Aerospace and Mechanical Engineering

1st Exam of Computational Fluid Dynamics

January 27th, 2023
Closed Book Exam
Duration: 2 hours
Name: ________________________________________________________________________ Number: ______________

We are all expected to abide by the IST standards for academic excellence and integrity.

Problem 1 (2.0+1.5+1.5 = 5.0 Marks)

Consider the fully developed Couette fluid flow between two parallel plates. The bottom plate is
stationary, and the top plate is moving with velocity U, as illustrated in the figure below.

The simplification of the continuity and Navier-Stokes equations for these mentioned flow
conditions yields the following equation:
𝑑 𝑑𝑢 𝑑𝑝
(𝜇 ) = −
𝑑𝑦 𝑑𝑦 𝑑𝑥
where the viscosity 𝜇 is assumed to be constant and the pressure gradient is assumed to be also
constant.

a) Consider the case where the pressure gradient is known and is equal to a constant:
𝑑𝑝
= 𝐶𝑡𝑒.
𝑑𝑥
Use the Finite Volume Method and a second order central scheme to obtain the finite
difference equation for each of the four Control Volumes shown in the figure (∆𝑦 = 𝐻/4).
Write in matrix form the resulting system of algebraic equations.

1
 𝑑𝑝
b) Consider the case of zero pressure gradient, = 0. Use the Gauss-Seidel iterative
𝑑𝑥
method to get the solution of the system of equations and obtain the velocity at each of
the Control Volumes as a function of U moving plate. Consider as initial guess:
10𝑈 3𝑈 5𝑈 7𝑈
𝑢1 = ; 𝑢2 = ; 𝑢3 = ; 𝑢4 =
8 8 8 8
Perform two iterations at most.

2
 𝑑𝑝
c) If you find a linear velocity profile for the case of pressure gradient = 0 , what is the
𝑑𝑥
truncation error? Does the error change if you consider ∆𝑦 = 𝐻/40?

Jacobi: Gauss-Seidel:
Formulae

3
4
Name: ________________________________________________________________________ Number: ______________

Problem 2 (1.0+1.0+1.0+1.0+1.0 = 5.0 Marks) (Scattered Questions)

a) Present three possible reasons (of different nature) to define/limit the Time Step in a
general numerical method.
R1:_____________________________________________________________________
________________________________________________________________________

R2:_____________________________________________________________________
________________________________________________________________________

R3:_____________________________________________________________________
________________________________________________________________________

b) Identify the temporal discretization scheme A, B or C above, with Crank-Nicolson, Implicit

Euler or Explicit Euler as options.

A__________________________ B__________________________ C__________________________

c) Suppose that you are calculating a specific flow and you got the results from different
meshes, presented below.
Compute the estimated order of accuracy and the Grid
Convergence Index, GCI, (For the GCI assume the theoretical
order of accuracy is 2 and the Safety Factor is 1).

Grid Spacing Local Parameter’s Value

1 120
0.5 103.5
0.25 100.875
0.125 100.22

5
 Formulae
𝑝 𝐿𝑜𝑔(𝜖) = 𝐿𝑜𝑔(𝛼) + 𝑝𝐿𝑜𝑔(ℎ)
𝜖 = 𝜙(ℎ) − 𝜙𝐸𝑥𝑎𝑐𝑡 = 𝛼ℎ + 𝐻. 𝑂. 𝑇.
𝜙 3 − 𝜙2 𝜙𝐸𝑥𝑎𝑐𝑡 (𝑥) = 𝜙ℎ (𝑥) + 𝛼ℎ𝑝 + 𝑂(ℎ𝑝+1 )
𝑝 = 𝐿𝑜𝑔 ( ) /𝐿𝑜𝑔(𝑟)
𝜙2 − 𝜙1
𝜙ℎ − 𝜙ℎ 1 𝑝 𝑝+1 𝑝+1
|𝜙𝐸𝑥𝑎𝑐𝑡 − 𝜙ℎ1 | = |[ 𝑝2 𝑝 ] ℎ1 | + 𝑂(ℎ2 ) + 𝑂(ℎ1 )
ℎ2 − ℎ1
ℎ𝑖+1 𝐹𝜖 |𝜖| 𝐹𝜖 |𝜖|𝑟 𝑝
𝑟= 𝐺𝐶𝐼𝑓𝑖𝑛𝑒 = 𝑝 𝐺𝐶𝐼𝑐𝑜𝑎𝑟𝑠𝑒 = 𝑝
ℎ𝑖 𝑟 −1 𝑟 −1
𝜙1 − 𝜙2 𝐹𝜖 |𝜖|
𝜙ℎ=0 = 𝜙𝑓𝑖𝑛𝑒 + 𝑝 𝐺𝐶𝐼 = 𝑝
𝑟 −1 𝑟 −1
𝐺𝐶𝐼2,3
𝐴𝑠𝑦𝑚𝑝𝑡𝑜𝑡𝑖𝑐 𝑟𝑎𝑡𝑒 𝑜𝑓 𝑐𝑜𝑛𝑣𝑒𝑟𝑔𝑒𝑛𝑔𝑒: 𝑝 𝑠ℎ𝑜𝑢𝑙𝑑 𝑏𝑒 ≅ 1
𝑟 × 𝐺𝐶𝐼1,2

