Best Practices for Aerospace Aerodynamics. Peter Ewing

Transcription

1 Best Practices for Aerospace Aerodynamics Peter Ewing

2 Agenda Pre-processing Geometry Origin/Import Geometry Prep Surface Mesh Volume Mesh Solver Settings Defining Flight Physics Setting Up Solvers Post-processing Automated Data Extraction Plotting Scenes Automated Reporting

3 Agenda Pre-processing Geometry Origin/Import

4 STAR-CCM+ Parts Geometries ultimately conglomerate in Parts Laser scans, extracted mesh topology External CAD modelers, e.g. CATIA, NX STAR-CCM+ 3D-CAD Mesh Operation Parts Common Denominator: tessellated surfaces STL or surface meshes dummy or flattened surface meshes Discrete Mesh Operations Detached mesh operations are green 3D-CAD/CAD Parts Analytic representation, blue or solid grey User should be aware of geometry quality Especially for flattened Parts! STAR-CCM+ requires clean, closed geometry: To use Boolean operations To generate a volume mesh Import Prep Surface Volume

5 CAD is Preferred Aero surfaces & leading edges are complex swept geometries These features matter! Hierarchy of geometry fidelity : STAR-CCM+ 3D-CAD CAD-Clients CAD Exchange X_B /X_T then STP/STEP DBS, STL, IGES Direct Link SolidWorks Parameter Transfer CAD geometry allows several benefits over flattened parts Project to CAD CAD-based Mesh Operations Feature aligned meshing Parametric design changes 3D-CAD and CAD-Clients Persistent Part naming STAR-CCM+ Bi-directional link Import Prep Surface Volume

6 Agenda Pre-processing Geometry Origin/Import Geometry Prep

7 External Aerodynamics Geometry Preparation Split the body into multiple Part Surfaces: Inflow/Outflow/Freestream definitions Allows tracking of physical convergence Trailing Edge for custom controls Rounded edges DPW4 Geometry (upper) Naming conventions enable filtering and efficient identification, e.g.: 00 Inlet, 00 Outlet, 00 Freestream, etc. 01 Wing, 01 Body, 01 Tail, etc. 02 Symmetry Plane 3D RAE2822 Airfoil for 2D simulation 03 Interface (Sliding or Overset) DPW4 Geometry (lower) Filter selection box Import Prep Surface Volume

8 Low-Speed Far-field Boundary Preparation Velocity Inlet Atmospheric flight: Upstream boundary: Typically velocity inlet in a round/bullet shape Distance is characteristic lengths Outflow boundary: Typically a outflow flat plane cut Distance is characteristic lengths Pressure Outflow Example bullet domain Wind tunnel configurations should be matched: Duplicate the geometry Inlet distances typically set as free stream * Outlet distance should follow free stream distance Side walls typically set to symmetry * * If inlet conditions are well measured, duplicate Import Prep Surface Volume

9 Transonic Far-field Boundary Preparation Freestream settings: Circular domain will use Freestream boundary condition Upstream position characteristic length scales Downstream position characteristic length scales Freestream Boundary Body Sample transonic circular domain Wind tunnel sections can be difficult to reproduce Transonic wind tunnels typically have slatted configurations Simulations may contain shock reflections to disrupt upstream flow Unless specific configuration is well documented, run in Freestream Import Prep Surface Volume

10 Supersonic and Hypersonic Far-field Boundary Preparation Upstream placed fairly close and aligned with shocks generated by the body The shock should not interact with the freestream boundary Outlet boundaries can either be Pressure Outlet or Freestream Hypersonic cases Outlet can be set to Pressure field function to extrapolate Pressure Outlet Freestream Axis or Symmetry Body Example of hypersonic domain for Mach 12 sphere Import Prep Surface Volume

