finally got conjugate heat transfer (CHT) running properly. w00t!



You can see the thermal proogation from one source to the other and the effects of it.
It's just a really simple block and a pipe and fluid flowing through the pipe. (50 C water, 10 C water in block, 20 C pipe.)

There is one tiny perculiarity with it where when I was analyizing the pipe and the water only; there is one spot that was hotter than everywhere else (not by much though) and it's sitting right next to the coldest spot too.

Statistics about that run:
3 domains
17,247 elements
1 m 32 s (total run time)

next up: Northbridge heatsink.



Here's what the mesh looks like so far: (~2.1 M elements)



At 2 GB of RAM, with 1.1 GB free, it's telling me that I don't have enough memory.

Attached Thumbnails

     

Nice cant wait to see what else you do.

Quote: Originally Posted by Cpt.Planet Nice cant wait to see what else you do.

Yes, part of the reason for doing/learning the CHT analysis was so that I can figure out how cold you should be able to get the bottom of your container to be.

I actually don't know much in the way of properties for solid CO2 (i.e. dry ice). What temperature should I be putting it in? I am thinking of putting it slightly about the freezing point because that way it would be all vapor and it would be easier for me to model on it (and that will also increase the probability of it running successfully.)

I'm still working on doing the northbridge heatsink CHT analysis.

Apparently, it's yelling at me because it can't address/request enough memory (allocate) and that I need to increase the memory stack size for "real" (i.e. as in the number type), and the online documentation doesn't tell me how to do that.

So, I have to try and find out tomorrow how to do that; and hopefully, then it'd be able to run properly. (My guess is because of the number of surfaces that are involved in the heat transfer, so I am going to try using fewer heat transfer surfaces (just more towards the center) and closer to the point source.

Interesting though. If you have an aluminum heatsink, with a 25 W point (heat) source; it can take it to almost 100 C. Another thing that I can try is to specify that as a wall boundary and apply the heat source over the entire area. (It was kind of unusual for me to see W(atts) represented as kg.m^2/s^3.)

*edit*

Another interesting thing that I just tried. Since I can't get the heat transfer to go, I do have it set up so that it can take air blowing across it. At 0.8 m/s, minimum temperature was 353 K (80 C). At 1.5 m/s, that drops down to 349 K (76 C).

*edit*

With 8 segments of the "fins", I can get the CHT to go. I have about 100 MB free RAM (which is unusually high, so probably means that I can add more segments to it) and a 1.89 GB swap file at the start of the run.)

Won't be able to fine out much more about it, but it's taking a LONG time to solve it. Check in tomorrow morning on the results.

Latest update:

I think that I fixed the memory allocation issue so it's running now.

I also ended up simplifying the geometry and also tried to drop the overall number of elements so now it's running at 852k elements.

Down side (thus far, after 17 iterations) is that it doesn't appear to the converging to a solution, which is bad.

I am certain that most of you are familiar with Cpt. Planet's Project Opteron -ve 146. His dry ice container is being scheduled in for analysis because I think that I would be able to solve it and get some pretty decent results out of it now.

We'll have to see how well the current heatsink runs go, as I have pretty much the next one queued up already.

*edit*
Update: Mon Feb 6 03:16 EDT 2006

Finally finished the simplified NB analysis. Spent the better part of today (now technically yesterday) trying to finish that up and also getting the actual (unsimplified) heatsink analysis to go.

Looks like that even with the memory allocation overrides, there's still too much for the system. Quite unfortunate, because I would have really liked to see how well it would run with the 2.1 M elements and thermal transfer. Oh well, I suppose.

Another interesting thing about the program is that for heat transfer, there are a couple of thermal models and methods that are available. Of which, one is a point source, another is to apply a wall BC to the model, and there's a source "tab" in the BC definition, and then there's heat flux.

As a source, it is given in units of [kg.m^2.s^-3], which actually does translate to W, however, that was kind of rather unusual notation. In any cae, setting that as 25, I end up at like 400 C on the surface of the heatsink. Given the what I(/we) know physically about the parts and the operating conditions, that definitely doesn't make any sense. Not reasonably, and not logically. So, back to the drawing board.

As a heat flux, that is given in units of [W.m^-2]. So, while the units are actually a lot closer to what I expected them to be, if I take the geometry of the part into consideration (i.e. 25 W divided by the area over which it is dissipated, (which works out to something like 0.0013 m^2), I get a heat flux of around 18,000 ([W.m^-2]). While that process/idea looks to be mathematically correct, the results are similar (if not identical to setting it as a source term rather than a heat flux.)

So, obviously, that isn't right either. So, neglecting part geometry, and setting that as 25; I get the result (4 C over ambient). Doing the "logical/gut" check; that certainly makes a whole lot of sense (at least from the result side). What I still don't quite get is why there's such a big disparity between what I know to be "mathematically correct, in taking the geometry of the part into consideration), vs. taking it as is.
Are all thermal dissipation values (that you find online, or however, whatever, wherever) really in units of [W.m^-2]?

