AIAA Paper 2004-1056 42nd AIAA Aerospace Sciences Meeting and Exhibit, Reno, Nevada 5 - 8 Jan 2004
A NUMERICAL STUDY OF TRANSONIC BUFFET ON A SUPERCRITICAL AIRFOIL
Q. Xiao1 and H. M. Tsai2 Temasek Laboratories National University of Singapore Singapore 119260
F. Liu3 Department of Mechanical and Aerospace Engineering University of California, Irvine CA 92697-3975
Abstract Self-excited oscillation has been investigated both experimentally and computational since the early work 1 The steady/unsteady flow of the BGK No. 1 of McDevitt et al. (e.g., Refs.1-18) Previous supercritical airfoil is investigated by the solution of the experimental and numerical studies of transonic flow Reynolds-Averaged Navier-Stokes equations with a over an 18% thick circular-arc airfoil have indicated two-equation k-ω turbulence model. The dual-time that, the flow over a circular-arc aerofoil exhibit varied stepping method is used to march in time. Two steady behavior depending on the flow conditions. Three distinct regions have been observed for a fixed free- cases ( Ma = 0.71,α =1.396 and Ma = 0.71,α = 9.0 ) stream Reynolds number but varying Mach number. and one unsteady case ( Ma = 0.71,α = 6.97 ), both Below a critical Mach number, the flow is steady and with a far-stream Re = 20x106 , are computed. The characterized by a weak shock wave near the mid-chord results are compared with experimental data obtained with trailing-edge flow separation. For larger Mach by Lee et al. The mechanism of the self-excited numbers, shock induced separation is encountered and oscillation for supercritical airfoils is explored. The the flow becomes unsteady, with unsteady shock oscillation period calculated based on the mechanism motions on the upper and lower surface that are out of proposed by Lee is consistent with the Fourier analysis phase with each other. As the Mach number is further of the computed lift coefficient. increased, a steady shock reappears. It is sufficiently strong to induce flow separation.
Introduction Compared to the non-lifting circular-arc airfoil, the study of other lifting supercritical airfoils is less Transonic buffeting flow appears in many aeronautical matured. The computation and experiment of applications ranging from internal flows such as around NASASC(2)-0714 have been reported in the paper of 7 8 turbo-machinery blades to flows over aircraft. Unsteady Bartels and Bartels et al. . An extensively set of data shock/boundary-layer interaction affects aerodynamic for the buffeting problem studied at the National performance. For flows around an airfoil self-excited Research Council, Canada of the flow over the Bauer- shock oscillations can cause flow separation and Garabedian-Korn (BGK) No. 1 supercritical airfoil at 9 10 increase drag significantly. The capability to accurately transonic speed has been reported by Lee , Lee, et al. , 11,12 13 predict such phenomena is of technological significance Lee , and Lee, et al. . The experiments were and a challenge. The accuracy of computation is conducted over a range of Mach number and angle of influenced by both the accuracy of the numerical attack. A map of the shock induced trailing edge technique and the accuracy of the physically model, separation regions together with steady and unsteady particularly, the turbulence model. pressure measurements were reported for various shock/boundary-layer interactions. The investigation of the supercritical airfoil indicates, different from the circular-arc airfoil, the supercritical airfoil has a much ––––––––––––––––––––––– weaker trailing edge separation in the onset region and 1. Research Scientist, email: [email protected] experience shock oscillation only on one side of the 2. Principal Research Scientist, email: [email protected] Member airfoil. AIAA. 3. Professor, email: [email protected] Associate Fellow AIAA. Copyright © 2004 by the authors. Published by the American The mechanism of the self-excited oscillation is not Institute of Aeronautics and Astronautics, Inc with permission well understood although some explorations have been done by several researchers. (Refs. 11, 13-18). The investigation on the bi-convex airfoil suggested that the
1 American Institute of Aeronautics and Astronautics transonic periodic flows are initiated by an asymmetric Sofialidis and Prinos25 were studied with limited unsteady disturbance. The shock induced separation success. changes the effective geometry of the airfoil, which would cause the forward and rearward movement of One difficulty in most turbulence models is the inability shock depending on whether the streamtube decreases to account directly for non-equilibrium effects found in or increases. The necessary but not sufficient condition separated flows. A new class of model was proposed by for the periodic flow to appear is the shock wave strong Olsen & Coakley26, which is termed the lag model. The enough to cause boundary separation. The Mach basic idea of the lag model is to take a baseline