11 Wrapping What does it do? Enables fast turn-around of broken geometry Standard use case is for unification of assemblies of broken (i.e. not clean and closed) Parts How do I know if I should wrap? Inefficient control over the CAD or Parts are flattened Extensive* Surface Repair work is required: Inefficient (or no) control of CAD workflow Many CAD based-errors (e.g. too many pieces) to fix efficiently in CAD Too many tessellation errors to efficiently fix in Surface Repair Simulation fidelity is independent of intricate details affected by Wrapper Features worth investigating: Works well in the PBM structure Maintains Part Surface naming convention Operation can be Detached to create new Part Partial Wrapping Speeds up the wrapping process Project to CAD Used by permission: Sikorsky / American Helicopter Society Import Prep Surface Volume

12 STAR-CCM+ Surface Repair Comments What does it do? Checks triangulations for valid clean/closed geometry Manipulate underlying triangulations (tessellations) How do I know if I should Surface Repair? The underlying Part is not clean/closed manifold There is no control of the CAD to fix within CAD If a Part Requires Repair: Don t panic! Undo/Forward-do buttons Surface Repair can repair the parts: Up-to-date guide flags remaining fixes Create new Part Surfaces where needed Create new Part Curves where needed Keep in Mind: It s like sewing up a bundle of triangles: Connect dots, zip edges Goal is to create a manifold, air-tight surface Import Prep Surface Volume

13 Agenda Pre-processing Geometry Origin/Import Geometry Prep Surface Mesh

14 Automated Surface Mesher Settings Automatic Surface Repair Model: Off Default settings Surface Remesher Settings: Increase minimum face quality to 0.20 Curvature=76 Surface mesher settings: Base Size to Characteristic Length/10, e.g.: Chord length/10 Characteristic Body length/10 Surface Curvature: Surface Growth Rate: Custom Surface Controls: Edge proximity on bodies to 3 Lifting Surfaces: Basic Curvature to 76 Growth rate to Target Size: Chord/100 Trailing Edges: Minimum Target Size to ¼ of t.e. thickness Inlet/Outlet/Freestream/Symmetry Boundaries: Target Surface Size to be at least characteristic length Chord/100 Proximity NACA0010 Growth = 1.05 Import Prep Surface Volume

15 Agenda Pre-processing Geometry Origin/Import Geometry Prep Surface Mesh Volume Mesh

16 Quasi-2D Core Volume Mesh Models 2D Automated Meshing (PBM): Requires an initial 3D body 2D section lies on z-axis Does not need to be CAD Applications: Airfoil analyses Test mesh settings Testing of physics settings Supersonic 2D/Axisymmetric NLF-0416 Directed Mesher (PBM): Ordered style grids High quality grids for supersonic flows Best practice topology for hypersonic cases Requires an initial 3D CAD body Workflow tip: Split patches in the CAD-Client or in 3D-CAD On Geometry transfer, choose All CAD Edges option Choose to Initialize Patches by CAD Edge Allows for macro automation 2D Axisymmetric Hypersonic bi-conic Import Prep Surface Volume

17 Core Volume Mesh Models Trimmer or Polyhedral are both acceptable topologies Refinement in flow regions of interest are key to capturing flow features in the simulation Polyhedral mesh: Aerospace cases mesh in serial Pseudo-random orientation of faces reduces numerical dissipation Smooth growth away from bodies Optimizer can increase mesh quality Prefer to control mesh based solely on remeshed surface Volume controls to catch the hard spots Trimmer mesh model: Massively parallel Faster, requires less memory Aligning the trimmer mesh model to the main flow directions can reduce numerical dissipation Mesh refinement/coarsening in factors of 2 Use of volume control to control location of transitions Lockheed Martin Public Release: ORL Import Prep Surface Volume

18 Polyhedral meshing for Aerospace Polyhedral Mesher Settings Growth Rate: Can be off or on Reduces cell count between geometry gaps Optimization Cycles Increase Optimization cycles to 1-4 Effective in aiding Adjoint case convergence Polyhedral Controls If Volume Growth Rate On Volume Growth Rate to 1.2 Maximum cell size to characteristic length Mesh Density Leave at defaults If a volume control exists in the mesh Volumetric Control Blending to 0.5 Growth Rate On Off Import Prep Surface Volume