Another interesting thing is that heat flux, as far as I know and understand it to be solved with a Gaussian closed-surface integral. (That is something that is used in electronics/electrical with charge density). I actually did enlist some help from my roommates as well. One to help verify if the numbers (i.e. the 25 W and the 4 C over ambient) made sense based on his experience. Another, we had a discussion about the difference between heat source and heat flux and the difference in the result that I am getting a) as a result of that and b) with and without consideration for geometry. And, one more to verify (without derivation) whether a Gaussian closed-surface integral can be used for heat flux instead of charge density, or flux.

What a way to spend a Sunday eh?

I supposed the upside to all of this is that I can probably do CHT analysis upto about 1 M elements.

Another thing that's running right now is to see what happens if I were to half the size (in the Z-direction) of the heatsink.

*edit*
Update: Mon Feb 6 04:59 EDT 2006

Completed the half-height/size analysis. Negligable difference.

Interesting.

Now running without turbulence. Figured that the flow velocity should be low enough that I can take that out; which should speed up the run. (est. 17% faster.) Also figure that the Reynolds number should be fairly low (ratio of shear stress to viscosity.)

Update: Mon Feb 6 06:45 EDT 2006

Neglecting turbulence yielded some fairly interesting results on the half-sized run.

The RMS residuals had a much higher standard deviation, which I didn't expect. As a result, that also changed the solution quite dramatically.

It reported that the max temperature was 2 C over ambient, NOT 4 C. And that makes me worry because it is underpredicting the worse case scenario.




Conclusion from that: Run with turbulence enabled. Yes, it takes longer, but at least that I don't have to worry about the accuracy of the results.

Attached Thumbnails

 

Just finished Cpt. Planet's Dry Ice Container CHT analysis.

Pretty interesting. There's going to be a slight change/update after my classes because I forgot to set the initial conditions of the container to be ambient temperature.

Cross reference to his thread.

Um...other interesting stuff about it. 70k elements. 800 MB RAM usage. Runs in about 9 minutes.

Trying to get him to verify some of the results.

Differences might exist because he had modified the container after it was made, and my CAD model wasn't updated to reflect those changes. So, we'll have to do that retroactively I suppose.

Quote: Originally Posted by alpha754293 Just finished Cpt. Planet's Dry Ice Container CHT analysis. Pretty interesting. There's going to be a slight change/update after my classes because I forgot to set the initial conditions of the container to be ambient temperature. Cross reference to his thread. Um...other interesting stuff about it. 70k elements. 800 MB RAM usage. Runs in about 9 minutes. Trying to get him to verify some of the results. Differences might exist because he had modified the container after it was made, and my CAD model wasn't updated to reflect those changes. So, we'll have to do that retroactively I suppose.

It still hasnt been modified yet. Also you had the cpu wattage higher than the stock wattage is. Stock is 67W. However I dont think that the wattage was really increased until I really pushed the CPU. I wouldnt be suprised your results may be fairly accurate.

Quote: Originally Posted by Cpt.Planet It still hasnt been modified yet. Also you had the cpu wattage higher than the stock wattage is. Stock is 67W. However I dont think that the wattage was really increased until I really pushed the CPU. I wouldnt be suprised your results may be fairly accurate.

I thought that you changed the thickness on the bottom of the container (the surface that sits on top of the CPU?)

Quote: Originally Posted by alpha754293 I thought that you changed the thickness on the bottom of the container (the surface that sits on top of the CPU?)

Not yet. I need to to get better performance. I want to take 4mm or so off.

Quote: Originally Posted by Cpt.Planet Not yet. I need to to get better performance. I want to take 4mm or so off.

Ahh. So, all of the tests that you've done so far is as it without any modifications.

In that case, I am going to need to spend some time making sure that the analysis model is true and accurate before modifying the CAD model (-4 mm).

Will talk with you later on today/tonight to get the rest of the details so that I can set up my analysis and make sure that it is set up properly.

*edit*
Update: Thu Feb 09 10:14 EDT 2006

Been busy with school lately. Did the analysis for Cpt. Planet with the 4 mm reduction. Ended up being about 4 C warmer than without the reduction. Haven't tried making it thicker actually to see what would happen to it. Unfortuately, I don't have access to my CAD files now (at school; didn't load them up to the server. Didn't think that I would need them.) So, will have to do give that a shot just because to see what the results would be if I modified it the other way.

Will post the results with the 4 mm reduction in a while.

*edit*

Quote: Originally Posted by Cpt. Planet Not yet. I need to to get better performance. I want to take 4mm or so off.

Now the optimizations begin.