Computational Technology Laboratory, Report KTH-CTL-4023

Simulation of 3D unsteady incompressible ow past a NACA 0012 wing section

Johan Hoffmana , Johan Janssona,, Niclas Janssona
a

Computational Technology Laboratory, CSC/NA, KTH, Royal Institute of Technology, SE-100 44 Stockholm, Sweden

Abstract We present computational simulations of three-dimensional unsteady high Reynolds number incompressible ow past a NACA 0012 wing prole, for a range of angles of attack, from low lift through stall. A stabilized nite element method is used, referred to as General Galerkin (G2), with adaptive mesh renement with respect to the error in target output, such as aerodynamic forces. Computational predictions of aerodynamic forces are validated against experimental data.

1. Introduction In this work we present computational simulations of a NACA 0012 wing prole, in three-dimensional unsteady high Reynolds number incompressible turbulent ow for a range of angles of attack , from low lift through stall. It is still a challenge to accurately and efciently compute aerodynamic forces (lift and drag) and predict stall of a wing at realistic ight conditions [1, 2]. The ability to perform such simulations at an affordable cost could greatly enhance the virtual design process of for example, airplanes, wind turbines and propellers. Direct numerical simulation (DNS) of turbulent ow is limited to moderate Reynolds numbers and simple geometry, due to the high computational cost of resolving all turbulent scales in the ow. The standard e.g. in the automotive industry is simulation based on Reynolds averaged Navier-Stokes equations (RANS), where time averages (or statistical averages) are computed to an affordable cost, with the drawback of introducing turbulence models based on parameters that have to be tuned for particular applications. An alternative to DNS and RANS is Large eddy simulation (LES) [3], where only the largest scales of the ow are resolved, combined with subgrid models to take into consideration the effect of the smallest unresolved turbulent scales. Subgrid models are either explicit, based on physics theory or experiments, or implicit through the numerical discretization of the equations. The simulation method in this paper is an adaptive nite element method, referred to as a General Galerkin (G2) method [4], where the residual based stabilization acts as a subgrid model similar to an Implicit LES (ILES). An adaptive algorithm is used, driven by a posteriori estimation of the error in chosen target output based on the solution of an adjoint problem. An efcient open source implementation (also suitable for high performance computer architectures) is available in the form of the computational software Unicorn [5], and the G2 simulation method has been validated for a number of turbulent ow benchmark problems [6, 7, 8, 9, 10, 11, 12]. At the high Reynolds numbers considered in this paper, boundary layers are turbulent and for small angles of attack the ow is attached until the separation at the rear of the wing, with small drag and high lift. Under increasing angles of attack the ow stays attached with a corresponding increase in lift and drag, until stall is reached where the ow separation moves upstream which results in a decrease in lift and a dramatic increase in drag. To model the effect of the turbulent boundary layer at very high Reynolds numbers, corresponding to realistic ight conditions, a slip with friction boundary condition is used [4]. We here approximate the small skin friction by zero skin friction, corresponding to a free slip boundary condition, an approach that previously has been validated in [10, 11]. The outline of this paper is the following:
Corresponding

author Email addresses: jhoffman@kth.se (Johan Hoffman), jjan@kth.se (Johan Jansson), njansson@kth.se (Niclas Jansson)

Computational Technology Laboratory, Report KTH-CTL-4023 Method: Here we describe the adaptive G2 method applied to uid ow and its implementation in Unicorn. Results: Here we present computed aerodynamic quantities for different angles of attack, compare to experimental data and show mesh convergence sequences.

=4

=4

=10

=10

=12

=12

=17

=17

=20

=20

Figure 1: Magnitude of the velocity on the surface of the airfoil together with isosurfaces left and surface pressure with isosurfaces right.

