Effect of moisture enthalpy in douglas fir case

[ericvmueller]

Based on a user question, I was looking at recent changes to the results for the tree_2_m_49_pc.fds douglas fir case.

I noticed that after the updates to the solid phase enthalpy, and associated variable heat of reaction, there seemed to be a big change. Previously, we were specifying fixed heats of reaction for everything. Using the most recent source, if I add ADJUST_H = F just to the MOISTURE material to have a fixed heat of reaction for drying I get the difference shown here.

Neither are a particularly great representation of the experiment, but should we expect such a large difference?

[drjfloyd]

With ADJUST_H=F, enthalpy is not conserved. It doesn't surprise me that this might result in large differences. The guide states:

"This process of adjusting enthalpies can be skipped by setting ADJUST_H=F on a MATL line. This should be done when material reactions do not represent actual chemical reactions."

Maybe this should be reworded to something like: This should only be done when a material reaction does not represent a phyiscal pyrolysis or phase chage reaction. This is an input that should rarely be used outside of verifcation type cases.

[drjfloyd]

Based on the kinetics, the reaction peak is at 312 K. At 312 K the heat of vaporization of water using the NASA-9 polynomials is 2410.5 kJ/kg. I changed read.f90 in my branch to reflect that value and reran the case. Now it peaks at ~15 s at 0.35. I think some of the big change might just be non-linear effects. A little more or less spread early on can have a big effect on the ultimate growth rate and amount of spread.

[ericvmueller]

I wonder if it is better not to hard code a default MOISTURE reaction and just throw an error if MOISTURE FRACTION is used without MOISTURE being defined? Also, should the GET_TMP_REF routine include a temperature term for reactions with N_T!=0, such as with the default moisture reaction?

[drjfloyd]

IMO it would be better to not have a hard coded reaction. Then what is happening isn't hidden in the source for someone reviewing an input file.

Even if N_T is non-zero there is still a peak reaction rate (TGA for the MOISTURE material FDS creates is below). Though I realize looking at the source that I forgot the N_T term in GET_TMP_REF. I will get that fixed.

[drjfloyd]

Rerunning with the fix I just pushed up and changing H_o_R to 2295 to reflect the change in computed reaction peak. Looking the same as before which make sense since the H_o_R(T) functions would be very close between the two.

[ericvmueller]

Okay, besides maybe tweaking the hard-coded H_R to this value, I think we can close this (and consider the idea of no longer hard-coding moisture).

I do think it is something to keep in mind though. Given the change between the previous release and the upcoming release, it may not be the last time we get a question about this.

[rmcdermo]

I'm confused, this looks like a pretty significant step backwards.

[drjfloyd]

If the pior was two wrongs making a right, how is it a step backwards?

[drjfloyd]

Looking at the original reference:

The three 49 % trees were M=48 to 50 % based on random sampling of different tree locations. A 10 % error is given for the moisture values. It isn't discussed if, for example, the needles are noticeably drier than the largest pieces of wood or if the moisture is uniform over the tree. We are assuming uniform moisture content at 49 %.

The 49 % trees had a total heights of 2.0, 2.1, and 2.1 with crown widths of 1.7, 1.8, 1.8. Photos of the trees show that the are approximately conical but not a perfect cone and not perfectly uniform foliage density throughout the tree volume. We model these as a perfect 1.9 m tall by 1.7 m wide cone with perfectly uniform foliage and wood density.

The tree is modeled as cyldindrical particles that are divided into four monodisperse particle sizes basead on one full and three partial measurements of trees that were not the trees that were burned. We assume the average of those measurements apply to the specific trees that were burned. The paper does not provide full details on uncertainty for these measurements. Those measurements are binned into foliage, < 3 mm wood, 3 - 6 mm wood, and 6-10 mm wood. The trunk is ignored as it is assumed it sees negligible mass loss (an assumption not later commneted on using post test observations and it may have an impact at the end tail of fire). The particles used in our files wind up at diameters of 1.2 mm, 1,5 mm, 4.5 m, and 8 mm or the midpoints of the bin sizes. The estimated mass fractions of the various bins are then uniformly applied over the cone. This is not, however, the true geometric distibution. There is a range of sizes over all foliage bins in a real tree. For example, some needles will be thinner and some thicker. Foliage is going to be primarly in the outer layer of the cone. Very near the trunk a tree will have little foliage. The largest bin size will be biased towards the trunk as branches get thinner as one moves away from the trunk.