19 Trimmer Mesher Settings Trimmer Mesh Model Settings Typically left at defaults Mesh in parallel Typical control settings Volume Growth Rate Slow to Very Slow Maximum Cell Size to characteristic length Maximum Core/Prism Transition Ratio Anywhere between 2-5 Import Prep Surface Volume

20 Prism Layer Mesher Model Settings: Stretching function: Hyperbolic Tangent Stretching Mode: Wall Thickness Minimum Thickness Percentage: 0.01 Layer Reduction Percentage: 0.0 Make conformal prisms in all layers Near Core Layer Aspect Ratio: =<1.0 Typically set to 1.0 or 0.75 Requires two inputs: Wall Thickness Prism Layer Total Height Translation: Wall Thickness = a low y+ mesh or high y+ mesh Prism Layer Total Height = Boundary Layer Thickness RAE2822 Airfoil HLPW4 Import Prep Surface Volume

21 High y+ mesh vs. Low y+ mesh High y+ mesh notes: Sub layer and buffer region is modelled by one grid cell Wall y+ value should be > 30 Wall y+ value < Typically has 8-14 prism layers Implicitly assumes that the boundary layer is turbulent and will try to reproduce the log layer behavior Low y+ mesh notes: Attempt to integrate/resolve entire boundary layer Wall y+ value should be ~< 1 Values << 1.0 will not improve results Should not be > 5.0 Has at least 10 prisms in y+ < 30 region Typically prism layers Flows that are not modelled with a transition model should not be taken as predictive transition modelling Explicitly model the trip on tripped boundary layers U Viscous sublayer Buffer-layer Log-layer Defect-layer Y+ Low y+: First grid point High y+: First grid point Import Prep Surface Volume

22 Prism Layer Techniques for Trailing Edges (or Hypersonic Leading Edges) Knife-edges lead to cells with high skewness angles High skewness angles create numerical instabilities Counter is to create refinements on the edge O-Grid, Retract O-Grid, No Retract Custom Surface Settings on Trailing Edges Create models with finite trailing edges Use Prism Layer Thickness Reduction Avoids prism layer collapse on trailing edges Avoids oddly shaped cells in the rear TE Custom Settings Import Prep Surface Volume

23 Automated Mesh Refinement STAR-CCM+ can perform automated mesh refinement : Table based refinement Custom Field Function metric for refinement Tabulate cell size refinement metric Solution based refinement Initial JAVA macro-driven volumetric control Conceptually, flow field contains arbitrary cells that contain refinement metrics Threshold Derived Parts are exported as STL files STL files can be wrapped to create volumetric control Remesh 1 New Feature in Adjoint based mesh refinement JAVA macro can drive adjoint-based mesh refinement Original Remesh 2 Refined Blunt Nose*; Mach 6.8; AoA 20 *Courtesy Lockheed Martin Missiles & Fire Control AGARD RAE 2822 Adjoint Refinement Physics Solvers

24 Agenda Pre-processing Geometry Origin/Import Geometry Prep Surface Mesh Volume Mesh Solver Settings Defining Flight Physics

25 Turbulence: RANS RANS Reynolds Averaged Navier Stokes Most common choice for external aerodynamics Robust, well studied Steady state simulations: 2D, Axisymmetric, 3D Obtains the average of all resolved flow features Extra equations add a turbulent viscosity to the dynamic viscosity in the Navier-Stokes Equations HLPW4 is the turbulence model of choice Enables use of, transition model Does not preclude the use of and its variants All y+ wall model is the preferred choice Boundary conditions: Typically left as default, but can use measured values Decay of inflow turbulent quantities can be mitigated by activating the Ambient Source Term (ASM) Do not use with the transition model Solver settings: Not uncommon to increase Turbulent Viscosity Limiter, e.g.: 1e8 Physics NLF-0416 Solvers

