CE-203

FLUID MECHANICS

CFD-Computational Fluid Dynamics

ASSIGNMENT

Under Guidance of:

Mr. SUBASHISA DUTTA
Course Instructor:
CE-203: FLUID MECHANICS

PAWAN PATEL NIRANJAN

09010440
Question 1 :- A CFD code is used to simulate flow over a two-dimensional airfoil at an angle of
attack. A portion of the computational domain near the airfoil is outlined in figure 1 (the
computational domain extends well beyond the region outlined by the dashed line). Sketch a coarse
structured grid using four-sided cells and sketch a coarse unstructured grid using three-sided cells in
the region shown. Be sure to cluster the cells where appropriate. Discuss the advantages and
disadvantages of each grid type.

figure 1:-

Solution 1:- We have to sketch

(i) A structured and an unstructured grid near the airfoil

surface.
(ii) Discuss the advantages and disadvantages of each.

In either case it is wise to cluster cells close to the airfoil surface since we expect that a thin
boundary layer will exist along the surface, and we need many tiny cells within the boundary layer to
adequately resolve it.

There are numerous other ways to construct a grid around this airfoil.

Some simple, coarse meshes are drawn in figure:- ans1.

figure :- ans1
ans1 (a):- A coarse structured grid
using four-sided cells.
(a)

ans1(b):- A coarse unstructured grid using three-sided cells

(b)
Notice that the cells are clustered (more fine) near the surface of the airfoil since there is likely to be large
velocity gradients there (in the boundary layer).
 Discussion on Advantages and Disadvantages: The structured grid in figure:-ans1(a ) is
called a C-grid since it wraps around the airfoil like the letter “C”. The main advantage of the
structured grid is that we can get high resolution near the surface with few cells. The main
advantage of the unstructured grid is that it is somewhat easier to generate when the geometry
is complicated (especially for highly curved surfaces). Furthermore, it is easier to transition
between curved and straight edges with an unstructured grid. The main disadvantage of an
unstructured grid is that more cells are required for the same spatial resolution.

Question 2 :- In this exercise, we examine how far away the boundary of the computational domain
needs to be when simulating external flow around a body in a free stream. We choose a two-
dimensional case for simplicity-flow at speed V over a rectangular block whose length L is 1.5 times
its height D (figure :- 2a). We approximate the flow as symmetric about the centre-line (x-axis), so
that we need to model only the upper half of the flow. We set up a semicircular computational
domain for the CFD solution, as sketched in figure:-2b. Boundary conditions are shown on all edges.
We run several values of outer edge radius R (5 < R/D < 500) to determine when the far field
boundary is “far enough” away. Run FlowLab, and start template Block_domain.
(a) Calculate the Reynolds number based on the block height D. what is the experimentally
measured value of the drag coefficient for this two-dimensional block at this Reynolds number?
(b) Generate CFD solutions for various values of R/D. For each case, calculate and record drag
coefficient CD. Plot CD as a function of R/D. At what value of R/D does CD become independent of
computational extent to three significant digits of precision? Report a final value of C D, and discuss
your results.
(c) Discuss some reasons for the discrepancy between the experimental values of C D and the value
obtained here using CFD.
(d) Plot streamlines for two cases: R/D = 5 and 500. Compare and discuss.

figure :- 2

Solution 2:- We are to generate CFD solutions for external flow over a 2-D block. Specifically, we
are to compare drag coefficient for various values of R/D (the extent of the outer boundary of the
computational domain). In addition, we are to compare the calculated value of C D with experiment.

Assumptions:-
i) The flow is two-dimensional and incompressible.
ii) The flow is symmetric about the x axis.
iii) The flow is turbulent, but steady in the mean.

The fluid is air with ρ = 1.225 kg/m3 and μ = 1.7894 × 10-5 kg/m⋅s.

Analysis:-

(a) The Reynolds number is

Re = ρVD/ µ
= {(1.225 kg/m 3 )(2.0 m/s)(0.10 m)}/{1.7894 x 10-5 kg/m.s}
= 1.37 x 10 4
Experimental data indicate that the drag coefficient for this body is CD ≈ 1.9 at Reynolds numbers
greater than 104 .

(b) The CFD code is run for eight values of R, all else being equal. C D is tabulated as a function of R/D
in Table 1, and plotted in figure:- ans2. As the extent of the computational domain grows in size, the
drag coefficient decreases steadily, but levels off to three significant digits of precision by R/D ≈ 200.
Thus, a computational domain extent of R/D ≈ 100 is sufficient to achieve independence of C D. We
report a final value of CD = 1.34.

TABLE 1
Drag coefficient as a function of the normalized
extent of the computational domain for turbulent flow
over a rectangular block.
R/D CD
5 1.81927
10 1.50662
20 1.41076
50 1.36723
100 1.35282
200 1.34671
500 1.34408

figure:- ans2
Drag coefficient plotted as a function of the normalized
extent of the computational domain for turbulent flow
over a rectangular block.

