Improvement of DNS tools for Air-assisted
Transcription
Improvement of DNS tools for Air-assisted
Improvement of DNS tools for Air-assisted Primary Atomization Simulations A. BERLEMONT1, J.L. ESTIVALEZES 2, S. ZALESKI 3 CNRS-CORIA1, ONERA 2, UPMC 3 T2.2.1 FIRST Project Spray and Soot Workshop B1-15 FIRST Project Spray19-20 and Soot Workshop, 19-20 November November 2014, Derby, UK 2014, Derby, UK General context Atomization: a multi-scale problem Atomization is a complex multi-scale phenomenon Example: air-blast atomization of a liquid sheet Injector width ≈ 4 cm Finest droplets ≈ 10 µm A thin liquid sheet is injected between two parallel high-velocity air flows Growing aerodynamic instabilities triggered by the shearing effect induce the sheet disintegration Sheet thickness ≈ 0.3 mm channel length ≈ 30 cm A. Cartellier, A. Delon, J.P. Matas, LEGI B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 2 General context Atomization: a multi-scale problem • Some orders of magnitude: Liquid sheet thickness : between 300 and 600 microns Mean break-up length : between 5 mm and 25 mm ( depending on velocity ratio) Size of primary atomization zone : around 5 cm x5 cm x 5 cm Smallest droplet diameter : 10 microns Minimum droplet resolution needed for DNS : 10 to 15 grid points/ diameter High liquid-gas density ratio conservation problems High shear at liquid gaz interface • Uniform grid : 125 1012 grid points • AMR grid : 125 109 grid points primary atomization secondary atomization B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 3 General context Atomization: a multi-scale DNS approach • Fully resolved DNS are still out of reach even with AMR alone • Solutions: Parallel computing (High Performance Computing) AMR ( Adaptive Mesh Refinement) Improvement of mass and momentum conservation Multi-scale simulations (Euler-Lagrange coupling) B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 4 Coria Contribution-code ARCHER: Introduction Objective • Numerical simulations are known for becoming unstable for atomization processes with large density ratios coupled with large shear. Development of a new consistent mass and momentum flux computation using Rudman type technique in ARCHER code (coupled Level Set / VOF / Ghost fluid methods ) Simulations are performed on air assisted atomization on experiments from LEGI (T3.1) and presented in T4.1.4 A. Cartellier, A. Delon, J.P. Matas, LEGI A. Cartellier, A. Delon, J.P. MATAS, LEGI A. Berlemont, T. Ménard, G. Vaudor, CORIA • Development of adaptive mesh refinement algorithm for incompressible two phase flows B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 5 Work Description and Results • Problem description NS formulation • V .( V V) p .( D) g t .V 0 t Incompressible NS • V (V .)V p .( ( ) D) ( ) g • t V . 0 ( ) In incompressible NS, we lost consistency between mass flux and momentum flux at the interface: the V flux must be the same in both equations in NS equation, but disappears in Incompressible formulation. This induces numerical instabilities in simulations, or wrong results. We thus need to ensure consistency in ARCHER algorithms. G. Vaudor, A. Berlemont, T Ménard, M. Doring A Consistent mass and momentum flux computation using Rudman type technique with a CLSVOF solver, FEDSM2014, Chicago, USA, August 3-7, 2014 B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 6 Work Description and Results • Objective : Use VoF (=Mass) flux to compute momentum flux. • Mass fluxes are known on the faces of the black cell (A,B,C,D) • Problem: Faces for momentum equation are not in the same place (staggered grid) (orange and purple cells). We need fluxes A’,B’,C’,D’ • Flux A’ and C’ are unknown ( B’ and D’ not directly but easy) Two approaches: Rudman technique with a dual grid for CLSVOF (twice smaller) Our approach: mass balance on a half cell B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 7 Work Description and Results • Validation on turbulent round jet experiment (Sedarski et al): measurements of interface velocities 1 2 1 3 4 250 Simulation • Experiment Velocity [m.s-1] 200 150 4 100 50 0 0 200 Radial distance 400 [µm] 600 800 Good agreement for Interface velocities Sedarsky D., Idlahcen S., Rozé C. Blaisot J.B., 2013, Velocity measurements in the near field of a diesel fuel injector by ultrafast imagery. Experiments in Fluids, Vol. 54:2, 1-12. B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 8 Work Description and Results • Adaptive Mesh Refinement Done : Data structure Adaptive mesh Convective term Very difficult work developed by Poisson Solver M. Doring (moved in June 2014) Navier-Stokes equation Now developed by T. Ménard Coupled VOF – Level Set Validations Last developments Parallelisation Load balancing … New discretization of N-S (Consistent fluxes) B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 9 Work Description and Results • 3D axi-symetric jet (CNRS LEGI) ( see WP 4.1.4) 8 mm cylindrical water jet (0,27 m/s) surrounded by a 1.7mm thick air flow (22,6m/s) with new convection scheme is introduced AMR with load balancing • Example of Domain MPI cutting Refinement blocks B1-15 One level refined grid FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 10 Work Description and Results Adaptive Mesh Refinement • • • A. Cartellier, A. Delon, J.P. Matas, LEGI B1-15 642x128 +1 level – 64 procs 642x128 +2 level – 64 procs 642x128 +2 level – 128 procs A. Berlemont, T. Ménard, G. Vaudor, CORIA FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 11 Onera contribution-code DyJeAt: AMR and lagrangian tracking Coupled LS/VOF method is based on Sussman and Puckett (JCP 2000), Menard et al. (IJMF 2007) VOF : mass conservation Level-set : good description of the interface Geometrical computation of the fluxes [VOF Tools, Hernández and López 08] Adaptive Mesh Refinement Oct-tree block refinement “Near interface” refinement criterion Droplets generation Small liquid inclusions: Under resolved by CLSVOF (2-6 cells) Local mesh refinement triggered the spray would create an everywhere-refined mesh