Mobile

A Turbulent Low-Speed Preconditioner for Unsteady Flows About Wind Turbine Airfoils

Description
A Turbulent Low-Speed Preconditioner for Unsteady Flows About Wind Turbine Airfoils
Categories
Published
of 22
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  22nd AIAA Computational Fluid Dynamics Conference, 22-26 June 2015, Dallas, Texas A Turbulent Low-Speed Preconditioner for UnsteadyFlows About Wind Turbine Airfoils Reza Djeddi ∗ , Jason Howison † and Kivanc Ekici ‡ University of Tennessee, Knoxville, Tennessee 37996  A harmonic balance method is used to study the unsteady flows at low-speed regimestypically seen for wind turbines. The frequency-domain technique is applied to the Reynolds-averaged Navier-Stokes governing system of equations with the Spalart-Allmaras turbu-lence model. The convergence, stability and accuracy of the compressible solver is im-proved via a fully coupled viscous preconditioning designed for low speed flows that fall inthe essentially incompressible flow regime. The viscous preconditioner couples the Navier-Stokes equations to the turbulence model through the correct implementation of the pre-conditioning matrix and the subsequently matrix-valued artificial dissipation term. Finally,the viscous low-speed preconditioning is enhanced with a mixed-mechanism that would in-crease the stability and improve the convergence of the harmonic balance solver used forthe simulation of unsteady periodic flows. I. Nomenclature b  semi-chord length of the airfoil c  speed of sound E   total energy F , G  flux vectors H   total enthalpy k  reduced frequency,  k  =  b ω/U  ∞  p  pressure Pr l  laminar Prandtl number Pr t  turbulent Prandtl number S  t  source term for the turbulence model t  time U  vector of conservative variables U 0  vector of primitive variables u,v  Cartesian velocities U  ∞  free stream velocity V   control volume x,y  Cartesian coordinates ξ,η  cross-flow coordinates γ   specific heat ratio µ l ,µ t  laminar and turbulent dynamic viscosity ν   kinematic viscosity˜ ν   working variable for Spalart-Allmaras turbulence model ρ  density τ   pseudo-time ∗ Graduate Student, Department of Mechanical, Aerospace and Biomedical Engineering, Student Member AIAA. † Graduate Student, Department of Mechanical, Aerospace and Biomedical Engineering, Currently Assistant Professor, TheCitadel School of Engineering, Student Member AIAA. ‡ Associate Professor, Department of Mechanical, Aerospace and Biomedical Engineering, Senior Member AIAA.1 of 22American Institute of Aeronautics and Astronautics    D  o  w  n   l  o  a   d  e   d   b  y   U   N   I   V   E   R   S   I   T   Y    O   F   T   E   N   N   E   S   S   E   E  o  n   J  u  n  e   2   3 ,   2   0   1   5   |   h   t   t  p  :   /   /  a  r  c .  a   i  a  a .  o  r  g   |   D   O   I  :   1   0 .   2   5   1   4   /   6 .   2   0   1   5  -   3   2   0   2  22nd AIAA Computational Fluid Dynamics Conference 22-26 June 2015, Dallas, TX AIAA 2015-3202 Copyright © 2015 by Reza Djeddi, Jason Howison and Kivanc Ekici. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission. AIAA Aviation  II. Introduction The aeromechanical design of wind turbines is a complex and multidisciplinary task that requires not onlythe consideration of a very large number of operating regimes of very low to moderate Mach number flows withlow and high turbulence intensities but also the applicability to conventional and prospective unconventionalnew designs subject to challenging off-design conditions. This necessitates a high-fidelity, robust unsteadysolver that can cover all aspects of the present wind turbine designs while building a platform for an adjointdesign optimization tool to further enhance the state-of-the-art. Due to the fact that most of the fundamentalwind turbine unsteady problems are periodic in time, frequency-domain techniques can be utilized to avoidprohibitively costly computational effort of time-domain RANS solvers. The application of a harmonicbalance (HB) method 1 to Navier-Stokes (NS) solvers has proven to be a promising tool for turbomachineryand wind turbine flow problems. 