Computational Technology Laboratory, Report KTH-CTL-4023 2. Method 2.1. Fluid model As a basic model we use the incompressible Navier-Stokes equations for velocity and pressure, given as: t u + (u )u + p = g + 2 (u) u = 0 (, 0) = u 0 u with = /0 the kinematic viscosity. 2.2. G2 nite element method Our computational approach is based on stabilized nite element methods, together with adjoint based adaptive algorithms, and residual based implicit turbulence modeling, for related work see e.g. [13, 14, 15]. The General Galerkin (G2) method for high Reynolds number ow, including turbulent ow and shocks, takes the form of a standard Galerkin nite element discretization together with residual based numerical stabilization of the residual. We split the time interval I into subintervals In = (tn1 , tn ), with associated space-time slabs Sn = In , over which we dene space-time nite element spaces, based on a spatial nite element space W n dened for a spatial mesh Tn [4]. = (U, P ) which is continuous piecewise linear In a cG(1)cG(1) method [16] we seek an approximate solution U n n in space and time. With W a standard nite element space of continuous piecewise linear functions, and W0 the functions in W n which are zero on the boundary , the cG(1)cG(1) method for constant density incompressible ow with homogeneous Dirichlet boundary conditions for the velocity takes the form: for n = 1, ..., N , nd (U n , P n ) n 3 (U (tn ), P (tn )) with U n V0n [W0 ] and P n W n , such that
1 n )U n , v ) + (2 (U n ), (v )) ((U n U n1 )kn + (U n , q) (P n , v ) + ( U

in Q, in Q, in .

(1)

(2)

n n SD (U , P n ; v, q )

= (f, v ) , v = (v, q )

V0n

n = 1/2(U n + U n1 ) is piecewise constant in time over In , with the stabilizing term where U
n n SD (U , P n ; v, q ) n U n + P n f ) , U n v + q ) (1 (U

n , v ), +(2 U where we have dropped the shock capturing term, and where (v, w) =
K Tn 3 K

v w dx,

( (v ), (w)) = with the stabilization parameters

(
i,j =1

ij (v ), ij (w )),

2 2 1/2 1 = 1 (kn + |U n1 |2 h n )

2 = 2 hn where 1 and 2 are positive constants of unit size. For turbulent ow we choose a time step size kn min(hn /|U n1 |).
x

We note that the least squares stabilization omits the time derivative in the residual, which is a consequence of the test functions being piecewise constant in time for a cG(1) discretization of time [17]. 3

Computational Technology Laboratory, Report KTH-CTL-4023 2.3. Boundary conditions We model the turbulent boundary layer by a slip with friction boundary condition [4] which can be written: u n = 0, u k +
1 T

n k = 0,

k = 1, 2,

with n an outward unit normal vector, and k orthogonal unit tangent vectors at the boundary. Here can be chosen as a constant parameter, or as a function of space and time, similar to simple wall shear stress models [18, 19] For very high Reynolds numbers we nd that = 0 is a good approximation for small skin friction stress, which we have validated for a number of benchmark problems [10, 11]. We implement the normal component condition (slip) boundary condition in strong form [5], where the boundary condition is applied after assembling the left-hand side matrix and the right-hand side vector modifying the algebraic system, whereas the tangential components (friction) are implemented in weak form by adding boundary integrals in the variational formulation. The node normals are computed from a weighted average of the surrounding facet normals. Edges and corners are identied from the angles between facet normals, for which the velocity is constrained in 2 and 3 linearly independent directions respectively. Care needs to be taken when constucting the mesh near rounded surfaces of sharp radius, to avoid articially constraining velocities. This is particularly true for the trailing edge in the present problem, where we introduce a rounding by limiting the radius of curvature to 0.006m (the chord length is 0.734m). 2.4. Adaptive mesh renement The adaptive mesh renement algorithm is based on a posteriori estimation of the error in a chosen target output. In this paper the target output was chosen as the aerodynamic force. A posteriori error estimates for G2 are available from our previous work [6, 7, 8, 9, 20], and is based on the general framework for a posteriori error estimation for nite element methods developed in the 1990s, see e.g. [21, 22, 23]. Starting from an initial coarse mesh T 0 , one simple such algorithm implemented in Unicorn takes the form: let k = 0 then do Algorithm 1. Adaptive mesh renement 1. 2. 3. 4. For the mesh T k : compute the primal problem and the dual problem. Compute the error indicators as a product of the gradient of the dual solution and the residual. Mark some chosen percentage of the elements with highest error indicators for renement. Generate the rened mesh T k+1 , set k = k + 1, and goto 1.

