Вы находитесь на странице: 1из 10

ME 450/550 Computational Fluid Dynamics

CFD Final Project: Driven Cavity


Submitted to: Professor Tao Xing (xingtao@gmail.com)

Name: Cristofer Farnetti


University ID: 111-67743
E-mail: farn7259@vandals.uidaho.edu
Department: UI Mechanical Engineering

Date: 05/11/2016
I.

Problem Description and Simulation Design


Taking a look at flow characteristics in a driven cavity is a fundamental problem for those
looking to employ computational fluid dynamics (CFD) techniques. This analysis is on a 1m by

1m driven cavity with the top wall moving at a constant 1m/s. The effects of Reynolds number
will be assessed by running simulations using Re= 100 and Re=1000. The approach to achieving
a solution will be to utilize the Ansys Workbench 16.2 software suite. In this analysis the solution
obtained from Ansys will be the y-coordinate location of the main vortex in the driven cavity,
and this solution will also be used for a single grid triplet study at each Reynolds number for
verification and validation purposes. For the grid uncertainty, the Factor of Safety Method was
used. The purpose of this study is also to take a look at the velocity magnitude, vorticity
magnitude, velocity vectors, streamlines and pathlines, and to compare the u velocity along the
vertical centerline and the v velocity along the horizontal centerline with their respective
experimental values. For the geometry and coordinate system see Figure 1 below.

Figure 1. Driven Cavity Geometry

II. CFD Process


A list of the CFD process characteristics is displayed below and gives insight to how Ansys
Workbench was used to solve for the driven cavity geometry.

III. Data Analysis and Discussion


A.

Flow at Re = 100

Figure 1. Residuals

Figure 2. Contour of vorticity magnitude

Figure 3. Contour of velocity magnitude

Figure 4. Streamlines

Figure 5. Pathlines

Figure 6. Velocity vectors

Figure 7. Plot of u velocity along the vertical


centerline CFD vs EFD

Figure 8. Plot of v velocity along the horizontal


centerline CFD vs EFD

Table 1. Grid triplet study on the solutions for the main vortex location

Figure 9. Plot of the solutions for the location of


the main vortex

Figure 10. Plot of the error of the solutions for the


location of the main vortex

B.Flow at Re = 1000

Figure 1. Residuals

Figure 2. Contour of vorticity magnitude

Figure 3. Contour of velocity magnitude

Figure 4. Streamlines

Figure 5. Pathlines

Figure 6. Velocity vectors

Figure 7. Plot of u velocity along the vertical


centerline

Figure 8. Plot of v velocity along the horizontal


centerline

Table 1. Grid triplet study on the solutions for the main vortex location

Figure 9. Plot of the solutions for the location of the


main vortex

Figure 10. Plot of the error of the solutions for the


location of the main vortex

Along with the main vortex, what other vorticies were there? How did this change with Re?
Looking at Figures 4A, 4B, 5A, and 5B for the pathlines and streamlines in the driven
cavity geometry there are several features that are mentionable. For Re=100 the only resolved
vortex is the main vortex, but for Re=1000 there are two other vorticies located in the bottom
corners of the cavity. Additionally, about of the way up the wall on the left side there seems to
be another vortex forming. The resolution of the vorticies could likely be improved by either
increasing the number of grid points, or introducing zones with a bias factor that can better
resolve the vorticies near the wall and at the predicted location of the main vortex. This would
result in a cluster of grid points near the walls and just barely right of the center of the cavity. I

