Impact of Reaction and Location of a Pilot Jet on the Flow Structures in a Co-annular Swirl Burner

The paper investigates the impact of heat release and pilot jet location on the swirling flow structures in a coannular swirl burner without confinement. Two iso-thermal turbulent swirling flow cases and three turbulent premixed swirling flames are studied with the aid of large-eddy simulation. Two geometries are considered, one featuring a pilot lance flush with the outlet and one with retracted pilot lance. The Reynolds number is 35,000. Detailed comparisons with experimental data are performed for velocity statistics and generally good agreement is achieved. Furthermore, coherent structures are analyzed and local power spectra computed. In the non-reactive case with retracted pilot jet, a very strong precessing vortex core (PVC) is observed in both numerical results and the experiment. The PVC is weak in the non-retracted case. In the reactive cases where the whole central recirculation zone (CRZ) is enclosed in the high-temperature post-flame region, the PVC is almost completely suppressed. INTRODUCTION Turbulent swirling flows are important and widely used in energy production devices such as internal combustion engines, gas turbine combustors, industrial burners and boilers. A strong inlet swirl induces a central recirculation zone (CRZ) in the flowfield which, recirculates the hot combustion products and the radicals and anchors the flame at the burner. In lean premixed combustors, in addition to the CRZ, an extra pilot jet with richer mixture is often utilized for enhanced flame stabilization. In the resulting complex flow system, flow instabilities are observed in both experiments (Habisreuther, et al., 2006) and simulations (Fröhlich et al., 2008). The ability to predict the size, shape and the position of the CRZ is essential to the prediction of turbulent swirling flows with and without reaction. Generally, the CRZ is not stable but precesses around the axis, hence termed precessing vortex core (PVC). The occurrence of a PVC in isothermal swirling flow depends on several factors such as swirl number and geometrical setup etc. Under combustion conditions its behaviour is dependent further on equivalence ratio, mode of fuel entry, flow rate, etc. Numerous studies have been reported on the influence of different parameters such as the level of swirl, the inflow profile, the flow configuration, Reynolds number, and heat release etc., as reviewed in (Lucca-Negro and Doherty, 2001; Syred, 2006). The main objective of the present work is to investigate the impact of heat release and pilot jet location on the swirling flow structures in an unconfined co-annular swirl burner. To this end, two iso-thermal turbulent swirling flow cases and three turbulent premixed swirling flames are simulated using large-eddy simulation (LES). NUMERICAL METHOD To simulate the turbulent premixed combustion, the thickened-flame (TF) model (Colin et al., 2000) is employed. Although it is comparatively simple it has shown to perform well for the type of flow studied in this work (Selle et al., 2004). In the TF model, pre-exponential constants and transport coefficients are both modified by a factor F to yield a thicker reaction zone which can then be resolved on an LES mesh. The subgrid-scale wrinkling is accounted for by an efficiency function E related to the local subgrid-turbulent velocity and the filter width. The equation for the mass fraction of the kth species is thus modified to read ( ) ( , ) th th th k j th th k k k k j j j j Y u Y Y E D EF Y T t x x x F ρ ρ ρ ω ∂ ∂ ∂ ∂ + = + ∂ ∂ ∂ ∂ (1) where the superscript ‘th’ represents thickened quantities. With this modification, the resolved flame propagates at a turbulent flame speed equal to E times the laminar flame speed, while the resolved flame thickness is F times the laminar flame thickness. The wrinkling function E is determined as proposed by Colin et al. (2000). The —⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯ * Currently a guest scientist at TU-Dresden thickening factor F is a local quantity and generally estimated as 0 ( ) L F n n δ = Δ (2) where Δ is the characteristic grid cell length scale, and 0 L δ is the laminar flame thickness. Hence, the thickened flame is resolved with n grid cells. An