1–4 The aforementioned HB-NS solver can be enhanced with a one-equationSpalart-Allmaras 5 turbulence model that incorporates a rotation correction that reduces the eddy viscosityin the vicinity of regions where the vorticity surpasses the strain rate. 6,7 The spatial discretization of theconvective and diffusive terms includes the central difference formula that can be stabilized and furtherenhanced with an artificial or numerical dissipation that uses a blend of second and fourth differences. 8–10 Amain difference between wind turbine flows and other aerodynamic problems for which compressible harmonicbalance RANS solvers have been used is that the flow-field for wind turbines falls in the incompressible regime.This entails the use of a robust low-speed preconditioning to ensure the accuracy of the solutions in a widerange of flow speeds while guaranteeing a rapid convergence at each sub-time level of the harmonic balanceRANS solver.In this paper, we develop a viscous low-speed preconditioning that incorporates the Spalart-Allmarasturbulence model. This approach not only improves the accuracy of the numerical solution but also alle-viates the convergence problems that are inherited from applying a compressible solver to an essentiallyincompressible flow. The preconditioner couples the turbulence model to the rest of the governing equationsto ensure the correct scaling of the turbulence equation and its respective artificial dissipation term thatimproves the convergence rate and also the accuracy of the compressible solver in low-subsonic flow regimes.The preconditioning is further enhanced with a mixed-mechanism that deals with the unsteadiness of thefluid flow through the correct modification of the preconditioner matrix for the harmonic balance imple-mentation. We believe that explicit handling of all primitive variables including the working variable of theturbulence model for the improved preconditioner is presented for the first time in this work for unsteadyanalysis of wind turbine flows using a harmonic balance RANS solver. III. Governing Equations and Mathematical Formulation Unsteady flow past the wind turbine airfoils is governed by the Reynolds-averaged Navier-Stokes equationswhich include modified Spalart-Allmaras turbulence model. The two-dimensional conservation form of thegoverning equations for a moving grid then becomes: ∂  U ∂t  +  ∂  F ∂x  +  ∂  G ∂y  =  S  (1)where the vector of conservative variables is  U  = [ ρ,ρu,ρv,ρE,ρ ˜ ν  ] T  and the vectors of inviscid and viscousfluxes in  x - and  y -directions,  F  and  G , and the vector of source terms,  S , are defined as: F  =  ρu − ρu g ρu 2 +  p − τ  xx − ρuu g ρuv − τ  xy  − ρvu g ρuH   − τ  xH   − ρEu g ρu ˜ ν   − τ  xν   − ρ ˜ νu g  ,  G  =  ρv − ρv g ρuv − τ  yx − ρuv g ρv 2 +  p − τ  yy  − ρvv g ρvH   − τ  yH   − ρEv g ρv ˜ ν   − τ  yν   − ρ ˜ νv g  ,  S  =  0000 S  t  In general, the unsteady grid motion velocity is incorporated into the governing equations by  u g  and  v g in  x  and  y  directions, respectively. The pressure  p  and enthalpy  H   are defined in terms of the conservativevariables based on the assumption of an ideal gas with a constant specific heat ratio as: 2 of 22American Institute of Aeronautics and Astronautics    D  o  w  n   l  o  a   d  e   d   b  y   U   N   I   V   E   R   S   I   T   Y    O   F   T   E   N   N   E   S   S   E   E  o  n   J  u  n  e   2   3 ,   2   0   1   5   |   h   t   t  p  :   /   /  a  r  c .  a   i  a  a .  o  r  g   |   D   O   I  :   1   0 .   2   5   1   4   /   6 .   