two- number just upstream of the shock should be in the equation model and couple it with a third (lag) equation range of 1.14< M1 <1.24. For the supercritical airfoil to model the non-equilibrium effects for the eddy (Lee11), a model to predict the shock motion was viscosity. Recently, Xiao et al.27 incorporated the lag formulated. The model assumed that the movement of model with the baseline k-ω model and computed both shock leads to the formation of pressure waves which steady and unsteady transonic nozzle flows. Their propagate downstream in the separated flow region. On computations show notable improvements for strong reaching the trailing edge the disturbance generates shock cases, where strong non-equilibrium effect is upstream-moving waves, which interact with the shock present. For separated flows, good turbulence modeling wave and impart energy to maintain the limit cycle. The is necessary to compute the overall flow physics. The period of the oscillation should be the time for a lagged k-ω model also enjoys the advantage of not disturbance to propagate from shock to the trailing edge requiring specification of wall distance, and has similar plus the duration for an upstream moving wave to reach or reduced computational effort compared with the shock from the trailing edge via the region outside Reynolds stress models. the separated flow. In the present work we employ the lag model coupled A number of difficulties are particular to the simulation with the two-equation k-ω model for the study of of unsteady flow. Dual-time stepping method 19 transonic buffeting flow for the BGK No.1 supercritical introduced by Jameson provides an alternative to such airfoil, which is well documented in the paper of Lee9, direct computation of flow involving oscillation. This Lee, et al.10, and Lee11,12. The computational results will method uses an implicit physical time discretization. At be compared with the experiments for the each physical time step, the equations are integrated in steady/unsteady flow. The exploration of mechanism a fictitious pseudo-time to obtain the solution to the for the self-sustained oscillation will be conducted from steady state in pseudo-time. When marching in pseudo- the computational flow field. To the author’s time, this method permits the use of acceleration knowledge, no computational work has been done for techniques such as local time step and multigrid to this supercritical airfoil. speed up the steady flows calculations. The application of dual-time stepping method of self-excited flow 3 5 In the next section, the numerical implementation of the includes Rumsey et al. and Oliver et al. for the 18% dual-time stepping method for the two-dimensional circular-arc airfoil. Navier-stokes equations will be briefly presented. The computational results and discussions will follow. The assessment of turbulence models in unsteady aerodynamic flow featuring buffet and/or dynamic-stall has attracted less attention than the steady flow. In the early work of Levy2, computations were done using the Numerical Method Baldwin-Lomax20 model which showed the unsteadiness in the flow field. It is sufficiently strong to The governing equations for the unsteady compressible induce flow separation. In terms of accuracy, the turbulent flow are expressed as follows: algebraic turbulence models such as the Baldwin- Mass conservation: ∂ρ ∂ Lomax model do not perform well (Girodroux-Lavigne + (ρu ) = 0 (1) 14 6 ∂ ∂ j and LeBalleur ). Barakos & Drikakis more recently t x j presented an assessment of more advanced turbulence Momentum conservation: closures in transonic buffeting flows over airfoils. ∂ ∂ ∂ ∂τˆ ρ + ρ = − p + ji (2) Various turbulence closures such as algebraic model, ( ui ) ( u jui ) 21 ∂t ∂x ∂x ∂x the one-equation model of Spalart and Allmaras , the j i j Launder-Sharma22 and Nagano-Kim23 linear k- ε models as well as the non-linear eddy-viscosity model of Craft et al. 24 and the non-linear k-ω model of
2 American Institute of Aeronautics and Astronautics Mean energy conservation: ∂ ∂ Where Pr and Pr are the laminar and turbulent (ρE) + (ρu H ) = L T ∂t ∂x j j (3) Prandtl numbers, respectively. The other coefficients are: ∂ ⎡ ∂k ⎤ ⎢uτˆ + (µ +σ *µ ) − q ⎥ a = 0.35, R = 1, R = 0.01, ε = 5/ 9 , ε * = 1, ∂ i ij T ∂ j 0 T 0 T∞ x j ⎣⎢ x j ⎦⎥ β = 0.075, β * = 0.09 , σ = 0.5 , σ * = 0.5 . Turbulent mixing energy:
