You are on page 1of 11

Registration Number: 180218264

MEC320
Computational Fluid Dynamics
Aerospace Engineering
Assignment 2: Numerical Simulation Results
05/19/2021
Dr Alberto Marzo and Prof. Shuisheng He
180218264 MEC320

Computational Domain and discretization process


The solution domain for the backward facing step was designed using the Space Claim in Ansys Fluent.
The geometry consisted of a backward facing step of height 1.27 cm. The height of the solution domain
for the internal flow inside the tunnel 11.4 cm. There were four different boundaries for this geometry.
The boundary conditions are stated below:

• Inlet : Velocity inlet, with velocity normal to boundary of value 44.2 m/s.
• Outlet : Pressure outlet
• Step-side wall : Non-slip wall boundary
• Opposite wall : Non-slip wall boundary

The upstream inlet length was estimated using the following equation :
𝛿
𝑥= 1 (1)

0.16 𝑅𝑒𝑥 7

Where, Rex = 5000 , which is the Reynolds number based on momentum thickness. 𝛿 = 1.9𝑐𝑚, which is
the boundary layer thickness. Using these values provided us an estimate of the inlet length of about
40.1cm. The outlet length was estimated to be about 50 times the step height, which was about 60cm.

Following the construction of solution domain, the solution was discretized using finite volume method.
This method directly deals with the Navier-Stokes equation and does not require mathematical continuity.
In this method, the domain is divided into discrete control volumes and the governing equation is
integrated over the control volume. This gives a discretised equation at its nodal point. The integrated
transport equation of a one-dimensional convection and diffusion properties over a control volume is,
𝑑𝜙 𝑑𝜙
(𝜌𝑈ϕ)𝑒 − (𝜌𝑈ϕ)𝑤 = (𝛤 ) − (𝛤 )
𝑑𝑥 𝑒 𝑑𝑥 𝑤 (2)

Here, 𝜙 is the boundary values of interest and 𝛤 is the diffusion coefficient. The gradient (d𝜙/dx) is located
at the east(e) and west(w) of the control volume. Interpolation is required to calculate the values of
gradients and fluxes. In practical calculations we can only work with a finite and limited number of cells.
Therefore, the numerical results will only be close to reality only when the interpolation scheme satisfies
properties such as conservativeness, boundness and transportiveness. Central differencing scheme is not
considered here due to its limitations with not satisfying one of the requirements for boundness and
transportiveness property. This means the scheme cannot correctly identify the flow direction. Since,
second order accuracy provides higher order of accuracy for our solution, second-order upwind scheme
was selected. This method computes the unknown quantities at cell faces using a multi-dimensional linear
reconstruction approach. This approach helped us get the solution closer to the analytical value as first-
order upward scheme has high dissipation and often results to less accurate solution. However, this
scheme introduced unphysical extreme values as it is an unbounded scheme. Therefore, a limiter scheme
was used in combination with this scheme to address this problem.

2
180218264 MEC320

Fig 1: Mesh generation with 14400 nodes

As shown in fig.1, the mesh was refined in the areas of interest, such as the boundary layer region,
separation and reattachment regions. Pure quadrilateral structured mesh was used for this simple 2D
geometry. In a 2D structured mesh each node has a corresponding integer i and j index values, which are
unique and can be used to locate neighboring cells. The flow domain was sub-divided into three sub-
regions, and each of these regions were meshed separately using edge sizing. To capture the areas of
interest better the mesh was refined using edge biasing. This feature allowed the mesh to be defined such
that it is forced to be finer near the step-side wall using a large to small bias. A more complicated bias,
small to large in the middle and back to small at the other end of the edge was used in the step wall to
capture the recirculation and reattachment zones better. Face sizing was also used to control the mesh
size at all the faces of the domain.

Mesh-independent study is performed for this solution domain to determine the optimum mesh at which
we can get a solution very close to the empirical solution with expense of low computational resources.
For this study, three simulations are performed for coarse mesh (1146 nodes), medium mesh (14400
nodes) and fine mesh (65650 nodes) respectively to ensure the results are independent of the mesh. The
monitoring value chosen for this study is the velocity along the surface of the domain because velocity is
more sensitive to the mesh variance due to the changes in upstream and downstream velocities. The
results are plotted in figure 2, which gives a graphical representation of mesh performance. The difference
between the fine mesh and medium mesh is very small and the lines are extremely close to each other.
This shows that we have reached mesh independence and further refinement in mesh will not change the
outcomes significantly. In order to reduce the simulation time and computational costs, medium mesh is
used for the study of the domain.

3
180218264 MEC320

Velocity vs x-position for different meshes


50
45
40
35
Velocity (m/s)

30
25
20
15
10
5
0
0.4 0.5 0.6 0.7 0.8 0.9 1
Position(m)
Medium Mesh Fine Mesh Coarse Mesh

Fig 2: Determination of mesh-independent solution

Solution Process
Iterative method was used in the solution process. This process is based on the repetition of algorithm
until convergence is met. The solution of the transport equations for velocity components presents us
with the problem of pressure-velocity coupling. The velocity component appears in each momentum
equations and satisfies the continuity equation, while the pressure component only appears in the
momentum equations. For an incompressible flow, pressure cannot be obtained directly using the
equation of state. This introduced a coupling problem between velocity and pressure, where a known
pressure field applied to the momentum equation requires the resulting velocity field to satisfy continuity.
This issue of non-linearity is addressed by following a pressure-correction method.