2   0   1   5  -   3   2   0   2   p  = ( γ   − 1) ρ  E   −  12( u 2 +  v 2 )  H   =  ρE   +  pρ  =  γ γ   − 1  pρ  + 12( u 2 +  v 2 )While the inviscid fluxes depend on the conservative variables  U , viscous fluxes depend on the gradientsof the flow velocities, temperature and the working variable of the turbulence model as defined below for theshear stress terms of the Navier-Stokes equations: τ  xx  = ( µ l  +  µ t )  43 ∂u∂x  −  23 ∂v∂y  τ  yy  = ( µ l  +  µ t )  43 ∂v∂y  −  23 ∂u∂x  τ  xy  =  τ  yx  = ( µ l  +  µ t )  ∂u∂y  −  ∂v∂x  τ  xH   =  uτ  xx  +  vτ  xy  +   µ l Pr l +  µ t Pr t   ∂H ∂xτ  yH   =  uτ  xy  +  vτ  yy  +   µ l Pr l +  µ t Pr t   ∂H ∂y In the above equations,  µ l  and  µ t  are the laminar and eddy viscosities, respectively, where the former isdefined based on the Sutherland’s law and the latter is determined using the Spalart-Allmaras turbulencemodel, 5–7 which also defines the respective source term,  S  t , in Equation (1). IV. Harmonic Balance RANS Solver Wind turbine aerodynamics consist of unsteady but temporally periodic flows usually due to the pitchingblades or yawed wind regimes. In these cases, the conservative variables for which the governing equationsare solved, can be written in terms of a truncated Fourier series up to a predefined number of harmonics by: U ∗ ( x,y,t i ) =  A 0 ( x,y ) + N   n =1 [ A n ( x,y )cos( ωnt i ) + B n ( x,y )sin( ωnt i )] ;  i  = 1 : 2 N   + 1 (2)where  ω  is the fundamental frequency of excitation and  A 0 ,  A n  and  B n  are the Fourier series coefficientsdefined in terms of the conservative variables. The Fourier series is truncated in a way that the flow variablesare computed and stored at 2 N  +1 equally-spaced sub-time levels over a single period. Based on Equation (2)and using a discrete Fourier transform, the Fourier series coefficients can be determined from the solutionsstored at each sub-time level as shown below: U ∗ =  E ˆ U  (3)In a similar fashion, the conservative variables at the sub-time levels can be determined using the inversediscrete Fourier transform:ˆ U  =  E − 1 U ∗ (4)It is worth mentioning that the discrete Fourier transform,  E , and its inverse,  E − 1 are both squarematrices with a size corresponding to 2 N   +1 times  N   + N   +1 where 2 N   +1 is the number of sub-time levelsand there exists  N   sine coefficients,  N   cosine coefficients and one zeroth harmonic coefficient. Looking backat the unsteady governing equations in the time-domain, we have: ddt  ( V  U ) + Q C  ( U ) + Q V   ( U ) − Q D ( U ) = 0 (5) 3 of 22American Institute of Aeronautics and Astronautics    D  o  w  n   l  o  a   d  e   d   b  y   U   N   I   V   E   R   S   I   T   Y    O   F   T   E   N   N   E   S   S   E   E  o  n   J  u  n  e   2   3 ,   2   0   1   5   |   h   t   t  p  :   /   /  a  r  c .  a   i  a  a .  o  r  g   |   D   O   I  :   1   0 .   2   5   1   4   /   6 .   2   0   1   5  -   3   2   0   2  where  V   is the control volume. Here, the convective terms,  Q C  ( U ), are evaluated at the cell centers byaveraging the conservative variables at the corresponding cell vertex while a compact stencil central differenceformulation 11 is used for the viscous terms,  Q V   ( U ), which slightly reduces the odd-even decoupling in thesolution field. However, the flux balance at each vertex still requires a robust artificial dissipation term, Q D ( U ). The srcinal Jameson-Schmidt-Turkel (JST) formulation 8 is utilized to obtain the “scalar” artificialdissipation, which blends second and fourth differences of the conservation variables where the latter providesthe background dissipation that is essential for stability and convergence toward steady state. However, theuse of a low-speed preconditioning mechanism, which will be discussed in the following section, makes itinevitable to have a “matrix-valued” numerical viscosity that is scaled by the preconditioning matrix in afully coupled fashion. It is worth mentioning that the dissipation or smoothing