26 Unsteady Turbulence: URANS vs DES vs LES URANS Unsteady Reynolds Averaged Navier Stokes Run in 2D, Axisymmetric, 3D Adds unsteady term to the RANS equations Common choice for rotor performance Sliding mesh setup About 2 degrees per time step If nothing dynamically changes about the geometric configuration during the simulation, risks reverting to RANS Rotor wake from a ROBIN body DES Detached Eddy Simulation Legitimate in 3D simulations, always unsteady Popular choice for performance simulations Not prohibitively more expensive than 3D URANS IDDES = Improved Delayed DES default mode modern method Blend of RANS and Large Eddy Simulation RANS near-wall, LES everywhere else Far less turbulent viscosity in the LES regions TLG: DES Buffeting analysis 2012 STAR Global Conference Talk Physics Solvers

27 Turbulence: LES and Laminar LES Large Eddy Simulation Legitimate in 3D simulations, always unsteady Not particularly popular choice in external aero More expensive than RANS and DES High mesh counts required near walls Needed to properly resolve structures of transitioning flows Laminar Navier Stokes Equations solved directly without any turbulence model Low-speed to supersonic simulations will not likely use this Hypersonic simulations that are not interested in boundary layer will choose in conjunction with a high y+ (>100) mesh 2012 STAR Korean Conference: Satish Kumar B. et al. Transition flow and aero-acoustic analysis of NACA0018 ALM: Bow shocks on re-entry of the Crew Exploration Vehicle Physics Solvers

28 What can you get in a 2D vs. 3D simulation? 2D Simulation Enables: Fast testing for unknown physics phenomena Shock position for grid refinement Solver settings Simulations for 3D axisymmetric shapes RANS/URANS turbulence modelling Transition location using, _ Onset of trailing edge stall 3D Simulation Enables: RANS, URANS, DES, LES Complex geometry interactions Stall Prediction Physics Solvers

29 Unsteady Time Stepping Key Idea: Simulating a continuously transient behavior in a discrete fashion U(t) Time (t) T/20 is a good start 2015: STAR Global : Overset with Zero Gap Demo Physics Solvers

30 Agenda Pre-processing Geometry Origin/Import Geometry Prep Surface Mesh Volume Mesh Solver Settings Defining Flight Physics Setting Up Solvers

31 Physics Continuum Solver Choice Low Speed: P weak function of ρ, T High Speed: P strong function of ρ, T Segregated Solver SIMPLE Continuity and momentum yield a pressure-correction equation Mildly compressible flows, but not appropriate for shock capturing Flow regimes: Incompressible Low speed High speed, subsonic; Mach < ~0.5 Consider local flow Mach numbers! Lower memory requirements, faster than Coupled solver Coupled Solver Continuity, momentum, energy are solved simultaneously Equation of state yields pressure Flow regimes: Incompressible Low speed High speed, subsonic; Mach < ~0.5 All other flow speeds for Mach > ~0.5 Designed for hyperbolic nature of equations and shocks Higher memory requirements Physics Solvers

32 Solver Settings: Incompressible to Ma<0.5 If Using the Segregated Solver Simulations that use this solver should start with good initial conditions Constant velocity in the direction of the flow Smoothly ramping velocity from wall using field function Constant temperature set to flow conditions Turbulent quantities are typically default URFs are typically not ramped Rotor cases typically ramp or step RPM Unsteady simulation initialization Begin from steady state RANS solution Turn on Unsteady Solver If Using the Coupled Solver Roe FDS Initial condition: Constant velocity in direction of flow Constant temperature set to flow conditions CFL: Grid Sequencing Initialization On Expert Driver On Physics Solvers

33 Solver Settings: Transonic to Supersonic Coupled Solver Suggested Settings: Transonic (0.5 < Ma < 1.0): Roe FDS if no local Ma > 1.0 CFL from 5.0 to 50.0 Supersonic (1.0 < Ma < 4.0): AUSM+ CFL from 5.0 to 20.0, 20/Ma Grid Sequencing Initialization Turn on Expert Driver to On If no Expert Driver: Ramp CFL from 1 to 1000 Ratio of CFL Number : Explicit relaxation factor = 3:1 Physics Solvers