Lagrangian tracking of stable droplets mesh de-refinement, dilute flow hypothesis satisfied! Problem Dispersed phase models definitely not satisfied at the droplet creation! B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 12 Droplets detection procedure 1) Tag cells where C > 0.5 (“liquid cell”): integer index 2) Propagate the minimum index among all the tagged neighbour liquid cells 3) Extension of the tag to the neighbouring cells : C< 0.5 4) Parallel propagation of the tags 5) Size criterion 6) a) Ni number of liquid cells contained in an inclusion b) if Ni < 2m (m=ndim) point lagrangian particle small droplets c) For 2m < Ni < 6m medium droplets (volumic particle) d) In the others cases eulerian treatment is done (CLSVOF) Shape criterion (medium droplets) Droplet generated if a) Sphericity b) Volume B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 13 Medium droplets treatment “Medium” droplet treatment (kind of penalty method) 1) Mass and momentum projection on the Eulerian grid: 2) Two-phase N.S. solve + 3) Averaging the new velocity inside the droplet: 4) Move the droplet according to : (0) Old position B1-15 (1) Projection (2) NS solve (3) Lag transport FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK (4) New position 14 Small droplets treatment and reimpact “Small” droplet treatment : EulerLagrange two-way coupling (DPS-DNS) • Interpolate gas velocity @ particle location : • Solve the following equations for the point particle (Runge-Kutta 2): • With , , Droplet impact • Reimpact allowed if : • Level-set and VOF reconstruction inside the droplet • Momentum coupling : • Level-set redistance B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 15 Numerical tests: droplet splashing, We = 272 Mesh Δx Euler Mesh 2Δx Euler+Lagrange B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 16 Numerical tests: surface impact Classical treament Lagrangian projection ( medium droplet reimpacttreatment) B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 17 Numerical tests: impact on a pool B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 18 Atomization of a 3D liquid sheet, ambiant conditions Ug = {40, 60, 80} m.s-1 Ul = 2 m/s P = 1 bar M = 2.6 Air-water Liquid sheet thickness 2a =300 m Splitter plate thickness=150 m AMR-CLSVOF-Eul-Lag calculation Dxmin = a/6 B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 19 Atomization of a 3D liquid sheet, engine conditions Ul = 2 m/s Ug = 17 m/s P = 10 bar M = 2.62 Air-Kerosene Liquid sheet thickness =300 m Splitter plate thickness=850 m Air boundary layer thickness=1.5 mm AMR-CLSVOF-Eul-Lag calculation Dxmin = a/10 (see WP 4.1.4) B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 20 UPMC Contribution-code Parissimulator • • • • • • • • • • • • • New UPMC code previous results obtained with Gerris Gerris does not perform well on large numbers of processors (> 64) write a new code, Paris Simulator, with the following features: Free code under the GPL. Written in Fortran 90/95 Use a regular structured grid and finite volume discretisation. Uses MPI and a regular array of subdomains for parallelisation. Momentum : QUICK, ENO, Verstappen or a momentum-conserving method Implicit viscous terms. Either Volume-Of-Fluid or Front-Tracking may be used to follow interfaces. VOF uses either CIAM or Weymouth-Yue scheme for advection. Height function method for curvature estimation and surface tension. Parallelized Lagrangian Point Particles (LPP) with VOF to LPP and LPP to VOF conversion • Second order in time and space with projection method. • Uses either hypre or internal Poisson solver. • Immersed solid boundaries B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 21 Parissimulator performances i): Parissimulator Speed issues • Facing scalability problems with Gerris code • Rewriting of a new code Parissimulator • Vof method + lagrangian coupling B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 22 Parissimulator performances ii): Perfect scalability on massive HPC (IDRIS Turing) Normalized Speed 16 8 4 2 1 1024 IBM BlueGene Turing Ideal 2048 4096 8192 16384 # of cores B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 23 Point Lagrangian coupling i) • Add Lagrangian Point Particles (LPP) that obey a point-particle equation du p dt = Fp éë u p ,u f (×) ùû • The force is determined by the surrounding carrier fluid velocity field: u f (×) • It reacts on the Navier-Stokes equation through a smoothing Kernel G r Du / Dt = -Ñp + Ñ ×(2mD) + skd S n - Fp *G • where the strain-rate tensor D is 1 æ ¶u j ¶ui ö Dij = ç + . ÷ 2 è ¶xi ¶x j ø B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 24 Point Lagrangian coupling ii) The choice of resolved interface or LPP modelling depends on the type of simulation and on grid resolution B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 25 Point Lagrangian coupling iii) VOF to LPP conversion - High Reynolds – CORIA (Berlemont) jet time B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 26 Atomization of a plane sheared liquid film • Legi experiment (Descamps, Matas & Cartellier) • Reδ = 1000, density ratio 100 • Simulation 64 x 256 x 512 B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 27 Results B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 28 PDF of droplet diameters (a) t = 0.4s, (b) t = 0.41s, and (c) t = 0.42s. B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 29 Conclusions • Improvements of DNS codes AMR : Archer and DyJeAt codes Parallel Euler-lagrange coupling : Parissimulator and DyJeAt codes Mass conservation improvement : Archer, DyJeAt, Paris codes Momentum conservation : Archer, DyJeAt codes IBM : Parissimulator, Archer codes • First cross comparisons with experiments : see WP 4.1.3 and WP 4.1.4 This project has received funding from the European Union’s Seventh Framework Programme for research, technological development and demonstration under grant agreement no 265848. B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK 30 Next B1-15 FIRST Project Spray and Soot Workshop, 19-20 November 2014, Derby, UK