coefficients for the secondand fourth differences in the artificial dissipation model are defined using a pressure switch 8 which is furthermodified 9 to ensure total variation diminishing (TVD) properties. Finally, the system of semi-discretizedinitial-value ODEs is marched temporally toward a steady state solution with an explicit multistage Runge-Kutta scheme in which the viscous terms are frozen at the first stage and the artificial dissipation terms areevaluated only at the odd stages.Using the harmonic balance method, as explained herein, the governing equations of (1) are then rewrittensimilar to equation (5) in the semi-discrete form for all sub-time levels, i.e. ddt  ( V  ∗ U ∗ ) + Q C  ( U ∗ ) + Q V   ( U ∗ ) − Q D ( U ∗ ) = 0 (6)where  V  ∗ is the control volume and the rest of the left-hand-side terms are the convective, viscous andartificial dissipation terms evaluated at each sub-time level. Here, the time-derivative term in equation (6)is replaced by a pseudo-spectral operator  D , that gives: ω D ( V  ∗ U ∗ ) + Q C  ( U ∗ ) + Q V   ( U ∗ ) − Q D ( U ∗ ) = 0 (7)In order to obtain the steady state solution at each sub-time level, a “pseudo-time” derivative term isadded to the harmonic balance equation of (7) which enables us to use the solution algorithm explainedearlier in this paper. This equation now reads: ∂ ∂τ   ( V  ∗ U ∗ ) +  ω D ( V  ∗ U ∗ ) + Q C  ( U ∗ ) + Q V   ( U ∗ ) − Q D ( U ∗ ) = 0 (8)where  τ   is the pseudo time which enables us to march the above equation to steady state, vanishing thepseudo-time term. Thus, the pseudo-time harmonic balance equations of (8) can be solved in a similarfashion to the time-domain governing equations of (5).As shown earlier by Hall et al., 1 an explicit treatment of the pseudo-spectral term in an explicit time-marching scheme can cause instabilities for the numerical solver. While non-reflecting far-field boundariestypically remedy these stability issues for low reduced frequencies, extra care must be given to cases withhigher reduced frequencies or when too many harmonics are retained. In the present study, the CFL numberconstraint that was initially proposed by Van der Weide et al. 12 is used to stabilize the explicit solver for awide range of reduced frequencies and number of harmonics. The adapted stabilization technique introducesexcitation frequency and the number of harmonics ( N  ) into the calculation of the time-step where∆ t  = CFL  V  λ  +  ωN  V   (9)where   λ   is taken to be the maximum of the spectral radius in all directions. The implementation of the present harmonic balance solver in wind turbine applications was presented by Howison and Ekici 3,4,13 along with the validating test cases and the definition of boundary conditions at the airfoil surface and atfar-field. In this work, we develop a new low-speed preconditioner including the effects of turbulence modeland implement it into our HB solver. V. Low-Speed Preconditioning Incompressible flow regimes in which the density variations become small as the Mach number approacheszero, can be handled by a simplified incompressible form of the governing equations. However, in somesituations we have to stay with the compressible flow solver since our simulations should ideally cover a 4 of 22American Institute of Aeronautics and Astronautics    D  o  w  n   l  o  a   d  e   d   b  y   U   N   I   V   E   R   S   I   T   Y    O   F   T   E   N   N   E   S   S   E   E  o  n   J  u  n  e   2   3 ,   2   0   1   5   |   h   t   t  p  :   /   /  a  r  c .  a   i  a  a .  o  r  g   |   D   O   I  :   1   0 .   2   5   1   4   /   6 .   