am unsure of how this would affect the solution for the y-location of the main vortex and
whether it would increase or decrease the solution error.
How does the mesh size effect the solution? What was the behavior of the error between Re
and which is larger? Is this error negligible?
Mesh size in a uniform grid is a critical component to obtaining accurate solutions. For
Re=100 mesh 1, 2, and 3 generated solutions of 0.738114m, 0.7378614m, and 0.7363607m
respectively. The changes in solution result in an error of 0.505%, 0.471%, and 0.267% for mesh
1, 2 and 3 respectively, which means that for the case of Re=100 a decrease in mesh size
generated a lower error. This is contrary to intuition and one would think that an increase in grid
points would decrease the error. For Re=1000, mesh 1, 2, and 3 gave solutions of 0.565849m,
0.5675077m, and 0.5743191m respectively. The change in mesh size resulted in an error of
0.115%, 0.408%, and 1.613% for mesh 1, 2, and 3 respectively which means that the error
increases as the number of grid points is reduces, which is what one would think should happen.
An important note to make is that the error is very low and while mesh size does play an
important part in the solution, the solution is nearly independent of mesh size after a certain
number of grid points is established.
Verification and validation: How do the solutions for the different Re converge? How do
you know? Can validation be obtained? What was the grid uncertainty for the different
Reynolds numbers?
When looking at verification and validation studies in Tables 1A and 1B there are several
things to discuss such as the convergence behavior and validation attainment. For both Reynolds
numbers, the solution monotonically converges given that R= 0.167 for Re=100 and R=0.244 for
Re=1000 (monotonic convergence: 0 < R < 1). A note to make on validation is that it cannot be
determined for certain unless the EFD data uncertainty is known. The validation of Re=100
model cannot be attained since the grid uncertainty (U_G(%S1)= 0.043%) is less than the mesh 1
solution error (|E|= 0.505%). This aside, the value for the experimental uncertainty would have to
just be greater than 0.5032% for Re=100, which isnt difficult to imagine being the case therefore
it isnt entirely unreasonable to estimate that the model is validated. Validation is achieved for
Re=1000 since the grid uncertainty (U_G(%S1)= 0.1803%) is greater than the mesh 1 error for
the solution (|E|= 0.115%).
How well did the solutions for CFD match with EFD on the centerline velocities?
It is important to qualitatively discuss how well the CFD matched the EFD for the
centerline velocities. For the u velocity along the vertical centerline at Re=100, CFD seems to
slightly over-predict from 0<y<0.5, but it agrees very well from 0.5<y<1. Additionally, v
velocity along the horizontal centerline seems to slightly under-predict from about 0.1<x<0.3 and
over-predict from about 0.4<x<0.9. For Re=1000, the u velocity seems to be slightly overpredicted from 0<y<0.9 and agree very well from 0.9<y<1. The v velocity along the horizontal
centerline seems to be over-predicted by CFD from about 0<x<0.9, but agree very well from
about 0.9<x<1.

IV. Conclusions
A. Summarize your findings and relate them to classroom lectures or
textbooks
For this simulation on the driven cavity problem at Re=100 and Re=1000 with
different mesh numbers there were many interesting findings, but some are most
notable such as the effect of mesh number on solutions, the agreement of CFD and
EFD, number of vorticies for the different Reynolds numbers, and agreement with
the literature. The effect of mesh size on the solutions was interesting because of the
behavior from Re=100 to Re=1000. For Re=100, as the mesh size decreased from
161x161 to 41x41 the solution became even more accurate while in contrast the
Re=1000 simulation showed that as mesh size increased the solution became more
accurate. The Re=1000 simulation is the kind of behavior that one would expect to
happen for both Reynolds numbers. Dr. Tao Xing once mentioned a case where a
solution was obtained for a geometry and when the mesh was refined even more the
solution began to become more and more inaccurate, this could be the case with the
Re=100 flow although a reason for this behavior is difficult to pinpoint.
Overall the CFD solutions for the centerline u and v velocities agreed very well.
The CFD over-predicted very slightly for most of the Re=1000 simulation, but the
difference between EFD and CFD was so small that it can be considered negligible.
The Re=100 simulations both over-predicted and under-predicted but overall it agreed
very well and the differences in velocities was negligible from CFD to EFD.
The different flows presented by the different Reynolds numbers gave some
interesting characteristics. For the Re=1000 simulation, there were three vorticies and
what looked like the beginning of another vortex, but given the coarseness of the grid
it could not be resolved. Additionally, for the Re=100 there was only one vortex (the
main vortex) which was resolved. The number of voriticies and their locations agree
very well with the Ghia-1982 resource provided except for the Re=100 where there
should have been an additional two more vorticies in the bottom corners which were
not resolved by this simulation. Ghia-1982 had a uniform grid of 129x129 and were
still able to resolve the vorticies for the Re=100 and I was not able to with a 161x161
grid. Perhaps a bias factor that allocates more points near the walls and corners of the
driven cavity would help.
Figures 2A and 2B of vorticity magnitude also depict a flow that agrees well with
the literature from Ghia-1982. If the contour levels are set equal to the literature, the
figures become identical with the exception of a color gradient presented by the
simulations in this report.

B. Comments
Given the driven cavity problem was a fun simulation to run because it gave us
the chance to freely use Ansys to come up with a solution, and being able to see the
agreement with the literature was really rewarding. The interface provided by Ansys
was intuitive and made running the simulation simple without any outside help.

C. Suggestions and improvements

At first, I thought that the driven cavity problem would be difficult but since I had
the background from running other simulations it was actually very easy to obtain
accurate solutions very quickly. Something that I would have enjoyed as a final
project would have been a turbulent simulation or a simulation with some shedding
vorticies. For example, maybe a flat plate that is perpendicular to the flow. This
would undoubtedly present some complexity, but it would be rewarding nonetheless.
One last comment, giving a brief tutorial on how to use tecplot could help the class a
lot since Ansyss post processing software isnt very user friendly in my opinion.

Вам также может понравиться