(c) There are several possible reasons for the discrepancy between the calculated value of C D (1.34)
and the experimentally obtained value of CD (about 1.9). First of all, the actual flow is most likely
unsteady, with vortices being shed into the wake, whereas we are simulating a steady flow. In
addition, the unsteady shedding of vortices renders the flow no longer symmetric about the x axis,
whereas we are forcing our flow to be symmetric. Furthermore, the grid resolution may not be
adequate to achieve grid independence (this is checked in the following problem). Finally (and most
importantly), we are using a turbulence model to simulate this flow field. The CFD solution we obtain
is only as good as the degree to which the turbulence model correctly models the physics of the
turbulence. As discussed in the text, no turbulence model is universally valid for all types of
turbulent flows. Discrepancies between experiment and CFD will always exist regardless of how fine
the grid or how large the extent of the computational domain.
(d) Streamlines near the body are plotted for R/D = 5 and 500 in figure:-ans2. We notice that the
streamlines for the R/D = 5 case are more tight around the body compared to those for R/D = 500.
This is most likely due to interference from the outer edges of the computational domain, which are
too close for the R/D = 5 case.

Question 3:- Air flows through a conical diffuser in an axisymmetric wind tunnel ( figure :- 3(a)-
drawing not to scale). θ is the diffuser half-angle (the total angle of the diffuser is equal to 2θ). The
inlet and outlet diameters are D 1 = 0.50m and D2 = 1.0m, respectively, and θ=20o. The inlet velocity is
nearly uniform at V = 10.0m/s. The axial distance upstream of the diffuser is L 1 = 1.50m, and the axial
distance from the start of the diffuser to the outlet is L 2 = 8.00m. We set up a computational domain
for a CFD solution, as sketched in figure:-3(b). Since the flow is axisymmetric and steady in the mean,
we model only one two-dimensional slice as shown, with the bottom edge of the domain specified
as an axis. The goal of this exercise is to test for grid independence. Run FlowLab, and start template
Diffuser_mesh.
(a) Generate CFD solutions for several grid resolutions. Plot streamlines in the diffuser section for
each case. At what grid resolution does the streamline pattern appear to be grid independent?
Describe the flow field for each case and discuss.
(b) For each case, calculate and record pressure difference ΔP = Pin – Pout. At what grid resolution is
the ΔP grid independent (to three significant digits of precision)? Plot ΔP as a function of number of
cells. Discuss your results.

Figure :- 3

Solution 3:- We are to generate CFD solutions for several grid resolutions to test for grid
independence for flow through a diffuser. Specifically, we are to compare streamlines and pressure
difference at each level of grid resolution.

Assumptions :-
(i) The flow is incompressible.
(ii) The flow is axisymmetric about the x axis.
(iii) The flow is turbulent, but steady in the mean.

Analysis :-

(a) Streamlines are plotted for six grid resolution cases in figure: - ans3.1.
figure :- ans3.1 :-
Streamlines for steady, incompressible, axisymmetric, turbulent flow through a diffuser at various levels of grid
resolution; the number of cells is (a) 880, (b) 3520, (c) 4950, (d) 8800, (e) 13750, and (f) 19800.

At the very coarse grid resolution, the streamlines are not well defined (figure:- ans3.1(a)), and the
calculation is not reliable. As grid resolution improves, details of the flow separation region become
more refined – the boundary layer is unable to remain attached in such a strong adverse pressure
gradient. From these plots, it appears that grid independence has been achieved by about the fourth
case (figure:- ans3.1(d)), beyond which there is no noticeable change in the shape of the streamlines.
We note that simulation of flow separation and separation bubbles is often a very difficult task for a
CFD program. In this particular problem we must use an extremely fine grid in order to resolve the
details of the flow separation.

(b) ΔP is tabulated as a function of cell count in Table 2. As grid resolution improves, ΔP increases,
and becomes independent of grid resolution by the fourth or fifth mesh. This is also seen in
figure:- ans3.2 where ΔP is plotted as a function of cell count.

TABLE 2
Pressure difference from inlet to outlet as a function of
cell count for a conical diffuser with θ = 20o.
Cell count ΔP (Pa)
880 -34.5622
3520 -32.7267
4950 -32.4183
8800 -32.2021
13750 -32.1162
19800 -32.1099
figure:- ans3.2
Pressure difference as a function of the number of cells
in the computational domain. Turbulent flow through
an axisymmetric diffuser.

Discussion:- Newer versions of FlowLab may give slightly different results. The unphysical-looking
streamlines at the very coarse grid resolution (figure:- ans3.1(a)) are due to interpolation errors
when the CFD code calculates contours of constant stream function. Notice that even though there
is gross flow separation in this diffuser, there is still a pressure recovery through the diffuser (P in is
less than Pout). A better design (with even higher pressure recovery) would use a smaller diffuser
angle so as to avoid flow separation along the diffuser wall. The outlet of the computational domain
is reasonably far downstream (several pipe diameters) to avoid reverse flow at the outlet.