Synthetic eddies with HVAC fans

[Dave-McGill]

Morning All,

I've been playing with jet fans and synthetic eddies. Dan Swenson has posted examples on the Thunderhead site, and there is an example in the V & V files. In both of these situations a simple fan was used. I've been unable to demonstrate synthetic eddies when using an HVAC fan. (The ultimate goal is to generate turbulence in jet fans in parking garages with cells on the order of 0.125 m.)

Is this a bug, or an undocumented feature?

See the attached comparison. It comprises a,

• simple fan without eddies
• simple fan with eddies
• HVAC fan without eddies
• HVAC fan with eddies.

They are in separate meshes, so there is no interaction, and all surfaces are open. At best, I can imagine a slight increase in the turbulence of the HVAC fan with eddies when compared to the HVAC fan without, but that may just be wishful thinking.

Thanks

Dave

jet_fan_comparison_0998.zip

[mcgratta]

I'll take a look at it.

[Dave-McGill]

Thanks Kevin

[rmcdermo]

I would not be surprised if SEM does not work with HVAC vent. I can document that. Assign to me.

On Wed, Mar 2, 2016 at 10:05 AM, Dave-McGill notifications@github.com wrote:

Thanks Kevin

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-191277712 .

[rmcdermo]

Added error trap and user guide discussion in pull request #3596.

[Dave-McGill]

There are a lot of people modelling underground car parks with jet fans.

Is it possible to make synthetic eddies work with HVAC fans?

[rmcdermo]

This would take some work. I'll have a look today and see what we can do.

But please answer this question. Do we not agree that with sufficient resolution SEM is not necessary? And with low resolution (1 or 2 cells across a fan) the SEM parameters are just knobs to turn; they have no physical significance. Hence you need data for calibration. Where does that come from?

I am all for making things as easy as possible. But put yourself in my shoes. Our critics love to say that we enable poor CFD practice. How would this be good CFD practice?

On Thu, Mar 3, 2016 at 5:37 AM, Dave-McGill notifications@github.com wrote:

There are a lot of people modelling underground car parks with jet fans.

Is it possible to make synthetic eddies work with HVAC fans?

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-191699823 .

[Dave-McGill]

Randy,

Yes, with sufficient resolution SEM is not necessary.

Dan has built a small car park example that he used to demonstrate Dynamic Smagorisky. The overall cell size is 0.25 m, and the cell size in the region of the 4 fans is 0.125m, for a total of ~ 1/2 million cells. In his earlier tests using Deardorff, he found a cell size of ~ 0.01 m was necessary to resolve the jet fan turbulence. If I drop the cell size around the fans to ~ 0.0156 the overall cell count is ~68 million cells, which is not doable.

I'm not qualified to judge what is good or bad CFD practice. I'm trying to find a solution that is practicable for the average FPE, who has limited computing power. Car parks with jet fans are being run all the time, and that's not going to change; the the challenge is to find the best, or if you prefer, least bad, way of doing it.

Dave

[rmcdermo]

Dave,

I will go on record to say I think what Dan is doing is bad CFD practice. There is no reason he could not make Pryrosim relatively easily resolve the fan regions but leave the rest of the car park at .25 cm resolution. I also take exception to use of the phrase, "resolve the jet fan turbulence". It is more like he "witnesses fluctuations".

Are you going to use DSMAG for the fire? I have said repeatedly that DSMAG is not a good model for the fire, especially at coarse resolutions.

Randy

On Thu, Mar 3, 2016 at 8:34 AM, Dave-McGill notifications@github.com wrote:

Randy,

Yes, with sufficient resolution SEM is not necessary.

Dan has built a small car park example that he used to demonstrate Dynamic Smagorisky. The overall cell size is 0.25 m, and the cell size in the region of the 4 fans is 0.125m, for a total of ~ 1/2 million cells. In his earlier tests using Deardorff, he found a cell size of ~ 0.01 m was necessary to resolve the jet fan turbulence. If I drop the cell size around the fans to ~ 0.0156 the overall cell count is ~68 million cells, which is not doable.