In Fig.2.4 we illustrate the adaptive mesh renement for 4 angles of attack, which shows the difference in mesh renement for attached and separated ow at stall. Note that for angles = 4 , = 10 and = 12 the mesh is rened upstream, explained by the behavior of the dual velocity in gure 4, as well as renement occuring close to the leading and trailing edges. For = 17 almost all the renement occurs downstream in the wake behind the airfoil. 2.5. Unicorn implementation Unicorn is solver technology (models, methods, algorithms and software implementations) with the goal of automated simulation of realistic continuum mechanics applications, such as drag or lift computation for xed or exible objects (uid-structure interaction) in turbulent incompressible or compressible ow (airplane or bird ight, car aerodynamics). The Unicorn software system is a free software/open source implementation as part of the FEniCS [24, 25, 5] software project for automation of computational modeling, with aspects such as abstraction of low-level parallel nite element assembly functions through a high-level description in mathematical language and abstraction of 2D/3D-specic implementation. Unicorn implements the G2 method described above through a description of the variational form in a Python high-level mathematical notation, which is then compiled into low-level nite element assembly C++ source code. 4

Computational Technology Laboratory, Report KTH-CTL-4023

=4, iter=0

=4, iter=7

=10, iter=0

=10, iter=7

=12, iter=0

=12, iter=6

=17, iter=0

=17, iter=6

=20, iter=0

=20, iter=7

Figure 2: Illustration of the adaptive mesh renement, with annotation of angle of attack and adaptive iteration number. Note the renement upstream for = 4, 10, 12 (before stall), compared to = 17, 20 (after stall) where the renement only occurs downstream in the wake.

3. Results 3.1. Overview Our focus in this paper is the prediction of aerodynamic forces. For a set of angles of attack we perform G2 simulations over the time interval I = [0, 10], where we for each angle of attack adaptively rene the computational mesh based on a posteriori error estimation. We compare against experimental results given by Ladson in [26]. We present visualizations and plots of the results organized as follows: 5

Computational Technology Laboratory, Report KTH-CTL-4023

Lift vs angle of attack

1.5 1.2 0.9 0.6 0.3 4

Unicorn Ladson 1 Ladson 2 Ladson 3

8 12 16 angle of attack 20 24 28

C_L

Drag vs angle of attack

0.25 0.20 0.15 0.10 0.05 0.004

Unicorn Ladson 2 Ladson 3 Ladson 4

8 12 16 angle of attack 20 24 28

Figure 3: Computed drag and lift coefcients plotted against the angle of attack, together with experimental results from Ladson [26].

Time-series of cD and cL We plot the drag and lift coefcients over time for the nal three adaptively rened meshes in gure 7. Mesh convergence of cD and cL We plot the mean value over time of the drag and lift coefcients as functions of adaptive iteration number to show the convergence behavior in gure 6. cD and cL against angle of attack We plot the drag and lift coefcients against the angle of attack, and include experimental results from Ladson in gure 3. Visualization of the velocity eld We visualize the velocity eld for all angles of attack as a color plot of magnitude of the velocity on the surface of the airfoil, together with isosurfaces for specic values to show the 3D eld in gure 1. In gure 4 we zoom in on the trailing edge. Visualization of the pressure eld We visualize the velocity eld for all angles of attack as a color plot of the pressure on the surface of the airfoil, together with isosurfaces for specic values to show the 3D eld in gure 1. Visualization of the dual velocity We visualize the dual velocity, which acts as a weight on the residual in the adaptive mesh renement algorithm. We plot isosurfaces for specic values to show the 3D eld in gure 4 6

