Grid Size Justification

[Kenny Leong]

Hi all,

Previously, an approach recommended by Bounagui et al. (Bounagui, A; Kashef, A; Benichou, N, 2003) was utilized in quantitatively determining the suitability of the grid resolution adopted. An important parameter that was referred to in determining the suitability of a grid resolution is the characteristic fire diameter, D*. The grid size ratio is then calculated to look at number of cells in the region. Higher ratios signified better numerical model predictions. FDS program (NIST) previously suggests the grid size ratio should be greater than 4.

Is the above still applicable or is there other methods of determining the suitability of the selected grid size?

Thanks

Kenny

[o...@aquacoustics.biz]

While D*, Q* or rules of thumb can provide a ready estimate of required resolution, there may be other factors such as the geometry of the model, its purpose, and the available time that can may be determining factors.

Computational mesh resolution is actually all about simulation time, otherwise we’d all run models at arbitrarily fine resolutions.  A model that takes months to run is useless when the project time line is six months!

In my experience model development is done progressively, adding features at a relative course resolution sufficient for reasonable representation of the problem geometry in the first instance.

Once the model is developed and debugged for the duration of the simulation a simple grid sensitivity analysis can be used to validate the grid resolution for the output of interest.  If the mesh resolution is reduced progressively by a volumetric factor of 8 (halving the side dimensions of a cubic cell) does the output of interest change appreciably?

Remember that this is a modeling exercise and the output is at best a reasonable approximation based on simplified physics, chemistry and empirical relationships at a macro level.

For best computational efficiency a courser resolution is desirable.  A useful trick can be to use a higher resolution mesh to contain the fire and plume (with relatively high computational resources assigned to this mesh) and reduced resolution meshes in other spaces.

While models incorporating an entire building can look impressive, a question you might ask is, ‘is this really necessary to for the task at hand?’  In a high rise building you might limit a simulation to the floor of fire origin, safe paths, shafts, and perhaps the top floor, incorporating representative leakage to other spaces.

If a fire reaches a quasi-steady state (perhaps through sprinkler control, ventilation or fuel limits) maybe you don’t need to run the model for 1,200 seconds?

In my experience folk often develop unnecessarily complicated models at arbitrarily fine resolutions resulting in excessive computational time with no benefit in the ‘accuracy’ of the results.  With appropriate refinement, sensitivity analysis, and computational resources some model run times can be reduced from weeks to hours.

[dr_jfloyd]

Section 6.3.6 of the User's Guide discusses grid resolution and it says much the same as the first response. I would add just a couple of comments in addition to the first response. The D*/dx >= 4 rule of thumb is most applicable for simulations where the primary interest is the smoke plume (for example smoke control). I disagree somewhat that mesh resolution is all about simulation time. It is true that we never have the luxury of unlimited time and unlimited computing power to do highly resolved calculations. We all look to select grid resolutions to meet project deadlines, and the tips and tricks in the first response can help us afford more resolution where needed. However, in the end we also need to ensure that the grid resolution we can afford given time and computation resources results in a solution that is good enough.    

[Kevin McGrattan]

The "rule" of D*/dx>4 is not a rule at all. This stems from a report of a validation study that we did back in 2007 in which we listed the D*/dx values for the different simulations. As I recall, the case where D*/dx=4 was one in which there was a small fire in a room of a 3 room enclosure (NBS Multi-Room). We gridded the entire space at 10 cm, as a matter of convenience. We did not do a grid resolution study. We just presented the results as is. We now routinely do simulations at different grid resolutions for the FDS Validation Guide. Not all cases, because it takes alot of CPU time to run all these cases, but some. We include the results of both coarse and crude calculations in the statistical scatter plots. We do this so that the reader understands the effects of the grid, but it is not meant as a recommendation. The grid you decide to use is a decision that you alone must make, depending on the objective of your study. 

[Khalid Moinuddin]

One can also look at the section “Measure of Turbulence Resolution” in the User guide

--
You received this message because you are subscribed to the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email to fds-smv+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/f3cba7e7-f169-477e-9d15-3691ec0e9ab7%40googlegroups.com.

[Hannah 11]

Dear dr_jfloyd,

        My question may be trivial. I would like to know how D* is arrived at (user guide,6.3.6). Where can I find it?. Thank you

[Tim O'Brien]

D* is the non-dimensional characteristic diameter of the fire derived from the Froude number.  Refer to Section 6.3.6 of the FDS Users Guide.  It is related to (and can be derived from) Q*.  See Zukoski 1975 although also attributed to others.

 

From: fds...@googlegroups.com [mailto:fds...@googlegroups.com] On Behalf Of Hannah 11
Sent: Thursday, 17 June 2021 7:53 PM
To: FDS and Smokeview Discussions
Subject: [fds-smv] Re: Grid Size Justification

 

Dear dr_jfloyd,

--
You received this message because you are subscribed to a topic in the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/fds-smv/e66yfvAn6CU/unsubscribe.
To unsubscribe from this group and all its topics, send an email to fds-smv+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/be7cfbb7-a023-4440-92de-37fc9daa3174n%40googlegroups.com.

[Hannah 11]

Thank you so much. I did have a look at the derivation of Q*.

 Are we taking the fire Froude number, Q* to be 1 to arrive at the equation for characteristic fire diameter D* (user guide 6.3.6)?. If so why?. TIA

[Tim O'Brien]

I’m not sure what you are asking Hannah.  Perhaps you could take a step backwards and tell me what you are trying to do with FDS?  I’m assuming you are trying to establish an appropriate mesh resolution.  But it might be that you are trying to understand why D*/dx within an upper bound may provide a reasonable initial estimate for mesh resolution given a HRR.

 

Note that the Froude number is not specific to fire.  It is simply a ratio (put simply: velocity to length).  Natural (buoyant diffusion) fires have been shown to exhibit Froude numbers within a range.  Plume velocity is driven by heat.  So we might expect that the ratio of HRR to cell dimension should fall within bounds to represent real fires.

 

If you are trying to establish an appropriate mesh resolution then you should complete a sensitivity study for the parameters of interest.       

To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/7f2b9330-00b0-4d95-9535-d1237fa75694n%40googlegroups.com.

[Hannah 11]

Thank you so much. I am not doing any simulations as of now. But I was just interested in the relation between Q* and D*.

[Kevin McGrattan]

Q* = (D*/D)^(5/2)