. 13. Comparison of the experimental and numerical local equivalent conductivities.
Computational domain
Heated cylinder
T = 40C
Flow
T = 20C
Fig. 14. Computational domain of ow past a heated cylinder.
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 659
Fig. 15. Initial nite element mesh for ow past a heated cylinder.
Fig. 16. Predicted temperature contours from the initial nite element mesh.
[17] which is in the form,
Nu = (0.22 + 0.58 Re
0.45
)(1
w
}1
i
)
0.113
(21)
where 1
w
and 1
i
are the values of temperature on the cylinder surface and inlet uid temperature,
respectively.
Due to symmetry of the ow solution, only the upper half of the circular cylinder can be analyzed.
Fig. 15 shows the initial mesh that consists of 1339 nodes and 2535 triangles with relatively uniform
element sizes. The predicted temperature contours is shown in Fig. 16. Fig. 17 shows the detailed
temperature contours near the surface of the cylinder. The error of the average Nusselt number
calculated from this initial mesh is approximately 52% of the empirical correlation in Eq. (21).
These suggest that smaller elements are needed near the cylinder surface to capture accurate ow
behavior.
The numerical solution obtained from the initial mesh is then used to construct the second adaptive
mesh. The second adaptive mesh consisting of 847 nodes and 1576 triangles is shown in Fig. 18.
Smaller elements are generated automatically in the region near the heated cylinder due to large
change in temperature gradients. With this second adaptive mesh, the nite element analysis is then
performed and the corresponding temperature contours are shown in Fig. 19 with detail given in
660 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
Fig. 17. Detailed temperature contours near the surface of cylinder in Fig. 16.
Fig. 18. Second nite element mesh for slow past a heated cylinder.
Fig. 19. Predicted temperature contours from the second nite element mesh.
Fig. 20. The average Nusselt number obtained from this mesh is approximately 9% dierent from
that given by Eq. (21).
The entire procedure is repeated again to generate the third adaptive mesh with 841 nodes and 1578
triangles as shown in Fig. 21. The corresponding ow solution contours of temperature distribution
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 661
Fig. 20. Detailed temperature contours near the surface of cylinder in Fig. 19.
Fig. 21. Third nite element mesh for ow past a heated cylinder.
Fig. 22. Predicted temperature contours from the third nite element mesh.
are shown in Fig. 22. Detailed temperature contours near the cylinder surface are also shown in
Fig. 23. This third adaptive mesh yields the average Nusselt number approximately 1% dierent
from that given by Eq. (21). Again, if a uniform mesh is used in order to obtain the same solution
accuracy as for the third adaptive mesh, the number of nodes will increase approximately 25 times
and may cause the computational time to increase by the factor of fty.
662 N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663
Fig. 23. Detailed temperature contours near the surface of cylinder in Fig. 22.
This example demonstrates that the adaptive meshing technique can increase solution accuracy
in the regions of complex ow behavior and, at the same time, can reduce the total number of
unknowns and computer memory by using larger elements in the other regions.
5. Conclusions
This paper presents the development of the streamline upwinding nite element method using the
three-node triangular element for the analysis of viscous incompressible thermal ows. The presented
solution algorithm employs equal-order interpolation functions for the velocity components, pressure
and temperature of the triangle. The nite element formulation and its computational procedure are
rst described. The procedure uses a segregated solution algorithm to compute the velocities, pres-
sure and temperature separately for improving the computational eciency. A streamline upwinding
formulation is applied to the convection terms in the momentum and the energy equations to sup-
press the non-physical spatial oscillation in the numerical solutions. The corresponding nite element
equations are derived and a corresponding computer program has been developed. The nite element
method is combined with an adaptive meshing technique to improve the ow solution accuracy. The
adaptive meshing technique generates an entirely new mesh based on the solution obtained from a
previous mesh. The new mesh consists of clustered elements in the regions with large change in the
temperature gradients to provide higher solution accuracy. And at the same time, larger elements
are generated in the other regions to reduce the computational time and the computer memory.
The capability of the nite element solution algorithm and the corresponding computer program
has been evaluated by examples that have prior numerical solutions and experimental results. The
combined nite element solution algorithm and the adaptive meshing technique has demonstrated
the eciency of the entire process for improved solution accuracy at the reduced computer memory
and computational time.
Acknowledgements
The authors are pleased to acknowledge the Thailand Research Fund (TRF) for supporting this
research work.
N. Wansophark, P. Dechaumphai / Finite Elements in Analysis and Design 40 (2004) 645663 663
References
[1] S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere, Washington DC, 1980.
[2] C. Prakash, S.V. Patankar, A control volume-based nite-element method for solving the NavierStokes equations
using equal-order velocitypressure interpolation, Numer. Heat Transfer 8 (1985) 259280.
[3] B. Ramaswamy, T.C. Jue, A segregated nite element formulation of NavierStokes equations under laminar
conditions, Finite Elements Anal. Des. 9 (1991) 257270.
[4] V. Haroutunian, M.S. Engelman, I. Hasbani, Segregated nite element algorithms for the numerical solution of
large-scale incompressible ow problems, Int. J. Numer. Methods Fluids 17 (1993) 323348.
[5] J.G. Rice, R.J. Schnipke, A monotone streamline upwind nite element method for convection-dominated ows,
Comput. Methods Appl. Mech. Eng. 48 (1985) 313327.
[6] O.C. Zienkiewicz, Y.C. Liu, G.C. Huang, Error estimates and convergence rates for various incompressible elements,
Int. J. Numer. Methods Eng. 28 (1989) 21912202.
[7] J. Peraire, M. Vahjdati, K. Morgan, O.C. Zienkiewicz, Adaptive remeshing for compressible ow computation,
J. Comput. Phys. 72 (1987) 449466.
[8] P. Dechaumphai, Adaptive nite element technique for heat transfer problems, J. Energy, Heat Mass Transfer 17
(1995) 8794.
[9] J.G. Rice, R.J. Schnipke, An equal-order velocity-pressure formulation that does not exhibit spurious pressure modes,
Comput. Methods Appl. Mech. Eng. 58 (1986) 135149.
[10] H.G. Choi, J.Y. Yoo, Streamline upwind scheme for the segregated formulation of the NavierStokes equation,
Numer. Heat Transfer, Part B 25 (1994) 145161.
[11] N. Wansophark, P. Dechaumphai, Enhancement of streamline upwinding nite element solutions by adaptive meshing
technique, JSME Int. J. Ser. B 45 (4) (2002) 770779.
[12] O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, 5th Edition, Butterworth-Heinemann, Oxford, 2000.
[13] J.T. Oden, G.F. Carey, Finite Elements: Mathematical Aspects, Prentice-Hall, Englewood Clis, NJ, 1981.
[14] B.V.K.S. Sai, K.N. Seetharamu, P.A.A. Narayana, Solution of transient laminar natural convection in a square cavity
by an explicit nite element scheme, Numer. Heat Transfer, Part A 25 (1994) 593609.
[15] G. de Vahl Davis, Natural convection of air in a square cavity: a bench mark numerical solution, Int. J. Numer.
Methods Fluids 3 (1983) 249264.
[16] T.H. Kuehn, R.J. Goldstein, An experimental and theoretical study of natural convection in the annulus between
horizontal concentric cylinders, J. Fluid Mech. 74 (1976) 695719.
[17] S.H. Kang, K.H. Hong, S. Kauh, A unied correlation of laminar convective heat transfer from hot and cold circular
cylinders in a uniform air ow, Int. J. Heat Mass Transfer 38 (1995) 752755.