Sei sulla pagina 1di 4

Conference on Turbulence and Interactions TI2006, May 29 - June 2, 2006, Porquerolles, France

Large-eddy simulation of multi-component compressible turbulent ows using high resolution methods
B. Thornber, , D. Drikakis, , D.Youngs ,
Fluid

Mechanics and Computational Science Group, Department of Aerospace Sciences, Craneld University, United Kingdom Email: b.j.r.thornber@craneld.ac.uk, Email: d.drikakis@craneld.ac.uk, AWE, Aldermaston, UK, Email: David.Youngs@awe.co.uk ABSTRACT

The ability of two very different Large Eddy Simulation (LES) codes to predict shock induced turbulent mixing has been tested through simulations of the Half Height experiment [3]. Initial simulations of homogeneous turbulence demonstrate the capability of the codes to reproduce turbulent behaviour such as the expected kinetic energy decay rates and growth of the length scales. Simulations of Richtmyer-Meshkov and Kelvin-Helmholtz instabilities in turbulent multicomponent ows demonstrate very good results in qualitative comparisons with experimental results

I NTRODUCTION

This paper focuses on the three dimensional simulation of shock induced turbulent mixing of two different gases via Kelvin-Helmholtz (KH) and Richtmyer-Meshkov (RM) type instabilities. The incident shock wave passes through a perturbed interface between two gases, inducing a shear ow. In this type of mixing the form of the initial pertur2 T EST C ASES bations affects the growth of the resulting mixing layer, and the initial perturbations should be allowed to grow without being dissipated (or enhanced) by the numerical method. Additionally, the monotonic- The rst test case examines the ability of the two ity of the various gas properties must be maintained, methods to resolve homogeneous freely decaying turbulent ow in a cube. The ow eld is initialised and the shocks captured accurately. following [5] with an initial kinetic energy spec4 With these requirements determined, two numeri- trum E(k) k 4 /kp exp 2 (k/kp )2 , where k is cal methods have been selected which differ signif- the wavenumber, the peak kp is chosed at k = 4, icantly in approach. CNS3D is a Finite Volume Go- the eld rescaled to Mach = 0.1 and kinetic energy dunov method using a characteristics based Riemann per unit volume is 0.5. Simulations were run at ressolver [2], where higher order accuracy is achieved olutions of 643 , 1283 , and 2563 with CNS3D utilthrough van Leers MUSCL limiting technique and ising the Minmod, van Albada, Superbee and van the recently proposed Total Enthalpy Conservation Leer limiting methods, and at 643 and 1283 with of the Mixture model is employed to model the gas TURMOIL3D [5]. Turbulent quantities were avermixture [4]. TURMOIL3D is a third order semi- aged over approximately 8 eddy turnover times. The Lagrangian scheme with a mass fraction advection kinetic energy decay rate was computed assuming equation for gas mixtures [5]. Both methods are a solution of the form K.E. (t t0 )p , thus based on a form of Large Eddy Simulation where u2 /u2 = pA(t t0 )p1 /A(t t0 )p = p/(t t0 ),

the numerical dissipation is employed as a subgrid model in lieu of explicit subgrid terms, to ensure positivity of the species concentration and the monotonicity of the resulting solution. This is referred to as Implicit LES (ILES). As the Kolmogorov length scale is less than the grid resolution, the viscous terms are neglected.

allowing an objective estimate of the decay rate [5]. The time origin t0 is xed where the skewness of the velocity derivative reaches a constant value. CNS3D gave 1.2 < p < 1.4 for all limiters at all mesh resolutions, apart from the Superbee limiter where the decay rate oscillated signicantly due an unphysical pile-up of kinetic energy at high wavenumbers. In addition the growth of the integral length scale follows a 2/7 power law to begin with, gradually moving to a 2/5 law at later times. This is in accordance with current turbulent theory. The van Leer limiter is chosen for further simulations as at lower mesh resolutions the velocity derivative skewness is smaller ( 0.4 at 2563 ) indicating that there is less dissipation at the higher wavenumbers. For TURMOIL3D the decay rate exponent p 1.41 in good agreement with theoretical predictions, and a Kolmogorov subinertial range is observed in the kinetic energy spectra [5]. Fig. 1 shows a schematic of the half height experiment (See Holder and Barton [3] for the full details). A Mach 1.26 shock wave in air passes through a block of SF6 initially held in place by a microlm membrane, and seeded with olive oil droplets. The lower half of the shock passes slowly through the block, and more rapidly through the air above the block, producing KH instability at the upper interface, with RM instabilities at the two vertical interfaces. Transition to turbulence occurs in the spiral of the large primary vortex promoting mixing dominated by turbulent transport of the individual gases. The shock reects off the end wall of the tube, increasing the mixing rate at the interfaces and producing a complex set of reected, refracted and transmitted shocks.