The ignition source is a ring of 400 particles. Looking at the paper there appears to be ~80-100 holes in the ring burner, and the burner looks to be a standard one inch copper tube. The particles are given a surface temperature of 1500 C with an emissivity of 1 (highly oxidized copper is ~0.8 to 0.9 and the photo in the paper shows fairly bright tubing). With ambient temperature propane gas flowing through an intialliy ambient temperature copper tube, the tube/gas temperature is not going to reach 1500 C over the 30 s ignition source time. Back of the napkin calc suggests the bulk copper pipe is getting to at most ~400 C (assuming a high emissivity which copper tubing does not have) before the ignition source is turned off. At 1500 C the radiation emission alone from the particles is equal to the output of the fire itself and physically it can't be more than the heat trasfer from the fire to the tube which is maybe a kW at most?

I ran five alternate cases: reducing burner to 400 K (cb), double grid resolution (hr), adding more particle size bins assuming gaussian distribution within the stated ranges (mp), and dropping (mo) and increasing (mo2) the moisture 10 %. Particle binning and grid resolution have little impact. Moisture content shifts the time of peak and the peak MLR by ~10 %. The cb case never takes off meaning our choice of an highly radiative burner is a large driver of the result.

I then ran four more cases increasing the burner temperature 200 K from the cb case (600,800,1000,1200) to demonstrate the effect of this parameter on the results. It is significant.

[ericvmueller]

Not surprised - the ignition has been known to be a big issue here and I think was tuned to get good results in previous cases (acknowledging that the representation of the actual burner at that scale is poor). But then we were probably compensating for other uncertainties. We can pick a new value like 800 (do you mean C or K?) that gives better results, but if you're saying it is still not the 'correct' value then maybe we need to state or demonstrate the sensitivity in the guide.

[drjfloyd]

argh mixed units up in post. Should have been 1500 C not K, I've edited the post. The true value is likely quite low during the time when the burner is running. 30 kW fire with ~10 kW radiative output means the copper tube is seeing maybe 1 to 2 kW of incident radiative flux. The e for relatively bright looking copper tube is ~0.3? At steady-state (which may not be reached during the ignition period) this is the copper tube emitting 0.3 to 0.6 kW or a particle surface temperature of 200 C to 300 C and the 400 C I ran didn't ignite. In reality it will start at ambient and rise in a near linear fashion to a peak when the burner is turned off.

I didn't try looking at sensitivity to particle distribution in space but looking at the paper the inner 1/2 to 2/3 of the branches down low have few needles

[rmcdermo]

I like the idea of demonstrating the sensitivities in the guide.

Does anyone have an idea about the qualitative change to more of a sharp peak in MLR compared to the more rounded curve we saw previously? This is something sensitive to geometry or fuel distribution, showing this sensitivity might be good.

[drjfloyd]

I haven't delved deep into why it changed but my suspicion is it comes from the char reaction. The changes for the adjust_h logic shouldn't have much impact on the virign material that only emits gas. The bulk of the mass loss occurs over a narrow temperature range where there shouldn't be a lot of impact going from H_R=constant to H_R=f(T). With only the fuel species produced (a single species) there were no changes to handling of enthalpy. For char, the adjust_h changes added a better accounting for the enthalpy for the gasses actually consumed and produced

[rmcdermo]

@ericvmueller When you get time, could you check if dialing back the char model is the issue with the shape of the MLR curve? Thanks

[ericvmueller]

I simulated with an 800C ignitor, based on Jason's plot. The alternate char model changes the peak and gives a slightly longer tail but the change in shape predates this. Or did you mean check before the fix related to the enthalpy for a lumped species char reaction?

[drjfloyd]

Looking at the log for the results in the out repository this case had results posted 06/08/22 and 01/05/23. The adjust_h was done in mid October of 22. Between the two there was a bug fix for density when moisture fraction is defined (https://github.com/firemodels/fds/commit/e28bc7a140a2ab812a3b5e4a56b9ef3bc21060f7). If pre-adjust h is a similar result to post, then this is another candidate.

[rmcdermo]

Thanks. No, I am not suggesting we dial back to pre-enthalpy fix. But I think it's important that we understand this qualitative change before we forget about it and it comes back to bite us a year from now. What is left to play with? Particle distribution? Position? Tree shape?