C_D

Computational Technology Laboratory, Report KTH-CTL-4023 3.2. Flow separation and stall For the angles = 4 , = 10 the ow is fully attached, with separation occuring at the trailing edge. At = 17 , the separation has moved upstream in the form of streamwise vortices attached to the upper wing surface, see gure 1. A similar separation pattern is also observed in experiments [27, 28]. For = 17 and = 20 we observe stall, in the form of a decrease in lift and a dramatic increase in drag. 3.3. Aerodynamic forces We perform simulations for a range of angles of attack (4, 10, 12, 17, 20) where we expect to go from low lift through a linear growth of the lift coefcient and growth of the drag coefcient to stall, where lift decreases and drag increases dramatically. For every angle we let the adaptive algorithm run to produce a sequence of meshes and outputs. The experimental values given here are for: Ladson 1 Re = 6.00e6, M = 0.15 Grit level 60W (wraparound). Ladson 2 Re = 5.95e6, M = 0.15 Grit level 80. Ladson 3 Re = 6.00e6, M = 0.15 Grit level 120. where the W signies a wraparound grit strip around the leading edge, where the other cases only have a thin grit strip which does not wrap around. The experiment can be divided into two regimes: pre-stall and post-stall. In the pre-stall regime different experimental setups give similar drag and lift coefcients for a given angle of attack. This means we can quantitavely compare against the experimental values to validate our simulation methods. In the post-stall regime however, the variation of the lift and drag coefcients for a given angle is large (partly because the stall angle is different), and this does not give a meaningful quantitative validation. Here we compare the post-stall lift (which appears to be roughly similar between the experiments) and check if get a similar low lift over drag in the simulations and experiments. We tabulate the values for = 4 , 10 , 12 and for post-stall in tables 1-4 for the Unicorn computation and the three Ladson experiments. Lift coefcient Ladson experiments The experimental lift values show a linear growth of lift from 4 until stall which appears between 14 and 18 , where the lift decreases again. They vary approximately 5% from setup to setup in the pre-stall regime. In the post-stall regime there is a decrease in lift to 1.05 20%. Lift coefcient Unicorn computations The Unicorn simulation results similarly show a linear growth of lift from 4 to 12 within 10% of the experiments and shows stall at 17 and 20 , with a decrease in lift. The convergence tolerances are 5% for all ve angles. Drag coefcient Ladson experiments The experimental drag values show a growth of drag from 4 until stall which appears between 14 and 18 , where the drag increases dramatically. They vary approximately 30% from setup to setup for 4 and 50% for 10 . For post-stall, the lift over drag ratio has decreased to below 5 for all experiments before or at 20 . Drag coefcient Unicorn computations From the Unicorn simulation results we can see a matching growth of drag from 4 (20% tolerance, within 25% of experiments) to 12 (30% tolerance, within 25% of experiments). For the 17 and 20 angles the computed drag value also increases dramatically, consistent with stall. For post-stall, the lift over drag ratio has decreased below 5 as in the experiments before or at 20 .. 7

Computational Technology Laboratory, Report KTH-CTL-4023

Table 1: Angle of attack: = 4 . Tabulated values of drag and lift coefcients.

Source ( = 4 ) Unicorn Ladson 1 Ladson 2 Ladson 3

cD 0.00668 20% 0.00975 0.00823 0.00876

cL 0.452 5% 0.423 0.425 0.428

cL /cD 67.7 43.3 51.6 47.9

Table 2: Angle of attack: = 10 . Tabulated values of drag and lift coefcients.

Source ( = 10 ) Unicorn Ladson 1 Ladson 2 Ladson 3

cD 0.0144 30% 0.0190 0.0119 0.0117

cL 1.11 5% 1.01 1.05 1.07

cL /cD 77.1 53.2 88.2 91.4

Table 3: Angle of attack: = 12 . Tabulated values of drag and lift coefcients.