2   0   1   5  -   3   2   0   2  wider range of flow regimes, which is the case for wind turbine applications. At low Mach numbers and ona fixed grid, convective waves that are traveling at speed  u  become much slower than the acoustic wavesthat are traveling at  u  +  c . What happens mathematically is that this disparity between the eigenvalues of the Jacobian matrices stiffens the governing equations, destabilizes the solution, slows down the convergenceand degrades the accuracy by resulting in unbalanced amounts of artificial dissipation. Also, when anexplicit time-marching method is used to obtain a steady state solution, the local time step depends onthe eigenvalues of the flux Jacobian matrices and the aforementioned disparity in wave speeds substantiallyimpairs the convergence rate. In order to overcome these problems, low-speed preconditioning methods havebeen introduced which basically scale the acoustic wave speeds to be more on the order of the convective wavespeeds. Various efforts have been made which are presented in the literature by pioneers such as Turkel, 14–16 Choi and Merkle, 17 Weiss and Smith, 18 and Van Leer 19 with varying degrees of success and a multitude of simplifications and limitations that are imposed along the way, especially in viscous cases.In simple words, almost all of these tools pre-multiply the time derivatives by a preconditioning matrixto overcome the stiffness problem of the system of equations. Also, more often, a set of primitive variablesis used to construct this matrix since it has been proven that it can ease up the derivation process. Thedifferences would then be in terms of the preconditioner matrix and the set of primitive variables used. In arecent effort, Collin et al. 20 have studied a robust low-speed preconditioning tool for viscous/turbulent flowsbased on the symmetric preconditioner of Weiss and Smith 18 and the viscous preconditioner of Choi andMerkle 17 in a combined so-called form of WSCM preconditioner. Also, Campobasso and Baba-Ahmadi 2 haveused a slightly different viscous preconditioner that srcinated from works of Venkateswaran and Merkle 21 and applied it to unsteady flow past horizontal axis wind turbine blades. More recently, Howison andEkici 3 used an inviscid preconditioner based on Turkel’s method 16 for unsteady harmonic balance analysis of wind turbine flows. However, in all of these earlier works, either an inviscid preconditioner was used or theEuler/Navier-Stokes equations were preconditioned separately and therefore a robust viscous preconditioningof RANS equations including the turbulence model has remained untouched. A. Preconditioning Matrix In this paper, a fully coupled viscous preconditioning of RANS equations that incorporates the Spalart-Allmaras turbulence model is studied and developed to tackle the convergence, stability and accuracyproblems of solving an essentially incompressible flow with a compressible flow solver. First, consider thetwo-dimensional RANS equations in primitive variables on a fixed grid: ∂  U 0 ∂t  +  ∂  F 0 ∂x  +  ∂  G 0 ∂y  =  S   + L ν  ( U 0 ) (10)where the primitive or non-conservation variables (pressure-velocity-entropy) are defined as U 0  = [  p,u,v,S,  ˜ ν  ] T  and the inviscid flux vectors in  x - and  y -directions are given as: F 0  =  u (  p  +  ρc 2 ) u 2 +  p/ρuvuS u ˜ ν   ,  G 0  =  v (  p  +  ρc 2 ) uvv 2 +  p/ρvS v ˜ ν   Also, the right hand side of equation (10) includes the vector of source terms,  S , and  L ν   is the differentialoperator for the viscous terms. Equation (10) can be rewritten in quasi-linear form as below: ∂  U 0 ∂t  + A 0 ∂  U 0 ∂x  + B 0 ∂  U 0 ∂y  =  R . H . S .  (11)where the Jacobian matrices of the flux vectors are given by: 5 of 22American Institute of Aeronautics and Astronautics    D  o  w  n   l  o  a   d  e   d   b  y   U   N   I   V   E   R   S   I   T   Y    O   F   T   E   N   N   E   S   S   E   E  o  n   J  u  n  e   2   3 ,   2   0   1   5   |   h   t   t  p  :   /   /  a  r  c .  a   i  a  a .  o  r  g   |   D   O   I  :   1   0 .   2   5   1   4   /   6 .   2   0   1   5  -   3   2   0   2
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks