Need help reducing simulation time

"Node news is good news."

SteveMartin · Apr 1, 2011

What do you consider 'ridiculous' (and corollary: what would be reasonable/acceptable)?

I'm used to running these thermal flow simulations for liquid cooling systems, not for external air flow. My simulation times would run from a few hours (relatively simple steady state) to 'over the weekend' (complex transient behavior). Anything transient was at least 8 hours.

Can you take advantage of symmetry to analyze only 1/2 the geometry?
Do you need a computational domain that large? (I'm not sufficiently familiar with external flow to say it is/isn't the right size)

kellnerp · Apr 1, 2011

It would be nice to see what you are doing. As I mentioned before, you either need to post a couple views or save it in 2009. If you can post some views of the mesh, especially any areas where it gets fine.

Your domain needs to be large enough down stream to allow eddies to form and die out. That may preclude using symmetry although symmetry is permissible in a CFD analysis.

If you are asking about run time please share with us what you are running on and whether you are taking advantage of multiple processors and lots of ram.

You can run a low quality run and see if there are any problems before committing to a higher quality run.

I've seen people spend a week on a FloWorks run. That was steam.

TOP
CSWP, BSSE

"Node news is good news."

kymaks · Apr 1, 2011

Here are three pictures of my model. The large rectangular box represents the pylon of a jet. The small square in the bottom right of that is a 1-in thick plate made of several layers. I made an indent in the pylon to allow the plate to fit flush with the surface of the pylon. The disk represents the jet nozzle. Hot air comes out of it at mach 0.91.

I've uploaded four images of my model and the simulation I'm trying to do. The image that shows airflows is from a simulation in which conduction in solids was disabled. The other three pictures are the simulation I would like to run.

kymaks · Apr 1, 2011

Oops here is the file.

kellnerp · Apr 2, 2011

If you can coarsen the mesh away from the parts that would reduce the model size. At the same time, if there are relatively thin bits of the solid those need to be finely meshed over. I know mesh isn't quite the right word for finite difference, perhaps subdivisions is a better term.

If your exhaust plume is at Mach .91 you might have localized areas with incompressible flow. That is going to make the analysis run longer too.

Another thing is that your domain needs to be shifted with respect to the parts so there is more downstream and less upstream. You probably don't need as much above, below and to the sides of the part either.

You are pushing your flat plat engine through the air. Doesn't that disturb the air flow in unrealistic ways?

TOP
CSWP, BSSE

"Node news is good news."

kymaks · Apr 2, 2011

Should I use the "high mach number" option since the exhaust is traveling at mach 0.91?

I think the flat plate does disrupt the air in unrealistic ways but I don't think it will be a problem for this simulation. My goal is to compare different materials in the plate, such as Aluminum and Titanium, to see how this affects surface heat flux and maximum temperature on the plate.

kellnerp · Apr 3, 2011

What simplification does the small flat "exhaust plate" have over modeling the whole engine as a cylinder with an appropriate inlet? In terms of your computational domain it is a small contribution.

TOP
CSWP, BSSE

"Node news is good news."

kymaks · Apr 4, 2011

I'm not sure I understand your question. Are you asking if there is a reason I modeled the engine as a plate instead of a cylinder? The reason is that I am concerned more with heat flow in the plate than a perfectly simulated flow field. Do you think I should change it?

kellnerp · Apr 4, 2011

Isn't having the engine as a flat plat going to disturb the flow tremendously? Does the flat plat have an inlet on one side and exhaust on the other?

TOP
CSWP, BSSE

"Node news is good news."

kymaks · Apr 4, 2011

The flat plate does have an inlet on one side, yes. Now that you bring it up I might remake it into a long cylinder anyway.

kellnerp · Apr 4, 2011

So back to your original problem, did you shift the domain and attempt to coarsen the grid away from the objects of interest?

Post a picture of the new grid.

TOP
CSWP, BSSE

"Node news is good news."

kymaks · Apr 6, 2011

Here is where I'm at:
I did coarsen the grid and customize the computational domain more. I got a simulation time of about 20 minutes on my 2.8GHz Intel Core i7 laptop with 8gigs of ram. Due to time constraints on this endeavor it will have to do. Now that I know that a 24-hour simulation is not unheard of I am more confident that I am not making a huge mistake somewhere.

I am going to try modeling a large wind tunnel around my parts and run the simulation with internal instead of external flow. I am curious to see if this will reduce the time even more. If so I can use a finer mesh for more accuracy.

I will post results and pictures of the new model.

kellnerp · Apr 7, 2011

I don't think a wind tunnel around you model will help much. The main thing is that your model should be shifted upwind so that there is enough distance down wind for eddies to develop. You can look at the distance around your parts as well in this regard. Don't have more distance around your part than what you need to develop any disturbances to flow.

I am curious as to whether FloWorks is using multiple processors on your system.

TOP
CSWP, BSSE

"Node news is good news."

kymaks · Apr 7, 2011

Flow Simulation is using 2 cores. I believe the Intel Core i7 is effectively a dual-core processor even if it only has one core. That's my non-expert take on it.

The wind tunnel is working better. The min/max table is not showing me unrealistically high mach numbers or low temperatures anymore. Previously there would be certain cells with mach 5 air flow or temperatures way lower than they should be.

Also, I only care about heat flux on the panel. It seems like I don't need to worry about eddies forming downstream from the panel. Does that sound right?

kellnerp · Apr 7, 2011

The i7 can be dual core or better. i7 Processors

I have a Hex Core cpu and when I watch task manager doing surfacing only four cores are used at any one time. PhotoWorks uses all cores. Many SW functions use one core. The task manager will show one window for each core. There is some question about how many processors various parts of Simulation will use.

TOP
CSWP, BSSE

"Node news is good news."

kymaks · Apr 7, 2011

I have an i7 M640. It's a dual core but task manager shows 4 cores. Flow Simulation only uses 2.

kellnerp · Apr 7, 2011

i7 640M
Windows should only show two processors.

Just switching to a Sandy Bridge would triple your speed (3.8/2.8*4/2) if you could go to desktop. For a big run this is nothing to sneeze at. 20 minute run down to 7 minutes. 24 hour run to overnight.

TOP
CSWP, BSSE