TIME-ACCURATE SIMULATION OF THE FLOW AROUND THE COMPLETE BO105 WIND TUNNEL MODEL Walid Khier, Thorsten Schwarz, Jochen Raddatz presented by Andreas Schütte DLR, Institute of Aerodynamics and Flow Technology Folie 1 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Outline Motivation Aerodynamics of the helicopter Flow solver Wind tunnel experiment Results Code performance Conclusion Other applications Folie 2 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Motivation h Demonstration of the capability of DLR s block structured flow solver FLOWer to simulate the flow around a complete helicopter h FLOWer is already validated for fixed wing applications and for isolated helicopter fuselages and rotors h Difficulty: complex geometry and unsteady flow h Work is part of the French-German CHANCE project Partners: Eurocopter, ONERA, IAG (Uni Stuttgart), DLR Folie 3 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Aerodynamics of the helicopter - a challenge for CFD solvers htransonic flow hdynamic stall hblade vortex interaction Folie 4 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Aerodynamics of the helicopter - a challenge for CFD solvers htransonic flow hdynamic stall hblade vortex interaction htail rotor interactions with rotor and fuselage hrotor fuselage interactions hflow separation at bluff bodies Phenomena affect: loads performance vibration noise Folie 5 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
DLR flow solver FLOWer (1) hfinite volume discretization of RANS equations on structured, multi block grids hspace discretization - cell centered or cell vertex discretization - central scheme with scalar dissipation or various upwind schemes htime discretization - flow equations: explicit multi-stage schemes (Runge-Kutta) with multigrid acceleration - turbulence equations: explicit multi-stage scheme or implicit DDADImethod hturbulence modeling: various 0-, 1-, 2-, 7-equation models, e.g. Spalart-Almaras, kω, kω-sst, EARSM, RSM hshape optimization by inverse design option or adjoint method Folie 6 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
DLR flow solver FLOWer (2) Numerics for unsteady computations himplicit time integration with dual-time stepping hoverlapping grid technique (Chimera) hmoving / deforming meshes High performance computing hparallelization based on MPI hoptimized for vector computers Folie 7 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Overlapping grid technique (Chimera) Folie 8 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Wind tunnel experiment (1) h BO105 wind tunnel model h experimental data were obtained during the HELINOVI campaign at the DNW in 2003 h inflow data: α fuselage = -5.2 M = 0.1766 M MR = 0.652 M TR = 0.63 Θ MR = 10.5-6.3 sin(ψ) + 1.9 cos(ψ) Θ TR = 8.0 Folie 9 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Wind tunnel experiment (2) wind tunnel model CFD model Folie 10 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Near field grids individual overlapping grids for hfuselage hspoiler and strut hskids hhorizontal stabilizer hmain rotor and tail rotor Folie 11 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Far field grid Number of blocks and grid cells hfuselage+spoiler+ stabilizer+skids+strut 48 blocks, 6.0 M cells hmain rotor 4*3 blocks, 4*0.8 M cells htail rotor 2*3 blocks, 2*0.3 M cells hbackground grid 414 blocks, 1.9 M cells htotal 480 blocks, 11.8 M cells Component grids are embedded in Cartesian background grid with hanging nodes Folie 12 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Unsteady flow computation Parameters of the computations: h Central discretization with scalar dissipation (JST-scheme) h Flow variables located at cell centers h CFL = 10.0, 3 level multigrid h k-ω turbulence model h time integration with dual time stepping - 50 inner iterations - one physical time step equals a 2 rotation of the tail rotor - one physical time step equals a 0.4 rotation of the main rotor h computation required four weeks using eight processors of NEC SX6 h 2.3 revolutions of main rotor were computed h more than 400 GB data produced Folie 13 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Unsteady pressure distribution Folie 14 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Variation of pressure distribution Folie 15 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Vortex cores (λ 2 -criterion) Folie 16 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Vorticity normal to symmetry plane Folie 17 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Vorticity component normal to plane through skids Folie 18 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Pressure distribution in symmetry plane Folie 19 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Variation of pressure on tail fin black: experiment, red: cfd Folie 20 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Pressure distribution on main rotor r/r = 87% Folie 21 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Pressure distribution on tail rotor r/r = 80% Folie 22 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Code performance analysis t = t + Δt t = 0 h time integration with dual time stepping method, within one physical time step execute: 1. move grids to new positions 2. cut holes and search for donor cells for interpolation (Chimera) 3. perform 50 iterations to converge the implicit time integration h separate performance analysis for Chimera and one inner iteration position grids Chimera hole cutting search 50 x one inner iteration of dual-time stepping one physical time step t = t end Folie 23 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Performance improvement of Chimera algorithms Execution time for Chimera (hole cutting and search procedure) and flow solver on eight Processors of NEC SX 6 Chimera flow solver one time step starting point on NEC SX 6 750 s 50 * 9.3 s 1215 s improved state on NEC SX 6 69 s 50 * 9.3 s 534 s Early tests on NEC SX 8 48 s 50 * 5.5 s 323 s Expected on NEC SX8 13 s 50 * 3.3 s 180 s Improvement of chimera performance Factor of ten Folie 24 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Parallel performance on NEC SX6 Chimera hole cutting and search procedure flow computation (time for one inner iteration) t (physical time step) = t (Chimera) + 50 * t (one inner iteration) seq: 3037 s, 8 proc: 532 s Folie 25 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Conclusion hsimulation of complete helicopter wind tunnel model successful hcomputation took four weeks on eight processors of NEC SX6 hpostprocessing of CFD results is very time consuming due to time dependent flow and large amount of data (0.4 TB) hunsteady pressure distributions and vortices in flow field analyzed hgood agreement for fuselage and main rotor, differences for tail rotor to be clarified hexperimental data not optimum for code validation, many uncertainties hexecution time per physical time step halved by optimizing Chimera algorithms hfurther improvement of vectorization and parallelization and use of NEC SX8 will increase execution speed by factor 3 Folie 26 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Other applications for HPC (1) EU-Project TILTAERO hnumerical and experimental investigation of tiltrotor configurations hrequires similar capabilities of flow solver as for BO105 hcalculations performed on NEC SX8 Folie 27 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte
Other applications for HPC (2) EU-Project GOAHEAD hproject lead by DLR-AS (Dr. K. Pahlke) hwind tunnel experiments with generic configuration in order to create a CFD validation database for helicopters hnumerical flow simulations including - elastic blade deformation by fluid-structure coupling - coupling with flight mechanics code to compute trim of helicopter - a converged solution will require approximately 15 revolutions of the main rotor strong need for high performance computers Folie 28 > HLRS 2005 > W. Khier, T. Scwarz, J. Raddatz, A. Schütte