∂ ∂ ∂u (ρk) + (ρu k) = τ i − β * ρωk It is known that conventional one- and two-equation ∂t ∂x j ij ∂x j j (4) turbulence models generate Reynolds stresses that respond too rapidly to changes in mean flow conditions ∂ ⎡ ∂k ⎤ + ⎢(µ +σ * µ ) ⎥ partially due to the need to accurately reproduce ∂ T ∂ x j ⎣⎢ x j ⎦⎥ equilibrium flows. As a result, these baseline turbulence Specific dissipation rate: models give unsatisfactory results for flows with significant separation under adverse pressure gradients ∂ ∂ ∂u or across shock waves. In the present computation, the ρω + ρ ω = εω τ i − βρω 2 ( ) ( u j ) ( / k) ij standard k-ω turbulence model is coupled with the lag ∂t ∂x ∂x 26 j j (5) model proposed by Olsen and Coakley to calculate the ∂ ⎡ ∂ω ⎤ turbulent eddy viscosity. + µ +σµ ⎢( T ) ⎥ ∂x ⎢ ∂x ⎥ j ⎣ j ⎦ The basic numerical method used to solve the above system of equations in this paper follows that described Turbulent eddy viscosity: in detail by Liu and Ji28. The integral forms of the ∂ ∂ conservation equations are discretized on quadrilateral (ρν ) + (ρu ν ) = a(R )ωρ(ν −ν ) (6) ∂t t ∂x j t T tE t cells using the finite volume approach. A staggered j scheme is used for the coupling of the Navier-Stokes ρ where t is time, xi position vector, is the density, equations and the k-ω and lag equations. A central µ difference scheme is used to discretize the diffusive ui velocity vector, p pressure, dynamic molecular terms and an upwind Roe’s scheme is used for viscosity, ν kinematic equilibrium turbulent eddy tE convective terms in the Navier-Stokes and the k-ω viscosity, k turbulent mixing energy, and ω the equations. specific dissipation rate. The total energy and enthalpy are E = e + k + u u 2 and H = h + k + u u 2 , After being discretized in space, the governing i i i i equations are reduced to a set of ordinary differential = + ρ = []γ − ρ respectively, with h e p / and e p ( 1) . equations with only derivatives in time, which can be The term γ is the ratio of specific heats. Other solved using a multi-stage Runge-Kutta type scheme. quantities are defined in the following equations: To accelerate the convergence, unsteady multigrid 19 µ = ρν (7) method proposed by Jameson and further T t implemented by Liu and Ji28 is applied in the present ν = ε * ω tE k (8) study for all 7 equations. = ρ µ ω RT k T (9)
1 ∂u ∂u S = ( i + j ) (10) ij ∂ ∂ Results and Discussions 2 x j xi 1 ∂u 2 The numerical method presented above is applied to the τ = 2µ (S − k δ ) − ρkδ (11) ij T ij 3 ∂x ij 3 ij Bauer-Garabedian-Korn (BGK) No. 1 supercritical k airfoil which is extensively investigated experimentally 1 ∂u τˆ = 2µ(S − k δ ) +τ (12) at the National Research Council, Canada. The ij ij ∂ ij ij geometry of the airfoil is shown in Fig. 1. The 3 xk experiments have been conducted at a chord Reynolds µ µ ∂h q = −( + T ) (13) 6 j Pr Pr ∂x number of 20x10 with free transition. The Mach L T j numbers in the investigation ranged from 0.5 to 0.792 ⎡(R + R )⎤ with angle of attack α varying from 1.3 to 11.5 degree. a(R ) = a T T 0 (14) T 0 ⎢ + ⎥ It is reported, from the force and pressure power ⎣(RT RT∞ )⎦ spectrum, strong shock oscillation occurs for Mach
3 American Institute of Aeronautics and Astronautics numbers between 0.69 and 0.733. For the present conditions develops into unsteady with shock-induced numerical simulation, three test cases are conducted oscillation. The average lift coefficient is 1.03 which is with a fixed chord Reynolds number of 20x106 and close to the experimental data of 1.016. Fourier Mach number of 0.71 at three different angles of attack, analysis result shown in Fig. 6 reveals a predominant = π case 1: α =1.396 ; case 2: α = 6.97 ; and case 3: reduced frequency ( k fc U ∞ ) equal to 0.16, which α = 9.0 . Based on Lee’s experimental results, the flow is about 0.4 of the bi-convex circular airfoil. This value is steady for the flow conditions of case 1 and case 3. is about 60% lower than the experimental value of 10 For case 2, however the flow is unsteady with shock Lee . induced oscillation. It should be noted that the Mach number and angle of attack used in the computation are The Mach number contours and skin friction coefficient the same as the experimental ones without any wind Cf plots at different time instants during one oscillation tunnel correction. period are presented in Fig. 7. The instantaneous shock positions on the upper surface of the airfoil are shown The computational grid has a C-topology. Most of the in Fig. 8. The t* equal to zero corresponds to the time computations are performed on a 640x64 mesh with instant at which the lift coefficient Cl is the minimum. 512 points on the surface of the airfoil, 128 nodes in the Both Fig. 7 and Fig. 8 reveal that, different from the wake. The computational domain extends non-lifting bi-convex circular-arc airfoil, the shock approximately 20 chords up and down stream. The oscillation here occurs on the upper side of the airfoil average minimum normal spacing at the wall is only. The shock motion is mainly of the type A 29 approximately 10-6 chord, which corresponds to reported by Tijdeman . Detailed flow features can be + obtained from the Mach number contours and the skin a y value of about 0.6 for an unsteady calculation at friction distributions in Fig. 7. At t*=0, a weak shock Ma = 0.71 with α = 6.97 . The grid independence test occurs at a location of approximately one quarter of the has been conducted and the results will be presented in airfoil, and there is a small separation region after the detail later. shock. During the first half cycle (0