Three-dimensional numerical simulations of flapping wings at low Reynolds numbers

OpenFOAM Workshop, Zagreb Frank Bos, Bas van Oudheusden, Hester Bijl 7-9 June 2007
1/22

Department of Aerospace Engineering
Delft University of Technology

Overview
1. Introduction 2. Background of insect aerodynamics 3. Problem and requirements 4. Equations and Methods 5. Fluent simulations, 2D 6. OpenFOAM simulations

• • •

2D validation 3D flapping wings Fluid Structure Interaction

7. Conclusions and (near) future work

Overview

Modelling

Fluent

OpenFoam 2D

OpenFoam 3D

OpenFoam FSI

Conclusions

2/22

Department of Aerospace Engineering

MicroBat (Darpa).Insect flight for MAV design Insects are interesting to develop Micro Aerial Vehicles Examples of application: 1. MFI (Berkeley) Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 3/22 Department of Aerospace Engineering . Espionage 3. Intelligence 2. Investigation of hazardous environments DelFly (TuDelft): Possible MAV designs: Left to right: DelFly (TuDelft).

Thomas. R.P.Background of insect flight • • • • • Flow dominated by low Reynolds numbers. Re = U L/ν. 303-316. enhancing lift! Limited predicted capabilities of quasi steady theories. Ellington and A.mech. B (1997) 352.P.html Source: A. Experiments or numerical simulations to study vortex interactions. Willmott. Phil. Trans. lond.R. Highly viscous and unsteady flow. C. Flapping motion leads to vortex generation. Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 4/22 Department of Aerospace Engineering .ac. Flow visualisation and unsteady aerodynamics in the flight of the hawkmoth.L.kogakuin. Dragonfly experiment: Hawkmoth experiment: http://fluid.jp/ iida/mav/dragonfly. Soc.

(Multiple) wing motion with large amplitudes. Re=O(100 − 1000). Hovering and forward flight.Problem Statement Problem definition: • • Performance in insect flight is strongly influenced by 3D wing motion. Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 5/22 Department of Aerospace Engineering . Fluid Structure Interaction / Body Dynamics coupling. For MAV design the forces on flapping wings need to be understood. CFD model requirements: • • • • • 3D flapping wings (6 DOFs). Low Reynolds number.

Governing equations and flow solvers Incompressible Navier-Stokes equations: ∂u ∂t 1 + (u · ∇) u = − ρ ∇p + ν∇2 u Moving mesh using Arbitrary Lagrangrian Eulerian (ALE) formulation: u → u − umesh Finite volume based general purpose CFD solvers: • • Fluent v6.3/1.4 → 2D/3D simulations and FSI Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 6/22 Department of Aerospace Engineering .2 → 2D simulations OpenFOAM 1.

P. exp..Insect wing motion modelling: 3D → 2D 3D parameters: • • • • • • Amplitude: φ Angle of attack: α Deviation: θ Start of upstroke q(t) f(t) Mean stroke plane a(t) Start of downstroke 2D parameters: Amplitude: x = Rg φ/c Angle of attack: α Deviation: y = Rg θ/c Rg = radius of gyration. and Dickinson M. J. The control of flight force by a flapping wing: lift and drag production. c = averaged chord Source: Sane S.H. Biol. 2607-2626 (2001) Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 7/22 Department of Aerospace Engineering . 204.

• • • Overview Moving inner mesh (not possible in 3D) Re-meshing far away Timestep restrictions Modelling Fluent OpenFoam 2D • • • Multiblock mesh (GridPro) Mesh deformation in OpenFoam Limited maximal motion amplitudes OpenFoam FSI Conclusions 8/22 OpenFoam 3D Department of Aerospace Engineering .Grid generation: 2D and 3D 2 different ways of grid generarion.

2D Fluent: Vorticity contours • • Vorticity is used to give lift and drag a physical meaning: ω = ∂Uy ∂x − ∂Ux . Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 9/22 Department of Aerospace Engineering . ∂y The flow fields generated by the different kinematic models are compared.