Source ( = 12 ) Unicorn Ladson 1 Ladson 2 Ladson 3

cD 0.0213 25% 0.0253 0.0132 0.0128

cL 1.24 5% 1.16 1.25 1.26

cL /cD 58.2 45.9 94.7 98.4

Table 4: Post-stall. Tabulated values of drag and lift coefcients.

Source Unicorn ( = 17 ) Unicorn ( = 20 ) Ladson 1 ( = 16 ) Ladson 2 ( = 18.5 ) Ladson 3 ( = 18.5 ) 4. Discusson

cD 0.0790 10% 0.230 10% 0.247 0.226 0.260

cL 1.09 5% 0.998 5% 0.893 1.06 1.19

cL /cD 13.8 4.34 3.62 4.7 4.6

We have presented computational simulations of the ow past a NACA 0012 wing section, for a range of angles of attack. We nd that the G2 simulations correctly capture the basic ow features found in experiments, and that quantitative predictions of lift and drag in the pre-stall regime are validated in experiments. The simulations also capture the stall phenomenon with dramatically reduced lift over drag ratio. In future work we will extend our computational studies to capture the maximum lift value and stall angle, and to other airfoil sections, including 3D sections, and we will also include uid-structure effects available in Unicorn [29]. Acknowledgments The authors would like to acknowledge the nancial support from the European Research Council, Swedish Foundation for Strategic Research and the Swedish Research Council. The simulations were performed on resources provided by the Swedish National Infrastructure for Computing (SNIC) at National Supercomputer Centre in Sweden (NSC) and PDC - Center for High-Performance Computing. References
[1] P. Moin, J. Kim, Tackling turbulence with supercomputers, Scientic American.

Computational Technology Laboratory, Report KTH-CTL-4023

=4

=4

=10

=10

=12

=12

=17

=17

Figure 4: Zoom in on surface velocity on trailing edge showing zig-zag pattern left and isosurface of dual velocity right, acting as a weight on the residual in the adaptive renement.

[2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]

P. Moin, D. You, Active control of ow separation over an airfoil using synthetic jets, Journal of Fluids and Structures 24. P. Sagaut, Large Eddy Simulation for Incompressible Flows (3rd Ed.), Springer-Verlag, Berlin, Heidelberg, New York, 2005. J. Hoffman, C. Johnson, Computational Turbulent Incompressible Flow, Vol. 4 of Applied Mathematics: Body and Soul, Springer, 2007. J. Hoffman, J. Jansson, C. Degirmenci, N. Jansson, M. Nazarov, Unicorn: A Unied Continuum Mechanics Solver, in: Automated Scientic Computing, Springer, 2011. J. Hoffman, Computation of mean drag for bluff body problems using adaptive dns/les, SIAM J. Sci. Comput. 27(1) (2005) 184207. J. Hoffman, C. Johnson, A new approach to computational turbulence modeling, Comput. Methods Appl. Mech. Engrg. 195 (2006) 2865 2880. J. Hoffman, Adaptive simulation of the subcritical ow past a sphere, J. Fluid Mech. 568 (2006) 7788. J. Hoffman, Efcient computation of mean drag for the subcritical ow past a circular cylinder using general Galerkin G2, Int. J. Numer. Meth. Fl. 59 (11) (2009) 12411258. J. Hoffman, N. Jansson, A computational study of turbulent ow separation for a circular cylinder using skin friction boundary conditions, in: Quality and Reliability of Large-Eddy Simulations II, Vol. 16 of ERCOFTAC Series, Springer Netherlands, 2011, pp. 5768. R. V. de Abreu, N. Jansson, J. Hoffman, Adaptive computation of aeroacoustic sources for rudimentary landing gear, in: proceedings for Benchmark problems for Airframe Noise Computations I, Stockholm, 2010. N. Jansson, J. Hoffman, M. Nazarov, Adaptive Simulation of Turbulent Flow Past a Full Car Model, in: Proceedings of the 2011 ACM/IEEE International Conference for High Performance Computing, Networking, Storage and Analysis, SC 11, 2011. Y. Bazilevs, V. Calo, J. Cottrell, T. Hughes, A. Reali, G. Scovazzi, Variational multiscale residual-based turbulence modeling for large eddy