I'm not qualified to judge what is good or bad CFD practice. I'm trying to find a solution that is practicable for the average FPE, who has limited computing power. Car parks with jet fans are being run all the time, and that's not going to change; the the challenge is to find the best, or if you prefer, least bad, way of doing it.

Dave

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-191762906 .

[Dave-McGill]

Randy,

I can't speak for Dan, and I don't want to put words in his mouth.

I'm not using DSMAG. In the tests using Deardorff do ~1 cm cells with not qualify as resolving the turbulence?

What about using a screen to create turbulence? I've never used screens before. I've cut and pasted from one the example files and and generating turbulence at 0.125 m resolution. (see attached)

Dave

[rmcdermo]

The question is why do you think one solution is better than any other? The only comparison with data or theory that I've seen is in our verification guide. It seems to me that this is the best guidance.

If jet fans in car parks are a critical need in FDS, we should undertake a serious effort to do this right. I get very frustrated by all the knob turning. Inevitably, I am the one who has to answer for it.

On Thu, Mar 3, 2016 at 9:47 AM, Dave-McGill notifications@github.com wrote:

Randy,

I can't speak for Dan, and I don't want to put words in his mouth.

I'm not using DSMAG. In the tests using Deardorff do ~1 cm cells with not qualify as resolving the turbulence?

What about using a screen to create turbulence? I've never used screens before. I've cut and pasted from one the example files and and generating turbulence at 0.125 m resolution. (see attached)

Dave

[image: with and without scree] https://cloud.githubusercontent.com/assets/13259079/13497658/95a9f3d0-e124-11e5-9ad6-2764e691d7bc.png

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-191796252 .

[rmcdermo]

By the way, any "serious effort" would start with actual data from a car park or tunnel. Can you point me to anything along those lines?

On Thu, Mar 3, 2016 at 10:29 AM, Randy McDermott randy.mcdermott@gmail.com wrote:

The question is why do you think one solution is better than any other? The only comparison with data or theory that I've seen is in our verification guide. It seems to me that this is the best guidance.

If jet fans in car parks are a critical need in FDS, we should undertake a serious effort to do this right. I get very frustrated by all the knob turning. Inevitably, I am the one who has to answer for it.

On Thu, Mar 3, 2016 at 9:47 AM, Dave-McGill notifications@github.com wrote:

Randy,

I can't speak for Dan, and I don't want to put words in his mouth.

I'm not using DSMAG. In the tests using Deardorff do ~1 cm cells with not qualify as resolving the turbulence?

What about using a screen to create turbulence? I've never used screens before. I've cut and pasted from one the example files and and generating turbulence at 0.125 m resolution. (see attached)

Dave

[image: with and without scree] https://cloud.githubusercontent.com/assets/13259079/13497658/95a9f3d0-e124-11e5-9ad6-2764e691d7bc.png

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-191796252 .

[Dave-McGill]

Nope

[rmcdermo]

Dave, Have you tried a "velocity patch"? See if the attached works for you. At this resolution, the core does not quite break down fast enough to give exponential decay of the centerline velocity component. But I suspect that one more level of refinement would do it.

jet_fan_patch.txt

[rmcdermo]

Here is some further analysis, which follows the centerline decay equation in the verification guide. While the decay is not a pretty exponential, it is within the uncertainty bounds on the experiments that we have available. So, even though it would be possible to dial in the patch decay more accurately, I see no reason to go beyond this level of complexity. Attached are the input files, scripts, and results for this simple case.

jet_fan_patch.fds.txt

plot_vel.py.txt

[Dave-McGill]

Randy,

I cut the cell size in 1/2 for jet_fan_patch.fds, and am running it now.