optimal value for the parameter n generally is in the range 5-10 (T. Poinsot, private communication). In the present work, the effect of n is studied, demonstrating that n=5 is a good choice for the flow cases under consideration. The dynamic procedure of (Legier et al., 2000) was used to determine F. It reduces F to unity remote from the flame and hence provides realistic transport coefficients there. To simulate the lean premixed methane/air combustion of the experiment, the two-step chemical scheme 2sCM2 (Selle et al., 2004) is used which takes into account six species (CH4, O2, CO2, CO, H2O and N2) and two reactions: (a) CH4 + 1.5O2 → CO + 2H2O (b) CO + 0.5 O2 ↔ CO2 . These models were implemented in the Finite-Volume code LESOCC2C (Wang et al., 2008 and 2009), solving the low-Mach number version of the compressible NavierStokes equations on block-structured body-fitted curvilinear grids. BURNER SETUP AND COMPUTATIONAL DOMAIN The investigated swirl burner by Büchner and his coworkers (2004) is sketched in Fig. 1a, together with the computational domain (not to scale). It consists of two coannular swirling jets discharging into the ambient air which is at rest. The tube in the centre of this setup (indicated by gray color) is the pilot lance. The Reynolds number of the flow based on the bulk velocity Ub=11.5 m/s, and the radius of outer jet nozzle R=55 mm, is Re=35,000. The mass flow rates are 180 m/h and 20 m/h at normal condition, and the theoretical swirl numbers are 0.9 and 0.79 for the main and the pilot jet, respectively. Swirl is co-rotating. The blades of the axial swirl generator for the pilot jet used in the experiment were represented geometrically using 12 inclined blades as seen in Fig. 1b. For the inflow of the main jet, 12 gaps where positioned regularly along the circumference of the cylindrical surface, with radial and tangential velocity component being imposed. Different swirl levels are achieved by adjusting the tangential velocity component and maintaining the radial velocity component. A slip condition was used on the outer boundary and a convective outflow condition was imposed at the exit. The Werner-Wengle wall function was employed at all solid walls. A very small uniform co-flow was imposed remote from the jets as validated in (GarcíaVillalba, et al., 2006). Two geometrical setups are studied: pilot jet flush with outlet (Fig. 1a) and pilot jet retracted by 40 mm (Fig. 1c). To study the impact of reaction and pilot jet location on the flow, five cases were simulated as listed in Table 1. In the reactive flow cases, the equivalence ratio is 0.667 and 0.833 for the main jet and pilot jet, respectively. (b) (c) Fig. 1. (a) Sketch of the computational domain, pilot jet flush with main jet, units is mm. (b) 3D view of the pilot swirler located at the position indicated by gray color in part (a) and (c). (c) Pilot jet retracted by 40 mm. Table 1: Overview over the cases simulated, definition in the text. cases Reactive Retracted n Grid/10 Δt/μs r1/2 I00 no no / 8.45 6.69 1.4 I40 no yes / 8.88 6.69 1.6 R00 yes no 5 4.34 2.84 2.0 R40 yes yes 5 4.56 2.52 1.7 R40B yes yes 10 4.56 1.25 1.7 RESULTS AND DISCUSSION Effect of Thickening-Factor The influence of the parameter n in Eqn. 2 is visualized with Fig. 2, displaying instantaneous reaction rates from case R40 and R40B. With small n, more flame wrinkles are resolved, which indicates that the interaction between turbulence and flame is better resolved so that the contribution from the SGS wrinkling model is smaller. In addition to this, a number of 1D laminar flame simulations were performed with different values of n and aimed to compare the integral heat release across the flame front. It was found that the relative difference among the results obtained from the thickened flames with n=5 and n=10 and the DNS result is smaller than 0.6%. The main benefit with small n is the larger time step allowed in the computation (see Table 1). Fig. 3 shows an instantaneous temperature field at the center plane obtained from case R40, and the corresponding distribution of the dynamic thickening-factor. It demonstrates that F is unity in the post flame zone and the unburnt region, but has large values in the flame brush to resolve the thickened flame with multiple grid cells. 40 30 57 70 D=110