2 0. Numerical Study of kinematic wing models of hovering insect flight. Bos. D. AOA is responsible for higher performance. Bijl. H.8 1 • • • AOA variation leads to an increase in early angle of attack. Lentink. AIAA 2007-482. [Ref] F.5 t/T [−] 0. low drag.2D Fluent: Influence of angle of attack harmonic model: ’bump’ angle of attack: 6 4 CL [−] 2 0 -2 -4 -6 0 0. van Oudheusden. Wing orientation → high lift. M. B. Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 10/22 Department of Aerospace Engineering .

Significant performance differences observed for different kinematic models. 2.Conclusions about 2D simulations 1. Fruit fly kinematics reduces drag considerably. increases performance and leads to a more balanced flight. Simplified kinematics increases drag. 4. Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 11/22 Department of Aerospace Engineering . • • The validity of these 2D results depends on 3D effects → 3D simulations using OpenFOAM! Wing flexibility may also play an important role → Fluid Structure Interaction using OpenFOAM. 3. Present simulations give good insight in the effect of kinematic motion modelling.

2D is OK. Parallel implementation (3D meshes require much computational effort!). Grid deformation. We simply don’t have enough Fluent licenses for large parallel calculations. OpenFoam is open-source such that user defined library development is easier. 2. Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 12/22 Department of Aerospace Engineering . ⇒ Before going to 3D some 2D validation results will be shown. while preserving high quality. rotation and translation. Difficult to maintain high quality moving 3D meshes. Why not use Fluent? 1. 3. but 3D is a problem.3D approach using OpenFOAM Solver Requirements: • • • • 3D flapping → 6 degrees of flapping. Wing flexibility → Fluid Structure Interaction (FSI).

1829 3 Lit.044 0.1845 1. Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 13/22 Overview Department of Aerospace Engineering .3 Validation: 2D cylinders.337 0.050 0.OpenFOAM 1. 1.333 1.7.3 CD CD pressure CL viscous f CPU time (days) 1.290 0. Re=150 OF 1.000 cells.1834 Plunging cylinder: • • • • • • Central scheme in space and Crank-Nicholson scheme in time. OpenFOAM results within 1% compared to Fluent and literature! OpenFOAM twice as fast. Moving cylinders still need some validation.289 0.5 Fluent 1. Grid size = 100. ∆t = period/800 → Comax ≈ 0.339 1.

periodical vortical patterns (compared to Fluent).3: 2D flapping airfoil.OpenFOAM 1. Euler scheme in time. ∆t = period/5000 → Comax ≈ 0. Possible less diffusion (compared to Fluent) causes more complex wake! Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 14/22 Overview Department of Aerospace Engineering . Re=110 • • • • • Central scheme in space.25! Mesh motion: quadratic diffusion. Very small timestep required. but still skew cells at wing leading edges. No regular.

tensor RCur = RstrokePlane & RxCur & RyCur & RzCur.value(). Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 15/22 Department of Aerospace Engineering .localPoints().deltaT().6 Degrees of Freedom Library Library development to define 6 DOFs body motion: 1 2 3 4 5 6 7 8 9 10 const pointField& oldPoints = mSolver.boundary()[bodyPatchID_]. motionUBodyPatch == (dispRot+dispTrans)/time().tetMesh().oldPoints. vectorField dispRot = (((RCur) & inv(ROld)) & (oldPoints . tensor ROld = RstrokePlane & RxOld & RyOld & RzOld.curRotationOrigin)) + curRotationOrigin .

backward in time. Force history shows good trend! Movie! Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 16/22 Overview Department of Aerospace Engineering .∆t = period/5000 → Comax ≈ 0.000 cells and domain far too small.OpenFOAM 1.25! Relative coarse grid of 100. Although mesh is coarse.3: 3D flapping wing (1/2) 3D Grid Z-Vorticity • • • • • Central scheme in space. the flow field seems OK. Initial mesh quality influence maximal wing displacements/rotations.

Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 17/22 Overview Department of Aerospace Engineering .OpenFOAM 1. the 3D flow effects are already visible! Mesh refinement leads to extremely small timestep: 1) make use of subsets 2) user-defined diffusion for mesh motion.3: 3D flapping wing (2/2) X-Vorticity Pressure Isosurface • • • • X-Vorticity shows 3D tip vortices. Although the mesh is too coarse. Pressure contours revealing 3D structure of the leading edge vortex.

3: Fluid Structure Interaction Setup: • • • • • • • icoFsiFoam with central scheme in space. Euler scheme in time. Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 18/22 Overview Department of Aerospace Engineering . To do: 1) Fix ’movingWallVelocity’ instability.OpenFOAM 1. No realiable simulation yet :-( Divergence with ’movingWallVelocity’ BC → unstable oscillations due to oscillatory corrections. It’s hard to find a proper testcase. Low density ratio (< 1000) causes divergence. 2) Maybe subiterating because of strong coupling. First testcase: Cylinder with flexible plate.

5 CPU time.U: new iterative solvers OpenFoam FSI Conclusions 19/22 Department of Aerospace Engineering .4 3 OF 1. p (s) CPU time. Pressure and Total Drag Coefficients [−] Total Drag 2.5 3 0 0 0.4 times faster Big gain due to new mesh solver Gain in p.5 Viscous Drag CPU time.4 is 4.3 OF 1.3 versus 1.4 2. Total (s) 2 2. mesh (s) 1 Pressure Drag 0.5 time / period [−] • • Overview Solutions not yet periodic → Maybe domain too small! More diffusion in 1.5 1 1.3 and 1.OpenFOAM 1.3 CD CPU time.3.4: Plunging cylinders Force comparison OpenFoam 1.1 41 354 203 687 2 1.5 OF 1. Movie! Modelling Fluent OpenFoam 2D • • • OpenFoam 3D OF 1. U (s) 91 450 2464 3018 OF 1.4 FontSize Viscous.

OpenFoam is very accurate. since the ’movingWallVelocity’ BC causes divergence. 4. 2D cylinder flow • • • • • • • • • Overview 2D cylinder flow. Necessary timestep is too small. which depends on mesh motion. Divergence at density ratios < 1000. which depends on mesh motion. Flow seems to have less diffusion compared to Fluent. 3. Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 20/22 Department of Aerospace Engineering . OpenFoam is fast for static and moving cylinders. 2D flapping wing More validation is needed. 3D flapping wing Even on a coarse grid. the flow seems plausible. 2. Fluid Structure Interaction No physical solution. Necessary timestep is too small.Conclusions 1.

only deformation far away! Obtain a working FSI solver with a working ’movingWallVelocity’ boundary condition. but needs tuning! Important topics to focus: • • Improve initial mesh quality.. Using subsets. Anyone? :-) Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 21/22 Department of Aerospace Engineering ....? Overall. OpenFoam is very well suited for our problem.What’s next . Improve 2D/3D mesh motion by: • User-defined mesh motion diffusion..

Thanks for your attention! Questions or Remarks? Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 22/22 Department of Aerospace Engineering .

which is reflected in lift-to-drag ratio! Models 1 and 2: decrease in performance in terms of lift-to-drag ratio of −25% to −31% Models 3 and 4: ’bump’ increases performance considerably. 25%! Deviation influeces force variation! Overview Modelling Fluent OpenFoam 2D OpenFoam 3D OpenFoam FSI Conclusions 23/22 Department of Aerospace Engineering .2D Fluent: Influence of wing kinematics Model 1) effect of ’Sawtooth’ amplitude 2) effect of ’Trapezoidal’ α 3) effect of ’bump’ in α 4) effect of deviation CL −8% −9% 0% −11% CD +22% +37% −13% −3% CL /CD −25% −31% +25% −8% • • • • Lift is comparable due to scaling. Mean drag is strongly affected.