Professional Documents
Culture Documents
(K x
) + (K y
) + (K z
) + f ( x, y , z ) = 0
z
z
y
y
x
x
= B
Kx
on B
nx + K y
ny + K z
n z + g ( x, y, z ) + h( x, y, z ) = 0 on B
y
z
Where g(x,y,z) and h(x,y,z) are known a priori and nx, ny and nz are the
direction cosines of the unit outward normal to the surface. A and B are
parts of the boundary, i.e. A+B=, the total boundary.
Boundary condition in above equation is known as the dirichlet condition
and B, the dirichlet data. Equation 2(b) represents the Cauchy boundary
condition.
If g=h=0, the Cauchy condition reduces to the Neumann boundary
condition, also called the Natural boundary condition.
A field problem is said to have mixed boundary conditions when some
portions of the boundary have Dirichlet boundary conditions while the other
portions have Cauchy or Neumann boundary conditions.
Physical interpretation of the parameters in eqn 1 depends upon the
particular physical problem and listed in the table below:
Kx, Ky and
Kz
Diffusion flow in
porous media
Hydraulic
head
Hydraulic
conductivity
Internal
sources
flow
Boundary
flow
Boundary
heat
generation
Convective
heat
transfer
coefficient
Boundary
velocity
Heat conduction
Temperature
Thermal
conductivity
Internal
heat
generation
Irrotational flow
Velocity
potential or
stream
function
Torsion
Stress
Function
Reciprocal of
shear
modulus
Seepage
Pressure
Permeability
Angle
of twist per
unit length
Internal
flow
1
2
2
2
2
1 2
K x ( ) + K y ( ) + K z ( ) 2 f d + ( g + h )d
z
2
y
x
It can be shown that J()=0 yields the Euler equations which are the same
as the equations 1 and 2.
Note: there are some slight modifications involved when eqns1 to 3 are
applied to a particular physical problem.
(k x
)+
(K y
) + Q ( x, y ) = 0
y
y
x
x
in
= B
Kx
Kx
on B
nx + K y
n y + q A = 0 on A
y
nx + K y
n y + q c + ( c ) = 0
y
x
on c
where = temperature
Kx=Thermal conductivity in x-direction
Ky= Thermal conductivity in y-direction
Q=Heat input per unit volume
qa and qc = specified heat input per unit area on A and C, respectively.
=Convective heat transfer coefficient
c=ambient temperature of the environment
Variational principle in two dimensions with thickness "t" take the
following form:
J ( ) = t
2
2
K x ( ) + K y ( ) Q dxdy + t q A d + t qC + ( c ) d
2
x
y
2
Equation 4-b for boundary condition on Ais valid only for transfer of heat
through conduction.
y
t= thickness (uniform)
A
B
A+B+C=
x
N ii
i =1
Where Ni are the shape functions, I are the nodal values of , n is the
number of nodes per element.
Finite element approximation is required to have only C0 continuity. That is
only needs to be continuous and no derivatives of it are required to be
continuous.
For the interior elements we do not have to consider the boundary integrals.
Hence, for interior elements:
Equation 6
J e ( ) = t
2 K
A
x(
2
) + K y ( ) 2 Q dxdy
y
x
A2
J e (i e ) =
where :
H ij e = t
i =1 j =1
i =1
J e (i e ) = t
1
2
i j
e
ij
i =1 j =1
[K N
x
i, x N j, x
fi ei
i =1
+ K y N i , y N j , y dxdy
fi = t
e
QN dxdy
i
A
e
[
= [f
{ e }T = 1e 2 e . . n e
{ f e }T
e
1
f 2e . .
f ne
2.2-
y
x
Stratification
Stratification
x
Then:
H ij e = t
[K
x N i, x N j , x
+ K y N i , y N j , y dx dy
The only dufference is that the derivatives of Ni are now taken with respect
to x and y, the local coordinates.
Again as commented before, there is no transformation needed from x and y
axes to x-y axes because is a scalar quantity. This then leads to a
considerable economy in computations.
3
(x3, y3)
3
2
(x2, y2)
2
1 1
(x1, y1)
For linear temperature variation within the element, using area coordinatea:
Equation 7
N 1 = L1
N 2 = L2
N 3 = L3
= L11 + L 2 2 + L3 3
ai
where A = area of the triangular element in the figure
2A
b2 = y 3 y1
b3 = y1 y1
bi
2A
b1 = y 2 y 3
N i, x =
N i, y =
a1 = x 3 x1
H ij e =
t
4A2
a 2 = a1 a 3
[K
x bi b j
a 3 = x 2 x1
+ K y a i a j dxdy
Kt
bi b j + a i a j
4A
Qdxdy + t
q Ad + t
[q
c ) d
Obviously, the contribution from line integrals comes only from that part of
boundary i.e. A and C where qA, qC and c are specified.
For Q constants:
{f }
e
1
3
1
= QA
3
1
3
lij
q Ad = tl ij
q A ( ) ( ) d
q2
on int egration
q1
2 q + q 2
tl ij 1
f2 =
q1 + 2q 2
6
0
Similarly , for elements with edge along C :
{ }
e
qA
x
2 g + g 2
tl ij 1
f3 =
g1 + 2 g 2
6
g ( ) = q C ( ) c ( )
{ }
e
where both q C and c are assumed to have linear var iation along edge i j :
g 1 = g (0)
g 2 = g (0)
d .
H ij = tl ij
C
i
N j C d
where NiC are shape function along edge i-j of the element that coincides
with C.
2 1 0
[ H ] 33
tl ij
1 2 0
=
6
0 0 0
10
(k x
) + (K y
) + Q ( x, y ) C
=0
x
x
y
y
t
in
2
2
C
K x ( ) + K y ( ) Q + & dxdy + t q A d + t q C + ( c ) d
2
x
y
2
2
note : & =
t
N i ( x, y ) i (t )
i =1
where nodal variable i(t) are now functions of time. Since we look for
stationary of J() at any time t, i.e.:
J ( ) = 0 at any time t
Therefore matrices [He] and [HC], the element heat transfer matrix and
contribution to it from boundary integral on C, are still the same. Load
vector {f1e}, {f2e} and {f3e} may vary with time. However a new matrix
needs to be derived, i.e. element heat capacity matrix [Ce] from t C&dxdy
2
c ij = t
e
CN ( x, y) N
i
j ( x,
y )dxdy
N 2 = L2
N 3 = L3
CL L dA
i
Where [H], [C] (if variable with ) and {F} are assigned their mid interval
values, and {&} has been replaced by:
Equation 9
{ } t + t { } t
t
Also note for linear var iation within the time int erval
1
{ } t = ({ } t + t + { } t )
t+
2
2
{&} t + t =
[ H ] + [C ] {} t = [C ]{}t {F }t + t
+
t
t
t
2
1
{}
t+
t
2
= [ H ] + [C ] [C ]{}t {F }t + t
t
t
And from equation 9c, we can calculate {}t+t. Equation 8 and 9c provide
the recurrence process sought.
If Kx, Ky, C and depend on temperature, then the problem becomes
nonlinear and some special iterative techniques are required for a solution.
12