Computational Technology Laboratory, Report KTH-CTL-4023

=4

=17

Figure 5: Visualization using the ParaView Surface Line Integral Convolution lter of the time-averaged velocity over the time interval, illustrating the direction of ow on the surface of the airfoil, similar to an oil-lm visualization technique [30]

simulation of incompressible ows, Comput. Meth. Appl. Mech. Eng. 197(1-4) (2007) 173201. [14] O. Guasch, R. Codina, A heuristic argument for the sole use of numerical stabilization with no physical les modeling in the simulation of incompressible turbulent ows, Preprint Universitat Politecnica de Catalunya. [15] J. L. Guermond, R. Pasquetti, B. Popov, From suitable weak solutions to entropy viscosity, J. Scientic Comput. 49 (2011) 3550. [16] P. Hansbo, A crank-nicolson type space-time nite element method for computing on moving meshes, J. Comput. Phys. 159 (2000) 274289. [17] J. Hoffman, J. Jansson, R. V. de Abreu, Adaptive modeling of turbulent ow with residual based turbulent kinetic energy dissipation, Comput. Meth. Appl. Mech. Eng. 200 (37-40) (2011) 2758 2767. [18] U. Piomelli, E. Balaras, Wall-layer models for large-eddy simulation, Annu. Rev. Fluid Mech. 34 (2002) 349374. [19] U. Schumann, Subgrid scale model for nite difference simulations of turbulent ows in plane channels and annuli, J. Comput. Phys. 18 (4) (1975) 376 404. [20] N. Degirmenci, J. Hoffman, J. Jansson, An adaptive nite element method for unied continuum uid-structure interaction, Tech. Rep. KTH-CTL-4020, Computational Technology Laboratory, http://www.publ.kth.se/trita/ctl-4/020/ (2011). [21] K. Eriksson, D. Estep, P. Hansbo, C. Johnson, Introduction to adaptive methods for differential equations, Acta Numer. 4 (1995) 105158. [22] R. Becker, R. Rannacher, A posteriori error estimation in nite element methods, Acta Numer. 10 (2001) 1103. [23] M. Giles, E. S uli, Adjoint methods for pdes: a posteriori error analysis and postprocessing by duality, Acta Numer. 11 (2002) 145236. [24] FEniCS, FEniCS project, http://www.fenicsproject.org (2003). [25] J. Hoffman, J. Jansson, M. Nazarov, N. Jansson, Unicorn, http://launchpad.net/unicorn/hpc (2011). [26] C. Ladson, U. S. N. Aeronautics, S. A. Scientic, T. I. Division, Effects of independent variation of Mach and Reynolds numbers on the

10

Computational Technology Laboratory, Report KTH-CTL-4023

low-speed aerodynamic characteristics of the NACA 0012 airfoil section, NASA technical memorandum, National Aeronautics and Space Administration, Scientic and Technical Information Division, 1988. N. Gregory, C. OReilly, A. R. C. G. Britain), Low-speed aerodynamic characteristics of NACA 0012 aerofoil section, including the effects of upper-surface roughness simulating hoar frost, Reports and memoranda, H.M.S.O., 1973. B. G olling, Experimentelle Untersuchungen des laminar-turbulenten Ueberganges der Zylindergrenzschichtstr omung, DLR, 2001. J. Hoffman, J. Jansson, M. St ockli, Unied continuum modeling of uid-structure interaction, Math. Mod. Meth. Appl. S. G unter, Schewe, Reynolds-number effects in ow around more-or-less bluff bodies, Journal of Wind Engineering and Industrial Aerodynamics 89 (14-15) (2001) 1267 1289.