34 Solver Settings: Hypersonic (Ma > 4) Implicit Coupled Solver Suggested Settings: Typical CFL ~ Grid Sequencing Initialization Expert Driver Ramp CFL from 1 to 1000 CCA Turned On If no Expert Driver: Ramp CFL from 1 to 1000 Ratio of CFL Number : Explicit relaxation factor = 3:1 Mach 6.77 blunt cone: NASA TN-D1606 Physics Continuum Settings: AUSM+ May choose gradient reconstruction value between 1.0 and 2.0 Sometimes an almost 2 nd order will converge Mach 11.3 Blunt Biconic Mach 16 Shock-shock interaction on a cylinder Physics Solvers

35 Grid Sequencing Initialization (GSI) GSI Input: Physics Continuum Initial Conditions How it Works: Runs the Euler equations on successively refined series of grids, from coarse grid to finest (real) grid Wrapped Rocket GSI Initial After Condition GSI 1000 Iterations Result: Field starts at near-flight conditions Recommended Settings: Sweeps per grid level = 200 Tolerance = Notes No reason for special velocity initial conditions Develops preliminary shock locations Physics Solvers

36 Continuity Convergence Accelerator When: High Mach number aerodynamic cases CCA Input: Updated Coupled Solver flow field How it works: Solves an elliptic equation for pressure corrections Updates the cell pressures (w/underrelaxation) Corrects the face mass fluxes and cell velocities Updates density, total enthalpy, etc. appropriately Continuity Convergence Acceleration of a Density-Based Coupled Algorithm, Caraeni et al., AIAA Fluid Dynamics Conference, June 2013, San Diego, CA With CCA Without CCA Results: Can result in faster convergence for stiff problems Mixed high Mach and low Mach numbers Internal compressible flows Temperature dependent properties Settings: URF typically set Physics Solvers

37 Agenda Pre-processing Geometry Origin/Import Geometry Prep Surface Mesh Volume Mesh Solver Settings Defining Flight Physics Setting Up Solvers Post-processing Automated Data Extraction Plotting Scenes Automated Reporting

38 Interpreting the Residuals Mission statement: Simulations should be as accurate as possible. Residual values are a global metric of convergence Local convergence may get lost when only using residual values Residuals are used as a metric to judge overall quality of the simulation Used in both steady and unsteady simulations Example Residual Plots: Steady Unsteady Post

39 Checking Convergence with Engineering Criteria Both Steady and Unsteady Simulations Create Plots of Quantitative Data Skin Friction Coefficient Mass imbalance (especially for high speed flows) Lift Drag Moments Plot versus inner iteration, make sure metrics asymptotically converge onto a value For steady simulations, asymptotic behavior For unsteady simulations, asymptotic behavior within the prescribed time step s iterations Post

40 Optimate & External Aero Requires additional planning up front Testing CAD robustness Post-processing change in data sets Use JAVA to drive changes Benefits: Automated sweeps 3D-CAD parameterization CAD-client bi-directional capability Fire-and-forget Reduces burden on heavy scripting Small pieces of JAVA can be inserted into process Rotating the coordinate systems Visualization of large data sets Post-processing is collected in single tool Visualize multi-variable interactions Post

41 Post-Process Interactively on a Cluster Common practice to post-/troubleshoot on special big-memory machines External aerodynamics cases can be more than 20M cells Difficult to run multiple iterations for troubleshooting purposes Download, identify mesh issues, remesh, re-submit to queue, crash, re-download, make rhetorical statement: There s got to be another way. STAR-CCM+ client-server architecture Data is post-processed by parallel cores Visualization on workstation graphics Benefits: Increased framerates Volume rendering Line Integral Convolutions 28.8M cell DPW4 model on 4 64 cores Client location: Los Angeles, CA Server location: Detroit, MI Post

42 Agenda Pre-processing Geometry Origin/Import Geometry Prep Surface Mesh Volume Mesh Solver Settings Defining Flight Physics Setting Up Solvers Post-processing Automated Data Extraction Plotting Scenes Automated Reporting

43 Thank You Time for any questions