[rmcdermo]

We could do a revert on that commit just to check.

[drjfloyd]

Geometric distribution (foliage biased to the outside and large branches biased bottom and inward) is probably the biggest thing left untested. Size distribution didn't seem to have a large effect. Tree shape is a variable. It is sort of conical, but more a cone with a rounded bottom.

.

I think doing a local repo revert to check pre-adjust H shape and then the other commit if needed would be good to do.

[ruddymell]

The ignition particle temperature WFDS (the original study) was chosen (tuned) such that the gas phase temperature above the burner (without the tree present) averaged 800 C to approximate an average flame temperature. I believe the emissivity of the particles was zero. As you know, ignition is difficult to model. The drier trees ignited relatively easily. Our aim was to have a well-defined, axisymmetric, ignition procedure (actual and numerical), that was identical for BOTH tree-moisture cases.

The trees used were from a nursery and had needles much further into the interior than "natural" trees. That being said, the spatial heterogeneity of solid mass and void space is definitely not as uniform as is assumed. This oversimplification may be more consequential as the moisture rises and the vegetation ignites less easily.

There was a significant difference in the consumption of the tree between the "dry" and "wet" cases. The dry trees were essentially enveloped by the flame and complete consumption of the foliage and roundwood up to 10 mm diameter occurred (foliage was about 2/3 of the overall mass that was consumed based on the pre- and post-fire load cell measures of the total mass lost). Only a center portion of the wet trees was consumed (this was above the burner and, therefore, burner dependent). The volume containing consumed vegetation was not investigated to determine how much mass was lost in the different size classes.

On Wed, Mar 8, 2023 at 8:36 AM Jason Floyd @.***> wrote:

Looking at the original reference:

The three 49 % trees were M=48 to 50 % based on random sampling of different tree locations. A 10 % error is given for the moisture values. It isn't discussed if, for example, the needles are noticeably drier than the largest pieces of wood or if the moisture is uniform over the tree. We are assuming uniform moisture content at 49 %.

The 49 % trees had a total heights of 2.0, 2.1, and 2.1 with crown widths of 1.7, 1.8, 1.8. Photos of the trees show that the are approximately conical but not a perfect cone and not perfectly uniform foliage density throughout the tree volume. We model these as a perfect 1.9 m tall by 1.7 m wide cone with perfectly uniform foliage and wood density.

The tree is modeled as cyldindrical particles that are divided into four monodisperse particle sizes basead on one full and three partial measurements of trees that were not the trees that were burned. We assume the average of those measurements apply to the specific trees that were burned. The paper does not provide full details on uncertainty for these measurements. Those measurements are binned into foliage, < 3 mm wood, 3 - 6 mm wood, and 6-10 mm wood. The trunk is ignored as it is assumed it sees negligible mass loss (an assumption not later commneted on using post test observations and it may have an impact at the end tail of fire). The particles used in our files wind up at diameters of 1.2 mm, 1,5 mm, 4.5 m, and 8 mm or the midpoints of the bin sizes. The estimated mass fractions of the various bins are then uniformly applied over the cone. This is not, however, the true geometric distibution. There is a range of sizes over all foliage bins in a real tree. For example, some needles will be thinner and some thicker. Foliage is going to be primarly in the outer layer of the cone. Very near the trunk a tree will have little foliage. The largest bin size will be biased towards the trunk as branches get thinner as one moves away from the trunk.

The ignition source is a ring of 400 particles. Looking at the paper there appears to be ~80-100 holes in the ring burner, and the burner looks to be a standard one inch copper tube. The particles are given a surface temperature of 1500 K with an emissivity of 1 (highly oxidized copper is ~0.8 to 0.9 and the photo in the paper shows fairly bright tubing). With ambient temperature propane gas flowing through an intialliy ambient temperature copper tube, the tube/gas temperature is not going to reach 1500 K over the 30 s ignition source time. Back of the napkin calc suggests the bulk copper pipe is getting to at most ~400 K (assuming a high emissivity which copper tubing does not have) before the ignition source is turned off. At 1500 K the radiation emission alone from the particles is equal to the output of the fire itself and physically it can't be more than the heat trasfer from the fire to the tube which is maybe a kW at most?

