Hybrid Simulation of Wake Vortices during Landing HPCN-Workshop 2014 A. Stephan 1, F. Holzäpfel 1, T. Heel 1 1 Institut für Physik der Atmosphäre, DLR, Oberpfaffenhofen, Germany Aircraft wake vortices Motivation Simulation methods Code requirements Post-processing Applications Scaling, bottlenecks
Wake vortex roll-up circulation: Mg Γ = ρ (π / 4) B u
Folie 3 Wake vortex in reality turbulent vortex with laminar vortex core
Folie 4 Interests related to WV: Safety & efficiency 70s hazardous tests 70s separation rule Source: NASA Increase number of flights while satisfying safety criteria
Prepared for future challenges? 12% unaccommodated demands in most-likely future scenario airport capacity challenge as strong as ever bottleneck aircraft separations during landing Eurocontrol, Challenges of Growth 2013, Summary Report
Why approach and landing? highest risk to encounter WV in ground proximity: physics: WV cannot descend below glide path rebound lidar, LES: WV may live much longer than 2 min (5 NM) NATS incident reporting: most encounters in ground proximity possibilities to recover limited by low altitude potentially critical situation bottleneck aircraft separation What happens during landing? Can we actively promote WV decay in ground proximity?
Aircraft wake during landing http://www.youtube.com/watch?v=ppuftg_mxg8 - Embraer 170 - Condensation trails appearing at deployed flaps - End effects after touch down
3 landing phases Approach wake initiation phase Touch down instantaneous lift reduction Vortex decay interaction with environment
Ground effect with crosswind
10 Resolution requirements
Numerical approach LES code MGLET Finite-Volume compact scheme Fourth-order accuracy Velocity-pressure iteration method Multi-grid convergence acceleration Third-order Runge-Kutta time integration Lagrangian dynamic sub-grid scale model Grötzbach-Schumann wall model for ground surface Obstacles modeled as a strong drag force Highly complex flows around aircraft use RANS/LES coupling Equidistant Cartesian mesh, 300-500 million nodes Parallel, max. 2048 processors
Bridging near-field and far-field Sweep a RANS flow field through a ground-fixed LES domain Switching function y f(y,α,β) LES RANS (Misaka et al., AIAA Paper 2011-1003)
Approach Sweep an existing RANS flow field through a ground-fixed LES domain The RANS flow field is used as a forcing term in the LES (Fortified Solution Algorithm, Nudging technique) [ 1 f ( y, α, ] RANS V = f ( y, α, β ) V LES + β ) V 1 y β f ( y, α, β ) = tanh α + 1. 0 2 β y α, β : parameters y f(y,α,β) LES RANS
AWIATOR long range aircraft model TAU simulation for ONERA catapult wind tunnel experiments Flow conditions: U = 25 m/s, Re = 5.2x10 5, C L = 1.4 (high lift configuration) RANS mesh Reference quantities Γ 0 = 5.36 m 2 /s b0 = 1.756 m w 0 = 0.49 m/s t 0 = 2.0 s
Simulation steps 1. RANS Reynolds-averaged Navier-Stokes equations steady approach unstructured grid DLR TAU-RANS solver 2. LES 2.1 Turbulence initialization 2.2 Wake-vortex initialization 2.3 Wake-vortex evolution
Aircraft descent every step is interpolated for Δx = Δy = Δz an angle of 3.57 degree realized by 32 RANS fields RANS flow field removed as a forcing term at the instant of touch down
Resolution requirements 1. RANS grid adaptation number of grid points depending on geometric complexity flaps, slats, engine pylon, wheels, etc vortex cores have to be resolved 2. LES long flight path determines x-dimension crosswind drift determines y-dimension Reynolds number determines mesh stretching at the wall Dimensions: 2048x768x256
Application: Wake vortex decay enhancement no obstacle obstacle, square profile 20 s Stephan, A., Holzäpfel, 28 F.,Misaka, s T. (2013) Aircraft wake vortex decay in ground proximity Physical mechanisms and artificial enhancement Journal of aircraft 36 s 44 s
Key mechanisms 2. Ω shape causes self-induced fast approach to primary vortex (PV) 3. after SV has looped around PV it separates and travels along the PV (again driven by self induction)
Decay triggered by plate line (z 0 = b 0 ) -plate dimensions for e.g. A340: -length height: 9 m 4,5 m -spacing: 20 m
Arrangement of plates at the runway (Munich airport)
Application: Simulation of WakeOP Cases flight experiment with HALO at airport Oberpfaffenhofen Lidar measurements from Falcon hangar smoke visualization documented by video and photo purpose: demonstrate functionality of plate line patent DE 10 2011 010 147 74 overflights at airport Oberpfaffenhofen flight altitude b 0 22 m above ground high-lift configuration, landing gear deployed weak wind & turbulence, poor visibility
23 Vortex Evolution 1
24 Vortex Evolution 2
Application: Aircraft Landing vorticity iso-surface vorticity layer at the ground end effects after touch down 80 frames, 1024x368x128, 200MB
400 frames, 1024x384x128, 1TB tracer initialized at high vorticity level tracer velocity color coded Landing animation Autodesk -Sdf
-Sdf Landing animation Autodesk
t* = -0.15 t* = 0.1 t* = 0.5 t* = 1.0
Processing Online Processing (parallel) turbulence statistics e.g. wind profiles not time consuming Post Processing (one processor for entire domain) vorticity computation structure tracking vortex cores and circulation computation time depending on grid size
Circulation decay with end effect w/o plate line with plate line End effects reduce vortex strength Circulation decays to ~ 65% Plate line reduces vortex strength Circulation decays to ~ 40%
Scaling, Bottlenecks, Outlook RANS/LES coupling currently 2048 procs used additional memory requirements for RANS fields, depending on descend angle LES MGLET is developing since 1980s was not intended for high processor numbers (up to 128) large communication costs memory scales with number of processors running with 8192 processors, 2 Billion grid points, in pure mode strong scaling has negative performance for more than 2048 procs. Outlook three dimensional structures getting into focus one line of research: increasing number of grid point
Conclusion hybrid RANS/LES approach is used to simulate aircraft wake vortices LES code MGLET (TU Munich) is used method effectively tackles landing simulations, plate lines etc. 2048 processors can be used for hybrid approach 8192 processors can be used for pure LES approach post processing needs parallelization fancy animations require large memory remote rendering would be nice scaling bottleneck: bad code performance