Professional Documents
Culture Documents
1987 - Rank, Zienkiewicz - A Simple Error Estimator in The Finite Element Method
1987 - Rank, Zienkiewicz - A Simple Error Estimator in The Finite Element Method
1987 - Rank, Zienkiewicz - A Simple Error Estimator in The Finite Element Method
SUMMARY
A technique of creating matrices of interface elements for a heat transfer analysis by the finite element
method (FEM)is presented. The method utilizes general convection matrices in the formation of an interface
element matrix.
INTRODUCTION
In the calculation of heat transfer by the finite element method (FEM), there is often a need, for
example, to model the heat resistance of an oxide film between different parts or to connect two
substructures into an overall FEM model. In those cases, an interface element is applicable.
However, it is quite difficult to model a film resistance by two-dimensional or three-dimensional
elements having one dimension very small compared with the others and having anisotropic
conductivity. This may introduce numerical difficulties in the formation of element matrices
because of the highly distorted geometry.' One possibility to avoid this is to use special thin
elements2 In the application of thin elements, the interpolation is always reduced to linear in the
thin direction of the element.
FORMULATION OF AN INTERFACE ELEMENT
In the following, special interface elements for a heat transfer analysis are introduced. The method
is very similar to the Coincident Node Technique reported by Samonds et d.,*
except that the
calculation of the interface element matrix is different.
Consider a surface element as shown in Figure 1. The element consists of two layers (I and 11)
of nodes joining two surfaces together and having a convection coefficient h between the surfaces.
The heat flow q through the surfaces obeys the convection law
q=hhT
(1)
0748-8025/87/040297-04$05.00
298
J . A . KARJALAINEN
where AT is the temperature difference between the surfaces. Considering the nodal temperature
differences between the nodes on the layers I and I1 and assuming a temperature interpolation
matrix [HI for expressing AT on the element surface in terms of the nodal temperature values,
we thus obtain
AT= [HI ( ( 7 - 1 1
{TI111
(2)
(4)
AT = [C] { T}
The potential Il can be expressed' in the form
(5)
[[A,
:Is
n=-
ATThATdS
(7)
where we used the transpose of a scalar for the further manipulations. Now we can substitute
expression ( 5 ) into (7) and thus obtain
n =j
(8)
Then we can replace the matrix [C] with the original form [HI [[I],-[I]] and have (8) in the form
n = 2 {TY-jjHI
" 4 7
[ W S {T}
(9)
which, in turn, can be expressed after some simple mathematical manipulations in the form
Now we notice that the integral part of the expression (10) is similar to that of the matrix for the
thermal convection or
[h] =
[HITh [HI dS
The final result is that the convection matrix of the interface element can be assembled into 2n
space, using the general convection matrix as a submatrix.
FURTHER REMARKS
Although result (12) was derived for the surface interface, the method itself is more general; it
can also be applied for point and line interfaces, using appropriate convection matrices, as shown
in Figure 2.
299
Convection matrix
P I l X l
Convection matrix
L
IQ
[hl =
htL 2 1
6 1 21
Figure 2. Point (a) and line interface elements (b) with the respective matrices. The line convection matrix may be
obtained from Reference 3
Compared with the Coincident Node Technique in Reference 2, the proposed method has
several advantages. First, in the proposed method, the geometry of the element can be described
using nodes on, for example, the first layer, and the second layer is only for topological purposes.
Hence it is possible to connect parts that are not actually coincident. However, the geometrical
compatibility between the connected parts should be considered. This property of geometrical
independence of the second layer can be misused in the definition of time-dependent convection
loading by connecting all nodes of the second layer to one node with the time-dependent temperature.
The second advantage is that there is no need for new element routines, because the interface
element matrix can be assembled by using general convection matrices available in any heat
transfer FEM code.
One difficulty in the application of the interface element is that the system coefficient matrix
may become ill-conditioned when the off-diagonal elements are large (the interface convectivity
is high). Then it is necessary to use enough digits (e.g. double precision) in the computer
arithmetical operations to ensure an accurate solution of the problem. When the method was
implemented in the cooling simulation of an injection mould, no numerical problems were
encountered when double precision arithmetic (14 significant digits) was used on an IBM 3083
computer. Some results of the use of the proposed interface element in the injection mould cooling
simulation are reported by the author in Reference 4.
CONCLUSIONS
A matrix to model an oxide film or to connect substructures in the FEM heat transfer analysis
can be quite easily obtained by assembling the general point or surface convection matrices in the
2n-space, where n is the number of temperature nodes in the convection matrix. Thus there is no
need to derive any special matrix to model interfaces, but the existing program can be modified
for the inclusion of the interface elements.
When using very high interface connectivities, special attention should be given to the numerical
stability of the solution, because the coefficient matrix may become ill-conditioned due to large
off-diagonal terms. By using enough digits (e.g. double precision) in the computer solution,
numerical problems can be avoided.
300
J . A. KARJALAINEN
ACKNOWLEDGEMENTS
The author would like to express his appreciation to the Research Council for Technology of the
Academy of Finland for supporting his research.
REFERENCES
1. K.-J. Bathe, Finife Elemenr Procedures in Engineering Analysis, Prentice-Hall, Englewood Cliffs, N.J., 1982.
2. M. Samonds, R. W. Lewis, K. Morgan and R. Symberlist, Finite element modelling of the mold-metal interface in
casting simulation with coincident nodes or thin elements, in Computational Techniques in Hear Transfer (Eds Lewis,
R. W. er a / . ) , Pineridge Press, Swansea, 1985.
3. J. Heuser, Finite element method for thermal analysis, NASA Technical Nore TN 0-7274, Washington D.C. (1973).
4. J . A. Karjalainen, Computer aided design of injection mold cooling, Preprints, 2nd Int. Conf. on Computer Applications in Production and Engineering, Copenhagen, 1986.