I'd never heard of a velocity patch until an hour ago, and I still have a pretty vague idea of what it is. Are suggesting that a patch with a cell size of 0.125 m is a practical solution?

Dave

[rmcdermo]

Yes. It seems to do the trick.

On Thu, Mar 3, 2016 at 1:36 PM, Dave-McGill notifications@github.com wrote:

Randy,

I cut the cell size in 1/2 for jet_fan_patch.fds, and am running it now.

I'd never heard of a velocity patch until an hour ago, and I still have a pretty vague idea of what it is. Are suggesting that a patch with a cell size of 0.125 m is a practical solution?

Dave

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-191904111 .

[Dave-McGill]

It appears that the most important parameter is the face velocity at the fan. Would it make sense to add it to the section in the manual on jet fans? The explanation would have to be simplified considerably.

Do you want me to make an initial attempt?

[rmcdermo]

I guess I don't understand what you are saying. Isn't velocity the usual parameter for a fan? What other parameters do you use?

If you want a louver, you would add a second small patch just at the exit and apply the velocity components there. I think it is best if we first put together a series of examples.

On Thu, Mar 3, 2016 at 1:46 PM, Dave-McGill notifications@github.com wrote:

It appears that the most important parameter is the face velocity at the fan. Would it make sense to add it to the section in the manual on jet fans? The explanation would have to be simplified considerably.

Do you want me to make an initial attempt?

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-191909671 .

[Dave-McGill]

The point I was trying to make was that it would be pretty much a cut 'n paste sort of operation for a user. All he/she would have to do is put in the face velocity of the specific fan that was being modelled, and get it lined up on the right axis, and at the right height.

[rmcdermo]

Yes, that's true. With this approach you can also have non-grid aligned fans (if you don't care about the aesthetics of the shroud).

On Thu, Mar 3, 2016 at 1:58 PM, Dave-McGill notifications@github.com wrote:

The point I was trying to make was that it would be pretty much a cut 'n paste sort of operation for a user. All he/she would have to do is put in the face velocity of the specific fan that was being modelled, and get it lined up on the right axis, and at the right height.

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-191913489 .

[Dave-McGill]

Morning Randy,

Using velocity patches with cell sizes of 0.125 and 0.0625, here is a comparison of the centerline velocities.

[rmcdermo]

Please email me the _line file for .0625 and I'll plot with data.

On Fri, Mar 4, 2016 at 6:56 AM, Dave-McGill notifications@github.com wrote:

Morning Randy,

Using velocity patches with cell sizes of 0.125 and 0.0625, here is a comparison of the centerline velocities. [image: velocity patch centerine velocities at different cell sizes] https://cloud.githubusercontent.com/assets/13259079/13526486/2a4a8778-e1d6-11e5-88ac-591807515dd2.PNG

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-192253953 .

[Dave-McGill]

You've got mail

[Dave-McGill]

In Dan's original sensitivity study he compared the simulations to experimental work by Geisen (which also cites earlier work by van Oerle). I've plotted their experimental data witht the two velocity patches. It's convenient, as they both used fans with a face velocity of 18 m/s. Their measurements extended out to 16 m, so I'm going to re-run the velocity patches to that distance.

I've also attached Geisen's paper. See figure 8, it has a collection of experimental observations, including wall jets.

11_envsoft_giesen.pdf

[rmcdermo]

Thanks for the paper. Let's set up a series of verification cases and non-dimensionalize in the same way we do the case in the verification guide now.

On Fri, Mar 4, 2016 at 8:25 AM, Dave-McGill notifications@github.com wrote:

In Dan's original sensitivity study he compared the simulations to experimental work by Geisen (which also cites earlier work by van Oerle). I've plotted their experimental data witht the two velocity patches. It's convenient, as they both used fans with a face velocity of 18 m/s. Their measurements extended out to 16 m, so I'm going to re-run the velocity patches to that distance.

