Surface Area Convergence Data
About
The surface area calculations have come under scrutiny when examining the burn rate, and the lack of convergance therein.
The point of this page, and the data presented within is to first determine whether the surface area calculations do converge over a set of reasonable resolutions.
If not, it will hopefully provide insight into a solution.
The current setup is to test several different geometries including: cylinders, holes, cubic shapes, triangular shapes. Results can be seen below.
04/08/09 - Finished wedge runs from 45-85 degrees in 5 degree incriments. Brings up fact that I need to output surface area at each cell, thus I'll be implementing that code on Friday (or so). Most of the 0.725 runs are complete and 0.675 runs are about halfway there. Should have the data visually availale online by late Friday as well.
04/06/09 - Considering unpacked geometry as the major focus for correction at this point, 0.65 is no longer a candidate for a new threshold value for burning cells. While it creates quicker convergence of logarithmic trends in packed geometry, the overall error is larger. Furthermore, it appears that it creates instability in the number of burning cells, as can be seen in several cases below, including: both cases with 16 unpacked cylinders, and both cases with unpacked holes. Currently, values of 0.675 and 0.725 are being tested to see if results are better. Prediction: 0.675 will have a smaller slope than 0.7 for error trend, also that total error might be focused at a zero, closer to 1mm spacial resolution.
As of now, the problem must lie within computation of density gradient. Thus, I'm pursuing more wedges of different angles, so that I can come up with some correction coefficient.
04/03/09 - Values of threshold value have been set to 0.65 and 0.75 and results have been added to this page. Mixed results occured, however, it still appears 0.7 is the best value. Most results, however indicated a value slightly lower (but nowhere near as low as 0.65) might be the best. However, as the graphs go up on this page, it becomes apparent that the threshold value is not the root of the problem. Where 0.65 increases rate of convergence for packed cylinders and holes (which is desirable), the value isn't appropriate for non packed cases. Furthermore, 0.75 appears to be better for some non packed cases, however, doesn't quite hold up for packed. Next step, to look at the wedge case, and really understand why the surface area calculations give error.
03/29/09 - Currently, the hypothesis being tested is that the threshhold value max mass/min mass for determining burning cell state is wrong. Different values are being tested to determine whether they provide better data. Overall, this idea arose from the exponential trend seen in the number of burning cells for circular geometries. The number of burning cells for double resolution should be a factor of 2 to 4 times that of the original value, most likely closer to 2. Extra cells would increase the amount of surface area, even to the point of overestimating the area, as can be seen in the unpacked cylinders.
Cases
NOTE: keyword "Resolution" denotes spacial resolution in both x and y directions (which are the same for all cases)
16 Cylinders Packed
- The setup for this simulation consists of:
- A domain 192mm by 192mm (double this for 64mm diamter case)
- 16 PBX cylinders w/ diameter 32mm and 64mm
- Resolutions(grid spacing dx dy): 0.125mm, 0.25mm, 0.5mm, 0.67mm, 1.0mm, 1.25mm, 2.0mm
- Atmosphere in between cylinders at 600K (in order to trigger "burning cell" flag
- Runs including tight packed and loose packed (45 deg array and rectangular array respectively
- Burn Rate set = 0 in code to isolate surface area calculations
Tight Packed
32mm diamter
- Plots of Surface Area Error:
- Logarithmic trend is apparent for the 3 different threshold values:
- 0.65 Threshold: |Ε|=0.000912*Resolution0.2855
- 0.70 Threshold: |Ε|=0.00061881*Resolution0.51787
- 0.75 Threshold: |Ε|=0.0005675*Resolution0.62372
- Input files can be found here.
64mm diamter
- Plots of Surface Area Error:
- Logarithmic trend is apparent for the 3 different threshold values:
- 0.65 Threshold: |Ε|=0.0026249*Resolution0.16273
- 0.70 Threshold: |Ε|=0.0021464*Resolution0.20996
- 0.75 Threshold: |Ε|=0.0020507*Resolution0.22823
- Input files can be found here.
- *NOTE* 0.125 mm case was not run do to memory requirements and CPU time required to run the case,
16 Cylinders Not Packed
- The setup for this simulation consists of:
- A domain 100mm by 100mm (200mm by 200mm in 32mm diamter case)
- 2 cases: 16 PBX cylinders w/ diameter 16mm and 32mm
- Resolutions(grid spacing dx dy): 0.125mm, 0.25mm, 0.5mm, 0.67mm, 1.0mm, 1.25mm, 2.0mm
- Atmosphere in between cylinders at 600K (in order to trigger "burning cell" flag
- Burn Rate set = 0 in code to isolate surface area calculations
16 mm Diameter
- Plots of Surface Area Error:
- Linear trend is apparent for the 3 different threshold values:
- 0.65 Threshold: |Ε|=-6.8289e-6*Resolutio+0.00016817
- 0.70 Threshold: |Ε|=8.5158e-5*Resolution-6.5328e-5
- 0.75 Threshold: |Ε|=9.2173e-5*Resolution-7.4939e-5
- Input files can be found here.