I ran five alternate cases: reducing burner to 400 K (cb), double grid resolution (hr), adding more particle size bins assuming gaussian distribution within the stated ranges (mp), and dropping (mo) and increasing (mo2) the moisture 10 %. Particle binning and grid resolution have little impact. Moisture content shifts the time of peak and the peak MLR by ~10 %. The cb case never takes off meaning our choice of an highly radiative burner is a large driver of the result.

[image: image] https://user-images.githubusercontent.com/12799217/223773304-ec3e4a9c-8f4a-4af8-91a2-5c34a1002ec7.png

I then ran four more cases increasing the burner temperature 200 K from the cb case (600,800,1000,1200) to demonstrate the effect of this parameter on the results. It is significant.

[image: image] https://user-images.githubusercontent.com/12799217/223773364-e6be8283-69da-4586-9597-fd013a864b68.png

— Reply to this email directly, view it on GitHub https://github.com/firemodels/fds/issues/11507#issuecomment-1460469695, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADBB5TVAM6FWSOBLH2W2NG3W3CYPXANCNFSM6AAAAAAVRXHQ4Y . You are receiving this because you are subscribed to this thread.Message ID: @.***>

--

William (Ruddy) Mell U.S. Forest Service Pacific Wildland Fire Sciences Lab 400 N. 34th St., Suite 201 Seattle, WA 98103 @. @. 206-430-2072

[drjfloyd]

Emmissivity in the 2 m 49 pct input file is 1. Obviously if the ignition source as used in the test can't be resolved on the grid to give us ignition, then if we want to try and use this case for validation we need to do something to fix that; however, it is clear that how we chose to modify the ignition source can drive the result one way or the other. Definitely tricky and I don't know that there is a good solution for this.

&SURF ID='burner', TMP_FRONT=1500., EMISSIVITY=1., HRRPUA=60., RAMP_T='ignite', RAMP_Q='ignite', RADIUS=0.01, GEOMETRY='SPHERICAL' /

[ericvmueller]

I'm still looking into the relative effects of the ignition and pyrolysis model ...but I'm also wondering, @drjfloyd, with WATER_VAPOR being included in THERMO_DATA, does it make sense to automatically calculate a heat of reaction for the special material MOISTURE? Then, if the user doesn't specify anything, it will be sure to be consistent with NASA-9 and whatever the reference temperature is from the specified kinetics, rather than the current value coded in read.f90?

[mcgratta]

I added the special material 'MOISTURE' as a convenience and because water and moisture are not necessarily the same thing, so I'm told. Feel free to do whatever you like with it. I'm not wedded to it.

[drjfloyd]

I think it is a reasonable idea to have this ability, but it isn't a 30 sec fix. It will take a little time to do this:

If an input file has a MATL producing a SPEC_ID with liquid properties, FDS has no way of knowing if that MATL is in fact the same liquid as the SPEC_ID. It may not be; a user could be using that gas species as a convenience or other reason. If we wanted this behavior, I think we need some input on MATL to tell FDS explicitly that the solid properties of the MATL should be taken as the liquid properties associated with the SPEC_ID. This would require error checking that only one SPEC_ID be present for the MATL and that there is no residue otherwise that would mean the solid can't be the liquid. It would also require changes to MATL processing. MOISTURE right now is being defined in READ_MATL which is before PROC_SPEC_1 so we don't have arrays of species properties defined yet at the point where we are setting up the MATL rho,cp,k, and heat of reaction. We can't do the simple solution of moving PROC_SPEC_1 before PROC_MATL since we need READ_RAMP before PROC_SPEC_1.

[ericvmueller]

Okay - definitely seems more trouble than it's worth, particularly if it's only for this one particular case.

[r-broek]

Hi,

Based on the discussion above, I believe the original problem is resolved by explicitly defining the moisture evaporation reaction, as in the NIST_Douglas_Fir input files. Is this correct?

Furthermore, I would like to ask if there have been any new insights that I should be aware of. When I run the example case for the $MC = 0.49$ tree, it still results in an incorrect peak MLR of ~0.6 kg/s.

[ericvmueller]

Depends what you mean by the original problem. If you want to modify the heat of reaction for evaporation of moisture, this can be done by explicitly defining the reaction, as you say. The issue of over-predicting MLR at high moisture remains in the most recent validation runs. There are many unknowns and sensitivities, as discussed in the posts above, so this is still very much in the realm of research.