I've also attached Geisen's paper. See figure 8, it has a collection of experimental observations, including wall jets. [image: velocity patch centerine velocities at different cell sizes and experimental data] https://cloud.githubusercontent.com/assets/13259079/13528119/14eb3786-e1e2-11e5-8ebe-420d61cfb94e.PNG

11_envsoft_giesen.pdf https://github.com/firemodels/fds-smv/files/158700/11_envsoft_giesen.pdf

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-192282707 .

[Dave-McGill]

Here's the comparison with measurements from the velocity patch (0.125 m cells) extended out to 16 m. The simulation with 0.0625 m cells is going to take all day to complete.

[rmcdermo]

Dave, This data is not behaving like a typical turbulent free jet. Probably boundary effects in the garage. Also, the plots I gave above have errors because my "origin" was not correctly specified. I am fixing this.

On Fri, Mar 4, 2016 at 11:07 AM, Dave-McGill notifications@github.com wrote:

Here's the comparison with measurements from the velocity patch (0.125 m cells) extended out to 16 m. The simulation with 0.0625 m cells is going to take all day to complete. [image: velocity patch centerine velocities at different cell sizes and experimental data] https://cloud.githubusercontent.com/assets/13259079/13532372/3448b4d4-e1f9-11e5-9206-2f929a7cf937.PNG

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-192336141 .

[rmcdermo]

I specified the velocity in the core region as part of the patch and corrected my exp data plot. I get the following. So, my conclusion is that this is a viable (in my opinion preferable) approach to modeling the fans (as opposed to using DSMAG). I do not think you can compare the data from the papers you sent with the FDS configuration we are testing. These papers should be treated like validation cases and a careful study of the full geometry should be done. I would appreciate it if someone else would take this on. If I put it in my queue, it will likely not see the light of day.

[rmcdermo]

Here is the low resolution input file.

jet_fan_patch.fds.txt

[Dave-McGill]

Randy,

Fig. 3 of Geisen's paper shows the test set-up for the free jet case; the walls aren't very close. Besides, Figure 8 shows the same behaviour for both the free jet and wall tests. This reminds me of the quote, “In theory, theory and practice are the same. In practice, they are not.”.

We've got 5 examples of jet fans exhibiting the same behaviour, which is a good thing in my mind.

Here's the graph updated with the 0.0625 cells and measurements taken out to 16 m. I'll mail you the _line.csv file.

Dave

[Dave-McGill]

OK, I'll do the validation study, although it might be the summer before I can get to it. I'll need some guidance as to how you want it done.

[dvswenson]

Randy,

The velocity patch looks like a fruitful approach to jet fans.

Could you comment on why using a velocity patch with a uniform velocity specified over a shrouded volume results in different downstream behavior than a vent with a uniform specified velocity? What was your thinking that lead to suggesting this approach?

Thanks, Dan

[dvswenson]

"led" not "lead"

[rmcdermo]

The only reason DSMAG appears to work better for some of your cases is that it lets the jet core break down sooner. With sufficient resolution one could apply Jarrin's SEM method to achieve the same result (see FDS Verification Guide). But apparently everyone insists on low resolution (2 cells across the fan), so SEM is out (you need grid cells to generate coherent eddies). The reason the patch works is that it sucks in random fluctuations from the inlet. Since the immersed boundary method used in the patch is not perfect and since we are not forcing the V and W components (in this case) then it seems that we persist enough fluctuations in the jet to allow the core to break down with 2 cells with Deardorff. I am much more comfortable with this because, as I've said before, DSMAG is a really bad idea for coarse LES calculations of fire.

On Mon, Mar 7, 2016 at 3:40 PM, dvswenson notifications@github.com wrote:

"led" not "lead"

— Reply to this email directly or view it on GitHub https://github.com/firemodels/fds-smv/issues/3592#issuecomment-193440394 .

[dvswenson]

Thanks.

[Monika-Samy]

how to model jet fan by velocity patch