[27] [28] [29] [30]

11

Computational Technology Laboratory, Report KTH-CTL-4023

0.50 0.48 C_L 0.46 0.44 0.42 0.40 0.020 0.016 C_D 0.012 0.008 0.004 0.000 300000 300000

4 degrees angle of attack Unicorn adaptive mesh refinement < 5% tolerance

C_L

1.28 1.20 1.12 1.04 0.96 500000 0.05 0.04 C_D 0.03 0.02 0.01 0.00 500000

10 degrees angle of attack Unicorn adaptive mesh refinement < 5% tolerance

600000

900000 1200000 No. mesh points

1500000

1800000

1000000 1500000 No. mesh points

2000000

2500000

Unicorn adaptive mesh refinement < 20% tolerance

Unicorn adaptive mesh refinement < 30% tolerance

600000

900000 1200000 No. mesh points

1500000

1800000

1000000 1500000 No. mesh points

2000000

2500000

1.6 1.4 C_L 1.2 1.0 0.8 400000 0.10 0.08 C_D 0.06 0.04 0.02 0.00 400000

12 degrees angle of attack Unicorn adaptive mesh refinement < 5% tolerance

C_L

1.28 1.20 1.12 1.04 0.96 200000 0.20 0.16 C_D 0.12 0.08 0.04 0.00 200000

17 degrees angle of attack Unicorn adaptive mesh refinement < 5% tolerance

800000

1200000 1600000 No. mesh points

2000000

400000

600000 800000 No. mesh points

1000000

Unicorn adaptive mesh refinement < 25% tolerance

Unicorn adaptive mesh refinement < 10% tolerance

800000

1200000 1600000 No. mesh points

2000000

400000

600000 800000 No. mesh points

1000000

1.28 1.20 C_L 1.12 1.04 0.96 400000 0.30 0.27 C_D 0.24 0.21 0.18 0.15 400000

20 degrees angle of attack Unicorn adaptive mesh refinement < 5% tolerance

800000

1200000 1600000 No. mesh points

2000000

Unicorn adaptive mesh refinement < 10% tolerance

800000

1200000 1600000 No. mesh points

2000000

Figure 6: Mesh convergence plot of computed drag and lift coefcients for the range of angles of attack plotted against the adaptive mesh renement iteration number.

12

Computational Technology Laboratory, Report KTH-CTL-4023

0.50 0.48 0.46 0.44 0.42 0.400

Lift coefficient at 4 degrees angle of attack iter4 iter5 iter6

1.15 1.10 1.05 1.00 0.95 0.900 0.05 0.04 0.03 0.02 0.01 0.000

Lift coefficient at 10 degrees angle of attack iter4 iter5 iter6

C_L

C_L

time (s)

10

time (s)

10

0.020 0.015 C_D 0.010 0.005 0.0000

Drag coefficient at 4 degrees angle of attack iter4 iter5 iter6

Drag coefficient at 10 degrees angle of attack iter4 iter5 iter6

C_D

time (s)

10

time (s)

10

1.8 1.7 1.6 1.5 1.4 1.3 1.2 1.1 1.0 0.90

Lift coefficient at 12 degrees angle of attack iter4 iter5 iter6

time (s)

10

1.5 1.4 1.3 1.2 1.1 1.0 0.90

Lift coefficient at 17 degrees angle of attack iter3 iter4 iter5

C_L

C_L

time (s)

10

0.20 0.15 C_D 0.10 0.05 0.000

Drag coefficient at 12 degrees angle of attack iter4 iter5 iter6

0.20 0.15 C_D 0.10 0.05 0.000

Drag coefficient at 17 degrees angle of attack iter3 iter4 iter5

time (s)

10

time (s)

10

1.8 1.6 1.4 1.2 1.0 0.8 0.60

Lift coefficient at 20 degrees angle of attack iter4 iter5 iter6

C_L

time (s)

10

Drag coefficient at 20 degrees angle of attack

0.6 0.5 0.4 0.3 0.2 0.1 0.00

C_D

iter4 iter5 iter6

time (s)

10

Figure 7: Computed drag and lift coefcients for the range of angles of attack plotted against time for the nal three mesh iterations in the adaptive renement.

13