32 mm Diameter
- Plots of Surface Area Error:
- Logarithmic trend is apparent for the 3 different threshold values:
- 0.65 Threshold: |Ε|=9.0387e-5*Resolution+0.00020358
- 0.70 Threshold: |Ε|=8.5529e-5*Resolution-0.00013539
- 0.75 Threshold: |Ε|=8.4381e-5*Resolution-0.0001484
- Input files can be found here.
Conclusions
- The general trends are quite similar when outliers are disregarded, both yielding a linear slope factor of ~8.5e-5 which greatly mirrors the Box case presented above.
This leads to the assumption that Surface Area calculations are fairly good for objects when complex geometry is not present, or when sharp corners or angles are present as with the packed cylinders.
Another interesting feature gleaned from these graphs is the point at which the trend predicts zero error in the calculations. In the 16mm case, this point exists around 0.75mm resolution. In the 32mm case, this point exists around 1.5mm. Both of these values are by mere inspection of graphs presented above, but seem to also indicate a linear trend in the resolution at which the least error is found. Simulations at these resolutions will follow to validate the assumptions above and this section will be updated.
16 Holes Packed
The setup for this simulation consists of:
- A domain 192mm by 192mm
- 16 cylinders w/ diameter 32mm cut in a solid block of PBX
- Resolutions(grid spacing dx dy): 0.125mm, 0.25mm, 0.5mm, 0.67mm, 1.0mm, 1.25mm, 2.0mm
- Atmosphere in holes at 600K (in order to trigger "burning cell" flag
- Runs including tight packed and loose packed (45 deg array and rectangular array respectively
- Burn Rate set = 0 in code to isolate surface area calculations
Tight Packed
32mm Diamter
- Plots of Surface Area Error:
- Logarithmic trend is apparent for the 3 different threshold values:
- 0.65 Threshold: |Ε|=0.0020225*Resolution0.13044
- 0.70 Threshold: |Ε|=0.00070245*Resolution0.55272
- 0.75 Threshold: |Ε|=0.00063158*Resolution0.6182
- Input files can be found here.
64mm Diamter
- Plots of Surface Area Error:
- Plots of Surface Area Error:
- Logarithmic trend is apparent for the 3 different threshold values:
- 0.65 Threshold: |Ε|=0.0014903*Resolution0.32661
- 0.70 Threshold: |Ε|=0.00094551*Resolution0.6283
- 0.75 Threshold: |Ε|=0.00082866*Resolution0.70061
- Input files can be found here.
16 Holes Not Packed
The setup for this simulation consists of:
- A domain 100mm by 100mm (200mm by 200mm for 32mm diamter)
- 2 cases: 16 PBX cylinders w/ diameter 16mm and 32mm
- Resolutions(grid spacing dx dy): 0.125mm, 0.25mm, 0.5mm, 0.67mm, 1.0mm, 1.25mm, 2.0mm
- Atmosphere in between cylinders at 600K (in order to trigger "burning cell" flag
- Burn Rate set = 0 in code to isolate surface area calculations
16mm Diameter
- Plots of Surface Area Error:
- Linear trend is apparent for the 3 different threshold values:
- 0.65 Threshold: |Ε|=-0.0001083*Resolutio+0.00016829
- 0.70 Threshold: |Ε|=-0.00010008*Resolution-3.8368e-5
- 0.75 Threshold: |Ε|=-9.786e-5*Resolution-7.4006e-5
32mm Diameter
- Plots of Surface Area Error:
- Logarithmic trend is apparent for the 3 different threshold values:
- 0.65 Threshold: |Ε|=-6.1445e-5*Resolution+0.00028806
- 0.70 Threshold: |Ε|=-4.3967e-5*Resolution+0.00010531
- 0.75 Threshold: |Ε|=-8.982e-5*Resolution-0.00015026
Wedge
Results of different angle wedges:
Wedge w/ Absolute Error
- ^ graph of wedge cases with absolute error
Wedge relative error
- ^ graph of wedge cases with relative error
The setup for this simulation consists of:
- Domain is 100mm by 100mm
- 1 case: a triangular block of PBX from upper left corner to lower right corner angled 45 deg to the grid and orthagonal to xy plane
- Resolutions(grid spacing dx dy): 0.125 mm, 0.25mm. 0.50mm, 0.67mm, 1.0mm, 1.25mm, 2.0mm
- Atmosphere in upper right is 600K to trigger "burning cell" flag
-
- Burn Rate set = 0 in code to isolate surface area calculations
Results:
- Plots of Trends:
- Logarithmic trend is apparent for the e different threshold values:
-
- 0.65 Threshold: |Ε|=-3.7001e-6*Resolutio-3.2585e-5
- 0.70 Threshold: |Ε|=-3.3709e-6*Resolution-3.3028e-5
- 0.75 Threshold: |Ε|=-3.174e-6*Resolution-3.3303e-5
- Input files can be found here.
Misc.
All file linked to by this page can be found in this directory.
Back to Revision 3 Results
Resolution Home
Home
Contact: josuf dot the dot uf at gmail
Updated: 04/09/09
Created: 02/23/09