and 1280320640 with TURMOIL3D to examine mesh convergence. The z-direction (vertical) wall boundary conditions are free slip, and periodic in the y-direction. Random initial perturbations with RMS amplitude of 0.1mm at wavelengths between 5mm to 50mm are added to the vertical surfaces to model the effects of the membrane. Note that the drainage hole was not modelled. Fig. 2 compares the experimental and computational results.The experimental images are more diffuse than those in the simulations because of multiple scattering effects [3]. In the rst two time steps the two different codes and the experiment give very similar results, and the only disagreement is at the primary vortex which remains more coherent in the simulations than in reality. At 1.006ms the rst differences between the numerical methods occurs as instabilities in the primary vortex in TURMOIL3D, which are suppressed in CNS3D. At this point the angled shock converges near the wall causing high pressures, densities and velocities, leading to the bump visible at the right hand interface. The incident shock, reecting from the end wall, passes through this bump triggering an RM instability producing the large mushroom shaped injection of heavy gas. At 2ms both codes show good agreement in terms of the position of the primary vortex, and the presence of signicant turbulent mixing in the vortex core. This mixing occurs at a smaller scale in TURMOIL3D than with CNS3D. The mushroom shaped injection of heavy gas is orientated upwards in both simulations, however in the experiment this remained horizontal. It is believed that this discrepancy is due to not modelling the drain hole.

The primary vortex reaches the top of the shock tube at 3ms in both experiment and simulations, indicating that the growth of the length scales present in the experiment are caputered well in the simulation. At 4ms the thin line connecting the remains of the block with the primary vortex is less mixed in the simulation than in the experiment. In the experiment this mixing is caused by an RM instability at a scale below that which is captured in the lower resolution simulation. The position of the shock wave and SF6 block have been taken from the experimental data Fig. 1. Schematic of the half height experiment and a comparison of these can be seen in Fig. 3. The Simulations have been run with mesh resolutions of agreement is good with only a small discrepancy in 600 160 320 for CNS3D and at 640 160 320 the angle of the refracted shock, which has a less

(a) 0.20ms

(b) 0.37ms

(c) 1ms

(d) 2ms

(e) 3ms

(f) 4ms

(g) Line average SF6 density at 4ms. TURMOIL3D ne resolution (left), medium (centre), CNS3D (right) Fig. 2. a) - f) Comparison of experimental images (left, c British Crown Copyright 2006/MOD) and mass of SF6 density (kg/m3 ) for TURMOIL3D (centre) and CNS3D (right)

steep angle in the experiment due to initial diffusive Fig. 2 g) shows the y-averaged SF6 density at time mixing at the upper SF6 /Air boundary. 3.996ms, and Fig. 4 the plane averaged quantities < f1 f2 > and < f1 >< f2 >, where f1 is the volThe level of convergence of the results is examined ume fraction of Air and f2 that of SF6 . An improveusing the high resolution TURMOIL3D simulation. ment can be seen in the level of mixing in the thin

C ONCLUSIONS

0.2ms

0.37ms

1ms
Fig. 3. Comparison of experimental shock and SF6 positions (dashed line) and numerical results (solid line) for TURMOIL (Left) and CNS3D (right)

strip at high resolution, however there there is good agreement in the magnitude and location of the peaks of molecular mixing in all simulations. The extent of the mixing in the primary vortex (0.12 < x < 0.23) does not change signicantly with a doubling in mesh resolution.

This paper has compared the performance of two very different MILES codes in simulating an experiment involving complex shock induced mixing. The two codes have been tested initially on homogeneous decaying turbulence where the best MUSCL conguration was selected for CNS3D, and TURMOIL3D shown to give good decay rates and spectra. Comparison with the half height experiment shows excellent qualitative agreement with the available experimental data in terms of the size, location and temporal growth of the main ow features. Convergence was tested against a high resolution simulation which gave improved mixing at the thin strip between the mushroom and primary vortex features, however the mixing in the primary vortex was not affected greatly. In addition, the shock positions are captured accurately demonstrating the effectiveness of the multicomponent models. In comparing the codes, TURMOIL3D allows more ne scale turbulent mixing than CNS3D with van Leer limiting, however CNS3D captures the shock waves more accurately. The two codes are in very good agreement for overall degree of mixing, as indicated in Fig. 4. The authors would like to acknowledge support given via an EPSRC-AWE CASE award.

BIBLIOGRAPHY

[1] Drikakis, D., Rider, W., High-Resolution Methods for Incompressible and Low-Speed Flows, Springer Verlag, 2004 [2] Eberle, A. Characteristic ux averaging approach to the solution of Eulers equations, VKI Lecture Series, Report 1987-04, 1987 [3] Holder, D.A., Barton, C.J. Shock tube RichtmyerMeshkov experiments: inverse chevron and half height, 9th IWPCTM, 2004 [4] Wang, S.P., Anderson,M.H. , Oakley, J.G., Corradini, M.L., Bonazza, R., A thermodynamically consistent and fully conservative treatment of contact discontinuities for compressible multi-component ows, J. of Comp. Phys. 195, pp. 528-559, 2004 Fig. 4. Comparison of plane averaged mixing parameters < f1 f2 > and < f1 >< f2 > at 3.966ms, where x is measured from the rst SF6 interface [5] Youngs, D.L., Three - dimensional numerical simulation of turbulent mixing by Rayleigh-Taylor instability, Phys. Fluids A 3(5), pp. 1312-1320, 1991

Potrebbero piacerti anche