ME 608 Numerical Methods in Heat, Mass, and Momentum Transfer

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

ME 608

Numerical Methods in Heat, Mass, and Momentum Transfer

Solution to Mid-Term Examination


Date: February 28, 2006
Instructor: J. Murthy
Open Book, Open Notes

1. The governing equation is:  


1 d dT
kr =0
r dr dr
(a) Integrating over the control volume rdr yields
dT dT
kn rn |n − ks rs |s = 0
dr dr
Using a linear profile assumption:
Tbn − TP TP − Tbs
2kn rn − 2ks rs =0
∆r ∆r
Collecting terms:  
2kn rn 2ks rs 2kn rn 2ks rs
+ TP = Tbn + Tbs
∆r ∆r ∆r ∆r

(b) Writing an interface balance at the north face:


dT
|n = εσ T∞4 − T 4

k
dr
Discretizing
Tbn − TP
= εσ T∞4 − Tbn
4

2k
∆r
Linearize the right hand side using:

∂q ∗
q = (T − T ∗ )
q∗ +
∂T
= εσ T∞4 − T ∗4


−4εσ T ∗3 (T − T ∗ )
= εσ T∞4 + 3εσ T ∗4 − 4εσ T ∗3 T

Thus, the interface balance becomes:


Tbn − TP
2k = εσ T∞4 + 3εσ Tbn
∗4
− 4εσ Tbn
∗3
Tbn
 ∆r

2k 2k
+ 4εσ Tbn
∗3
Tbn = TP + εσ T∞4 + 3εσ Tbn
∗4
∆r ∆r

(c) For the south boundary face


dT
k |s = q”0
dr

1
Discretizing:
TP − Tbs
2k = q”0
∆r
2k 2k
Tbs = TP − q”0
∆r ∆r
It is of course possible to eliminate the discrete equations for Tbn and Tbs from the cell balance for TP , but this is not
necessary.
(d) We choose the tangent to the q(T) line in order to get the best estimate of the dependence of the radiative flux on
the boundary temperature. If we chose a linearization with a more negative dq/dT we would have slowed down the
iteration (and perhaps gained more stability). If we chose a linearization with a less negative dq/dT , we could have
convergence problems because we would not have correctly accounted for the dependence of q on T.
(e) Since the problem is non-linear, our iterative process would take the following form:
1. Guess T everywhere, including boundaries.
2. Find all discrete coefficients, using prevailing temperatures for Tbn to account for non-linearities.
3. Solve nominally linear system to obtain Tbn , TP and Tbs .
4. Check for convergence. If converged, stop. Else, go to step 2.

q"
bn r

P
∆r
bs

r q"
0

Figure 1: Computational Domain for Problem 1

2. (a) The mass flux F1 through face f 1 is ρ V · A = 5 kg/s. Accounting for the mass source, we have:

F2 = F1 + ṁ∆V = 5 + 2 × 0.5 = 6
F3 = F2 + ṁ∆V = 6 + 2 × 1 = 8
F4 = F3 + ṁ∆V = 8 + 2 × 0.5 = 9

2
(b) Discretizing the scalar transport equation (with Γ = 0) yields

F1 φb + ṁφsource ∆V = F2 φ1 = (F1 + ṁ∆V ) φ1


F2 φ1 + ṁφsource ∆V = F3 φ2 = (F2 + ṁ∆V ) φ2
F3 φ2 + ṁφsource ∆V = F4 φ3 = (F3 + ṁ∆V ) φ3

Substituting numerical values, we obtain

6φ1 = 5.5
8φ2 = 6φ1 + 1
9φ3 = 8φ2 + 0.5

(c) The discrete equations obtained using the upwind scheme are diagonally dominant, and can be solved using the
Gauss-Seidel scheme. However, this is not necessary since information flows from left to right, and we can successively
evaluate φ from left to right.
(d) Solving from left to right in Fig. 2, we obtain the following answers using UDS:

φ1 = 0.916
φ2 = 0.812
φ3 = 0.777

These answers make sense because the mass being added through the mass source has less φ than the incoming mass
through the f 1 face. Therefore we would expect species A to be diluted in direct proportion to the mass being added.
(e)Using CDS, and using a simple average of φ at the face we obtain
φ1 + φ2
F1 φb + ṁφsource ∆V = F2
2
φ1 + φ2 φ2 + φ3
F2 + ṁφsource ∆V = = F3
2 2
φ2 + φ3
F3 + ṁφsource ∆V = F4 φ3
2
Substituting numerical values, we obtain

3φ1 = −3φ2 + 5.5


φ2 = 3φ1 − 4φ3 + 1
5φ3 = 4φ2 + 0.5

(f) Since the equations do not have diagonal dominance, we cannot solve them directly using the Gauss-Seidel scheme.
It may be possible to rearrange them so that diagonal dominance is obtained, however.
(g) Solving the three simultaneous equations, we obtain the following φ distribution for CDS:

φ1 = 0.986
φ2 = 0.847
φ3 = 0.777

These are plausible results, and are not wiggly, primarily because the mass addition, as well as the inlet and outflow
values, have been upwinded and there are only three cells.

3
Note that φ3 comes out the same with both schemes. This occurs because an overall balance over the whole domain for
both UDS and CDS yields exactly the same algebraic equation. (The interior faces are eliminated, and we are using the
same upwinded face fluxes at the inlet and the outlet for both cases.)
It is instructive to compute the exact solution for the problem, which is given by:

ṁφsource + (ρ u) f 1 φb
φ (x) =
ṁ + (ρ u) f 1
or
5+x
φ (x) =
5 + 2x
The exact numerical values at the cell centroids and faces are given below.

φb = 1.0
φ1 = 0.941
φf2 = 0.916
φ2 = 0.8571
φf3 = 0.8125
φ3 = 0.7999
φf4 = 0.7777

As we can see, UDS gets the face values exactly right in this 1D problem, but the cell-centroid values are wrong. CDS
gets neither correct, but will tend to the correct cell-centroid values more quickly than UDS as the mesh is refined
because it is a second-order scheme.

(0,1) (2,1)
Wall

Outlet
Inlet
2
V=5i f3
φ = 1.0 f2
b f4
1 3
All lengths are in meters
f1

(0,0) (1,0) (2,0)

Wall

Figure 2: Computational Domain for Problem 2

You might also like