How Sensitive is a Combustion Simulation to Changes in the way Chemistry is Handled?
A combustion simulation is a notoriously tricky thing to get right. There are any number of things which can affect the results, but for this page, we're just looking at the differences due to a few options for the treatment of chemistry.
The effects of changing the turbulence model are examined
As in that study, the case examined here was based on (but not identical to) the experiments of Burrows and Kurkov on supersonic combustion. In these simulations, hydrogen is injected at Mach 1 parallel to the main stream of air from a backward-facing step. The primary flow consists of vitiated air traveling at Mach 2.44. The hydrogen is ignited by the hot air and burns as it convects downstream.
Normally, combustion simulations would try to match both the location and the extent of the observed combustion region, as well as the distribution of chemical species, but because the simulations had a different chemical composition for the vitiated air than the experiments, direct comparison is not possible. For this study, we were more concerned with the differences between the simulations than the absolute value of any of the predicted quantities.
The Baseline Case
After tinkering with various options in the code, a “baseline” case was created. This simulation used the compressible formulation of Menter's SST turbulence model, the latest NASA curve fits (from NASA TP-3287) for thermodynamic data, and included the effects of third body efficiency in the chemical reactions. The reaction set itself, with seven species and eight reactions, was based on one by Evans and Schexnayder. Coefficients for both forward and backward reaction rates are specified. Static temperature contours for this case are pictured below.
Effect of Different Thermo Data Curve Fits
In the Wind-US solver, which was used for this study, thermodynamic properties (such as enthalpy, heat coefficient at constant pressure, etc.) for each chemical specie are computed using curve fits in one of three functional forms. One of the issues examined in this study was whether or not it mattered which curve fit was used. To do this, the case was re-run using the SPARK curve fit (which is itself a modification of the NASA Lewis CET86 high temperature thermo database).
The results of this run are shown below. From the plot, it is obvious that for this case the curve fit made almost no difference at all. This is good news for the Wind-US code, because most of the chemistry files that ship with the code still use the older formats.
Effect of Including Effective Binary Diffusion
The next case looked at the effect of using an effective binary diffusion calculation for species diffusion (to better account for the interaction of molecules of different species). Unlike the previous test, using the effective binary diffusion algorithm causes the reaction region to move upstream about four centimeters, as shown in the plot below. This change is of the same order as that observed when switching from one of the two equation turbulence models to the Spalart model (see
the page on the effects of turbulence model
for more on this issue).
Effect of Ignoring Third Body Efficiencies
The next test re-ran the case with one of the chemistry sets that ships with the solver. As with the baseline case, the reaction set is based on the Evans and Schexnayder mechanisms. This particular set used the SPARK curve fits (which we already saw do not change anything from the baseline) and also ignored third body efficiency in dissociation reactions.
Making these changes results in a significant delay to the onset of combustion, as shown below. Furthermore, even when ignition occurs, the temperature rise is greatly reduced, as shown by the lack of red contour levels in this plot.
Effect of Reaction Rate Algorithm
An even bigger change is seen when a chemistry reaction set is used which specifies only forward rates. The backward rate is found using the forward rate and the equilibrium constant. In this case, no reaction is seen at all before the flow exits the domain.
The conclusion from the above simulations is, above all, that the reaction set used for a combustion simulation is critical. Also significant are the effects of neglecting third body efficiency and effective binary diffusion. These latter two items are apparently just as important as the
choice of turbulence model was found to be.
In contrast, the form of the curve fit used for the thermodynamic data made no difference at all for this case.
Of course, every combustion simulation is going to be somewhat different, but the above findings are consistent with what I have seen elsewhere. Therefore, if the conclusions are not perfectly general, they are at least a good place to start when discussing the factors that affect CFD of reacting flows.
Another issue in combustion simulations, which has not been addressed here is the basic reaction mechanism used to model the chemistry. All the cases here used variations of the Evans and Schexnayder mechanisms. The more up-to-date Westbrook or Dryer mechanisms would likely be better choices, but they have not been set up for the Wind-US code which was used for these simulations.
If you want to learn more about the sensitivity of combustion simulations to various factors, see the pages on
the effects of turbulence model
the impact of various other code options.
When you are finished, you can
leave the effects of chemistry options on combustion simulations and return to the CFD applications page.
Or you can head back to the
main Innovative CFD page
and browse the other topics there.