d) Consider the convection discretization scheme for the face flux f , illustrated below with
positive 𝑽𝒇 > 0, for a uniform grid, as:
3 1
𝜙𝑓 = 𝜙𝐶 − 𝜙𝑈
2 2
Justify if this scheme is bounded.

For 𝑽𝒇 > 0 write the face flux

discretization with the other
schemes used in the MINMOD non-
linear scheme.

6
Formulae

7
e) Consider the domain below, where the boundary was defined, and the faces (boundary
edges) were numbered in a counter-clockwise way. Four points in the bulk were also
created to further define a mesh.
Following the Advancing Front mesh generation technique, define a meshing criteria for
the triangles to be constructed and thus conduct the mesh generation.
Number the next faces and triangles being generated.
Refer for any particular decision you have to take.
Mark with # the triangles automatically defined along the process.

8
Name: ________________________________________________________________________ Number: ______________

Problem 3 (1.0+1.0+2.0+1.0+1.0 = 6.0 Marks)

Consider the unstructured mesh
oh the right where the different
faces of control volume P are
highlighted with color.
The external normal are
represented as red vectors,
pointing outward, and the blue
vectors are the vector distance
from P centroid, to the centroid of
the neighbors.

As a methodology, the subscript

PA, for instance, refers to the
entities at face separating control
volumes P and A, or to the
distance vector of A relative to P.

For the sake of simplicity, some quantities/entities were previously computed and are
presented below.
Again, to clarify, if 𝜙𝑃𝐴 is the interpolation of 𝜙 to the left green face, then:
𝜙𝑃𝐴 = 𝑔𝑃𝐴 𝜙𝑃 + (1 − 𝑔𝑃𝐴 )𝜙𝐴

Normal Distance vector Interpolation factor

𝑛⃗𝑃𝐴 = (−1; 0) 𝑟𝑃𝐴 = (−1; 1) 𝑔𝑃𝐴 = 0.5
𝑛⃗𝑃𝐵 = (0; 1) 𝑟𝑃𝐵 = (0.5; 1.5) 𝑔𝑃𝐵 = 1/3
𝑛⃗𝑃𝐶 = (1; 0) 𝑟𝑃𝐶 = (1.5; 0.5) 𝑔𝑃𝐶 = 2/3
𝑛⃗𝑃𝐷 = (1; 0) 𝑟𝑃𝐷 = (1; 1) 𝑔𝑃𝐷 = 0.5
𝑛⃗𝑃𝐸 = (0; −1) 𝑟𝑃𝐸 = (−0.5; −1.5) 𝑔𝑃𝐸 = 1/3
𝑛⃗𝑃𝐹 = (−1; 0) 𝑟𝑃𝐹 = (−1.5; −0.5) 𝑔𝑃𝐹 = 2/3
2
a) Show, with all steps, that 𝑔𝑃𝐶 = 3.

