Best Practice Advice AC2-06
The confined TECFLAM swirling natural gas burner
Application Challenge 2-06 © copyright ERCOFTAC 2004
Best Practice Advice for the AC
Key Fluid Physics
The key physics of confined swirling flows is the formation of recirculation zones inside the combustion chamber. These recirculation areas are generated by introduction of swirl into the flow. The formation of the recirculation zones is often used in practical technical applications, although its physical aspects are far from being fully understood.
The overall characteristics of the flow field considered within this AC can be divided into three regions:
1. the mixing zone of fuel and air (combustion takes place predominantly here)
2. the inner recirculation zone (around the flame axis), and
3. the outer recirculation zones (close to the walls of the combustion chamber)
If the flow with chemical reactions is considered, especially the combusting flows, such a flow field can be generated by introduction of the swirl into flow of air or fuel as well as into both of them. For the case of two swirling jets, recirculation zones can be observed while the jets are in the co-swirl or counter-swirl state. The value of the swirl number has an important influence on the size and location especially of the central recirculation zone.
The formation of recirculation zones in such a flow type enhances mixing of fuel and air, increases heat transfer as well as stabilises the flame due to recirculation of hot combustion products. Introduction of swirl into the flow makes diffusion flame short and stable.
The size of each region as well as their locations in the flow are of great interest, since they play major role and have significant influence on combustion process. They are also very important from the environmental point of view, because the presence of central recirculation zone decreases the combustion temperature, which directly reduces emission of NOx. The complexity of the flow field and combustion processes causes that computational results obtained from CFD tools are often not satisfactory. Moreover, since the AC concerns flows with combustion the major as well as minor species like CO or NOx emissions are also of great importance and their predictions should be taken into account.
2.1 Numerical uncertainties
The first uncertainty of this AC concerns boundary conditions in the combustion chamber walls. The experimental setup consists of the combustor with water-cooled walls in course of which 50% of thermal load is transferred, while for the numerical calculations the adiabatic boundary conditions were assumed. The numerical results show however that such boundary conditions give good agreement with experiment of the temperature value and distribution inside the central recirculation zone, even that in the outer zone the temperature is slightly overestimated. Further downstream significant heat looses due to convection and radiation not taken into account in CFD predictions lead to significant overestimation of numerically predicted temperature compared to experiment data.
The second very important uncertainty can be derived from the studies performed within UFR4-02 related with the AC under consideration. It turns out that the boundary condition formulated at the outflow could influence significantly the flow field predictions. The character of the central recirculation zone can be completely different i.e. with presence of a contraction or expansion at the end of the test section.
2.2 Experiment uncertainties
The main experimental uncertainty concerns measurements of major species O2 and CO. Due to a photon statistics, the accuracy of single-pulse measurement is reduced, and for O2 with mole fraction of 0.2 at 2000K is 12¸15%, while for CO it even lower and for mole fraction 0.06 is about 20% and for mole fraction 0.02 it is 50%.
Computational Domain and Boundary Conditions
3.1 Computational domain
The calculations were performed for a variety of turbulence models in a 2D axi-symmetric case. The computational domain was bounded by the burner axis of symmetry, the burner wall, free outlet boundary and the inlet to the combustor. However, instead of modelling the geometry of the swirler, the measured flow field profiles were used.
3.2 Boundary Conditions
The inlet boundary conditions were assumed based on the measurements done with LDV. Distributions of the velocity and turbulent quantities were used from the measurements at 1 mm downstream from the nozzle exit. The flow rates of the fuel and air were adjusted to match values at the same measurement location and the dissipation rate was prescribed based on the integral length scale. At the centreline the symmetry boundary conditions were assumed, while at the wall adiabatic boundary conditions and for the turbulence wall function was used. At the outlet boundary the outflow was assumed to have a parabolic character.
Discretisation and Grid Resolution
The grid used in calculations comprised 80 points in the axial and 60 points in the radial direction for which the results were found to be grid independent. The mesh was refined in the reaction zone.
Discretisation scheme used was 2D elliptic finite volume method with SIMPLEC pressure correction and TDMA solver.
The calculations were performed for a variety of turbulence and turbulence/chemistry interaction models. These include standard and RNG k-ε as well as Reynolds-Stress models for turbulence and finite rate chemistry, flamelet approach, β-PDF, ILDM and Monte-Carlo-PDF models. The numerical results presented in AC documentation show that there are no significant differences between k-ε and RSM models as far as the mean velocity is concerned. However, it turns out that predictions obtained for both models are far from experimental data especially further downstream in the combustor. On the other hand the literature studies documented in UFR4-02 showed that for the confined coaxial swirling jets a significant improvement of the mean velocity field predictions could be attained using RSM instead of k-ε model. Moreover, in the UFR4-02 it was concluded that only when using LES the correct mean velocity could be obtained for swirling jets. In particular the CFD results presented in the AC documentation showed a significant underestimation of the velocity in the inner recirculation zone using both k-ε and RSM models which could significantly affect the results of combustion predictions.
The best results for the velocity profiles, mixture fraction, kinetic energy as well as CO and CO2 distributions are obtained by RSM turbulence model interacting iteratively with the Monte-Carlo PDF transport approach (with 100 particles per cell) and ILDM tables for chemistry tabulation, however for the Monte-Carlo PDF approach the calculation time was more then 10 times longer than for the flamelet model with presumed b-PDF.
Recommendations for Future Work
Modelling of combusting flows stabilised by the recirculation of hot exhaust gases generated by the swirl imposed on the fuel and oxidizer streams requires the precise predictions of the mean velocity field and turbulence structure which determines mixing intensity.
The AC2-06 and UFR4-02 documentations showed that RANS models lead to significant discrepancies of CFD predictions with respect to experimental data for mean and fluctuating velocity fields. Certain improvement of the mean velocity field predictions can be attained using RSM instead of k-ε model but turbulence structure in most cases remains poorly predicted. It seems that the further improvement of swirling flows predictions can be obtained only by the application of the LES model. However, it is necessary to keep in mind that the LES calculations require very long CPU-time especially in flows with complex geometry even for isothermal non-combusting flows. Another important factor limiting applicability of the LES technique in the AC considered is the presence of combustor walls near which a very high mesh resolution is required for LES application. It can be expected that the flow structure predictions in the inner and outer recirculation zones can be significantly improved with LES calculations while combustor walls influence should be for the time being neglected.
As it was showed in the AC2-06 documentation the PDF transported combustion model requires one order of magnitude longer computational time compared to flamelet presumed β-PDF model. Hence it is recommended for future work to examine the LES method for the confined TECFLAM swirling burner with the finite rate chemistry flamelet model with presumed β-PDF.
© copyright ERCOFTAC 2004
Contributors: Stefan Hohmann - MTU Aero Engines