Several methods were available in Fluent, among which SIMPLE algorithm was used in this problem. This
is a basic guess-and-correct procedure, which is used to guess a pressure field that will also satisfy the
momentum field. The process was iterated until the convergence of velocity and pressure fields. The
SIMPLE algorithm is quick with low computational costs. It also allows to perform a greater number of
iterations quicker compared to PISO or Coupled methods. Since, the pressure correction equation is
susceptible to divergency, an under-relaxation factor was used during the iterative process. Values of
under-relaxation factors less than 1 were used to ensure convergence and stable computations. Using a
larger under-relaxation factor would result in oscillatory or even divergent iterative solution and a very
small factor would result in extremely slow convergence. Default values in Fluent were enough to provide
a converged solution.

4
180218264 MEC320

Fig 3 : Residuals monitoring plot

The turbulence model used was a viscous SST k-ω model. The residual monitor was used to measure the
error in solving the conservation equations. Each cell in the model has its own residual value as shown in
fig. 3. There were a total five residuals contributed from the continuity, momentum and turbulence
model. Since, for an iterative solution, the residuals can never reach zero, so the solution is deemed
accurate if the residual values are low. The solution convergence criteria were set to 10-6. Furthermore, a
pressure coefficient monitoring point was plotted alongside the residuals to study the critical point of
interest in flow field. The point was set near the step and area-weighted average was used to study the
effects of adverse pressure gradient in the flow field. The monitoring point can be used to further validate
that our solution has converged as shown in figure 4 with a steady-state solution.

In order to initialize the solution, hybrid initialization was used because it normally speeds up the overall
computation time. The solution was then run and calculated for 1000 iterations. The convergence was
met at around 900 iterations to a convergence criterion of 10-6. This value of RMS residuals can be
considered as tightly converged and upon inspection from the residual and monitoring plots we can
conclude that the solution is converged.

5
180218264 MEC320

Fig 4 : Pressure coefficient monitoring plot

Validation of results

Validation helps us determine the degree of accuracy of the model to real world depending on the uses
of the model. We can quantify the uncertainties using this process. Contour plots of pressure and velocity
can be used to validate our results qualitatively as shown in figure 5 and 6. From fig.5, we can see the
regions of high and low pressures. The pressure abruptly drops when the upstream boundary layer
separates at the edge of the step. The pressure is low around the recirculation region near the step. The
shear layer can be seen getting larger as the flow proceeds downstream of the step. Finally, the flow
attaches at reattachment point on the step-side wall. The pressure starts rising from the reattachment
point as the flow moves further downstream towards the outlet. The pressure is constant throughout the
freestream flow as there is no disruption due to pressure gradient in the flow. From fig. 6, a similar drop
in velocity can be seen in the step region, where the flow separates. The flow downstream is relatively
slower than the flow upstream due to separation near the step. This matches with the experimental
results.

6
180218264 MEC320

Fig 5: Contour plot of total pressure

Fig 6: Contour plot of velocity

For the quantitative validation of our results, the numerical results obtained from this analysis was
compared with the experimental data. Two variables chosen for this analysis are the pressure coefficient
and skin friction coefficients. These parameters are plotted against the normalized position in figures 7
and 8, respectively. From fig.7, we can see that the numerical data is skewed slightly to the right compared
to the experimental data. Similar observation can be made from fig.8, where the numerical data is skewed
to the right. This inconsistency with the experimental data can be caused due to additional uncertainties
in the analysis. However, the difference between computed and experimental data is relatively small and
we can subjectively conclude that the model is validated. With access to larger number of test values from
the experimental data, we can determine a clearer difference against the computed value.

7
180218264 MEC320

0.2
0.15
Pressure Coefficient (Cp)

0.1
0.05
0
-20 -10 -0.05 0 10 20 30 40 50 60 70 80 90

-0.1
-0.15
-0.2
-0.25
-0.3
x/H
Numerical Result Experimental Data

Fig7 : Plot of Pressure coefficient against Normalized position for step-side wall

0.01

0.008
Skin-friction coefficient (Cf)

0.006

0.004

0.002

0
-10 0 10 20 30 40 50 60 70 80 90
-0.002
x/H
Numerical data Experimental data

Fig 8 : Plot of Skin-friction coefficient against Normalized position

8
180218264 MEC320

Further analysis is done here by changing the nominal value of inlet velocity and height of the step by
+10%. The plots of contour and velocity vector is displayed to illustrate the effect of increasing these
parameters. From figures 9, 10 and 11, we can see that the reattachment length is moved further
downstream by increasing the nominal values. The size of the recirculation region is also increased with
large formation of primary and secondary eddies near the corner of the step.

Fig 9 : Contour plot of pressure with +10% change

Fig 10 : Velocity vector plot with +10% change

9
180218264 MEC320

Fig 11 : Velocity vector plot of the original solution

10
180218264 MEC320

References

[1] H.K, Versteeg and W Malalasekera, An Introduction to Computational Fluid Dynamics: The Finite
Volume Method. Essex, England, UK : Prentice Education Limited, 2007.

[2] D.M. Driver and H.L. Seegmiller, “Features of a Reattaching Turbulent Shear Layer in Divergent
Channel Flow”, Moffett Field, California, USA, 1985. [Online]. Available: https://doi.org/10.2514/3.8890

[3] Ansys, Inc., Ansys Fluent Users Guide. Canonsburg, Pennsylvania, USA, 2013.

[4] Leap Australia, “Tips & Tricks : Size Controls in Ansys”.


https://www.computationalfluiddynamics.com.au/size-controls/ (Accessed : May 19, 2021).

[5] M. Kuron, “3 Criteria for Assessing CFD Convergence”. Engineering.com.


https://www.engineering.com/story/3-criteria-for-assessing-cfd-convergence (Accessed: May 19, 2021).

11

You might also like