9
b) Determine the Cell ‘average’ Gradient for cell P. Verify that:
̅̅̅̅̅̅̅̅ 1 1 1 1
∇(𝜙)𝑃 |𝑥 = − 𝜙𝐴 + 𝜙𝐶 + 𝜙𝐷 − 𝜙𝐹
4 6 4 6
̅̅̅̅̅̅̅̅ 1 1
∇(𝜙)𝑃 |𝑦 = 𝜙𝐵 − 𝜙𝐸
3 3

Formulae
1 1
̅̅̅̅̅̅
∇𝜙 𝑃 = ∫ ∇𝜙 𝑑𝑉 ̅̅̅̅̅̅
∇𝜙 𝑃 = ∫ 𝜙 𝑑𝑺
𝑉𝑃 𝑉𝑃 𝑉𝑃 𝜕𝑉𝑃
1
̅̅̅̅̅̅
∇𝜙 𝑃 𝑉𝑃 = ∑ ∫ 𝜙 𝑑𝑺 ̅̅̅̅̅̅
∇𝜙 𝑃 = ∑ ̅̅̅
𝜙𝑓̅𝑺𝒇
𝑓𝑎𝑐𝑒 𝑉𝑃
𝜕𝑉𝑃 𝑓=𝑛𝑏(𝑃)

10
c) Regarding the expressions above, justify:
- the only dependence on the variables appearing in the expressions
- the absence of 𝜙𝑃 in the expressions; and the reason why 𝜙𝑃 is not present. In general
it has a minor, but existing role
- the reason why this gradient is not considered to fully determine the diffusive fluxes
across the control volume

d) Using the generic function 𝜙(𝑥; 𝑦) = 𝑥 + 𝑦, show that the cell gradient is not second order
accurate. Justify that and the reasons behind.

11
e) Considering the transport equation ∇ ∙ (𝑈 ⃗ 𝜙) = ∇ ∙ (∇(𝜙)), apply the Finite Volume
method taking in mind the cell P and any uniform velocity field. Define the velocity field
illustrating it, clearly, with a vector in the sketch below. Discretize the transport equation
in the orange face, between cells P and C. For that consider first order upwind for
convection and no gradient corrections for diffusion.

12
Name: ________________________________________________________________________ Number: ______________

Problem 4 (20 x 0.2 = 4.0 Marks) Wrong answers discount 50% (-0.1 Mark)
Indicate if the statements below are True or False.
T F
1 von Neumann stability analysis can be used in linear problems with Periodic
boundary conditions
2 Bowyer-Watson algorithm triangulation may not ensure a Delaunay
triangulation
3 Rhie-Chow interpolation scheme is used to enforce the pressure-velocity relation
at the cell faces
The basic iterative methods for the solution of linear systems of equations can be
4 written in terms of an operator relating two consecutive iterations. If the spectral
radius (𝜌) of this operator is higher than 1 convergence may be accelerated
5 The discretization of the 1D linear wave equation with upwind schemes display
amplitude errors and phase errors
For the convection-diffusion equation, an essential requirement for boundedness
6 is that all coefficients of the discretized equations, 𝑎𝑗 , should have a negative sign:
𝜙𝑖 = ∑ 𝑎𝑗 𝜙𝑗
7 Mesh influences the eigenvalues of the iterative matrix but not the convergence
rate of the iterative method to solve the system of equations
8 CFL condition is necessary and sufficient for numerical stability
A necessary condition for the convergence of a finite difference method for a
9 hyperbolic PDE is that the mathematical domain of dependence contains the
numerical domain of dependence
10 For the constant coefficient wave equation, the numerical scheme should be
dispersive enough to match with the exact dispersion relation
11 First order upwind and central differences generates dispersion errors for the
linear wave equation
12 Only linear schemes can be made to be monotone while being at the same time
essentially of higher order of accuracy
13 Some limiters has been found to work well for all problems
The non-orthogonality angle is the angle between the face normal vector and the
14 vector connecting cell centroid and it leads to an explicit correction term in the
diffusion/Laplacian term that increases stability and deferred correction is
usually selected
For compressible flows the SIMPLE density correction modification of the
15 pressure correction equation was made to include a convection like term that
changes its type of pressure correction from hyperbolic to elliptic
16 1D Burger’s equation only displays contact discontinuities while Linear 1D
advection equation may display shocks and expansion waves
17 Inflow boundary conditions needed in the simulation of 1D Euler compressible
flows are different depending on the subsonic or supersonic nature of the
incoming flow
Godunov method employs a piecewise-constant reconstruction and evolves
18 according to the exact Riemann problem solution at interfaces of the control
volumes
For linear advection equation, despite the use of an exact solution for the
19 Riemann problem of the linear numerical equation, the resulting numerical
Godunov method is high order accurate both in time and space
20 The Godunov's method considers at each face all 10 possible wave patterns in the
solution of the Riemann problem for 1D Burgers’ equation

13
14