CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   STAR-CCM+ (http://www.cfd-online.com/Forums/star-ccm/)
-   -   Problem with convergence residual (http://www.cfd-online.com/Forums/star-ccm/87846-problem-convergence-residual.html)

 tom.j.87 April 30, 2011 12:00

Problem with convergence residual

3 Attachment(s)
Hi, Iīm student, now Iīm working on my thesis work on topic Car body. I have problem with value of residual.

I created model of sports car (half model exactly) and wind tunnel (size tunnel: 7 metres width, 6 metres high, 90 length). This all was imported to Star-ccm+. Mesh have 1 241 000 cells (polyhedral).

Flow conditions are: Three dimensional, Gas, Cell quality remediation, Coupled flow, Motion are stationary, Constant density, Time is steady, Viscous regime was selected Turbulent, Reynolds - Averaged Navier Stokes, K - Omega turbulence, SST (Menter) K-Omega, All y+ Wall Treatment.

On the start of tunnel is condition: velocity inlet (28 km/h, and in the end of tunnel is condition pressure outlet (0 Pa). Car and road have condition Wall, other side of tunnel have condition Symmetry.

Program calculted residuals as you can see on picture. I read somewhere about boundary of convergence is to be 10-4. During 800 iterations curves residual are decreased, and after were not change to the 2800 iteration. Then I stopped it . Iīm tried change turbulent flow on laminar on 2850 iterations. Iīm tried show of flow (pics enclosed), and I donīt know if the results with this mistake residuas can be right. My computer calculated this results 2 days, approximate numb is 50 iteration/hour (Pentium i3, 8gb ram)

 famerfamer April 30, 2011 17:38

Come on, your total mesh size is too small comapred to your geometry. And you were using your own computer to run the simulation.....

 abdul099 May 1, 2011 07:20

I'm feeling like a scratched record, I'm writing the same again and again:
You shouldn't give too much on the residuals. They are normalized during the first 5 iterations and therefore the final level strongly depends on the initial values.
Further you can see some oscillations in the residuals. That's an indicator for a unsteady nature of the flow while you are running the case steady state. That might be another reason why residuals drop less than you expected. That should not mean "you have to run it unsteady", that should mean "don't give too much on residuals".

And further I have to agree to famerfamer. The mesh is way to small to capture all flow phenomena. Just to give you an order of magnitude, a F1 car usually has about 100 Million cells, and even that still gives no mesh-independend solution. Anyway, that much will not be realistic for you, but your machine can handle at least six times as much as you are using now. I now, that will slow down your simulation, but that's how CFD works - using much computational power or waiting for ages.

Two other things that came into my mind: Have you applied rotation on the tyres and motion on the ground? How does the contact between tyres and road look like?

 tom.j.87 May 1, 2011 08:05

3 Attachment(s)
Hi abdul099, thanks for asnwer,

I tried rotation on tyres, motion on road I applied too (I forgot to upload). Mesh size of car is 22 mm, and I diluted the size of the tunnel elements by 7 blocks. (firts block mesh size is 50 mm, next 100, next 200, ... and last block have mash size 500 mm). I have to do somothing correction on design car body, so I need more accurately results. I proceeded through the thesis work of my teacher and there is written about the conditions convergence as I wrote. Now I tried to change the length the tunnel on 55 metre (from 90 metres). The residual curves increased a lot too (attached screen).

Donīt you have any idea? If I understood correctly, there is the problem with number of cells car compared the cells tunnel? My teacher told me that I canīt get number 2 milions cells.

 tom.j.87 May 1, 2011 09:46

I thought if I donīt fault in the direction speed of the road. I changed direction road speed against the direction of the x-axis. Before it was reversed. And the direction rotation tyres it was the same. I also change (direction rotation tyres).

 abdul099 May 1, 2011 11:14

Why can't you get even up to 2 million cells? Of course, it will take longer, but without an appropriate mesh resolution, you will not get good results. For me it sounds like
"uh, I need a car with a top speed of 300km/h. But the 5l V8 needs to much fuel, so let's try it with a scooter engine. When it doesn't reach the desired top speed, let's complain in a forum, for sure some smart guy will have some hints how to tune the scooter engine to reach a top speed of 300km/h".
It's up to you what mesh size you're using - but no hint can make a miracle to occur.

To judge convergence, you shouldn't bother about residuals which decrease not enough or maybe increase a little bit. As I mentioned before, the residuals are normalized within the first 5 iterations. Imagine, you would initialize with a "perfect" solution, therefore the non-normalized residuals would be very low. When starting the simulation from that perfect state, the residuals are normalized and they will never drop. So you can't judge convergence just by looking on the residuals!
You should have some plots for drag, lift or whatever value you are interested in. Look at this plots, when they reach a continous level and the residuals have stabilized (at what level ever), you can stop your simulation.

In general, I would use the longer wind tunnel. The cells can grow a lot towards the end, so the impact on the cell count will be very low. It's also necessary to dissipate eddies, pressure changes etc. before they reach the outlet which will be done by the growing cells.

 tom.j.87 May 1, 2011 11:49

Ok, I understand what you mean. Now I run the calculation, but the program reports me message on output message: Outlet: reversed flow on 17 faces. I think that I have short tunnel (90m). I will try to increase length tunnel at 200 meters, for sure. Thanks for your advice :) Regards, Tom.

 abdul099 May 1, 2011 13:47

Reversed flow is not that problematic, but you can also prevent it by increasing the cell size near the outlet. This will dissipate vortices etc. which I assume are the cause for the reversed flow at the outlet.

Good luck ;)

 tom.j.87 May 1, 2011 15:43

5 Attachment(s)
Quote:
 Originally Posted by abdul099 (Post 305795) Reversed flow is not that problematic, but you can also prevent it by increasing the cell size near the outlet. This will dissipate vortices etc. which I assume are the cause for the reversed flow at the outlet. Good luck ;)
Last question, Iīm sorry, that I ask so stupidly. Can I ask what sign would you choose for direction the rotation tyres and road speed to coordinate system? x-axis is directed against the direction ride, y -axis is directed to the center car.

I opted for the speed of the road a positive sign, this means that the speed of the road has the same direction as the x-axis. Direction of velocity has a negative direction (-90,032 rad). I used the enclosed picture. These directions are as I think it should be. So that now computing.

In thesis works other students is the x-axis is directed in the direction ride, y -axis is directed to the center car. But they have positive sign, as rotation of the wheels and direction of the road. So they have the opposite sign than I when now. When I followed the same principle as they are, so it suited me message: reversed flow.

So what rate ( + or - ) the road sign would you choose? (to boundary) This is last question, perhaps. :) Tom

 abdul099 May 3, 2011 03:30

The velocity of the road seems reasonable to me. It simply the opposite direction the car would travel, therefore the opposite direction of the incoming flow (when running without yaw).

To get the wall rotation of the tyres, use the right-hand rule. Point your thumb in the same direction as the desired rotation axis which will be the Y-axis (positive direction of Y!) in your case. Now the curvature of your relaxed fingers will show you in which direction the wall would rotate when a POSITIVE rotation rate would be applied. In your case, this is obviously the wrong direction, as the tyres would rotate against the road. Therefore a negative rotation rate is necessary in your case. You already did it right. It's possible to check it by plotting not the velocity magnitude in the scalar scene but the velocity in X-direction.

What diameter do your tyres have? I'm just wondering about the rotation rate.

 All times are GMT -4. The time now is 11:48.