Spatial Discretisation
Transcription
Spatial Discretisation
Spatial Discretisation Markus Widhalm Folie > Vortrag > Autor Dokumentname > 23.11.2004 Content spatial discretisation upwind scheme central scheme upwind flux basic gradient computation central flux basic artificial dissipation turbulence flux full viscous discretization thin shear layer approximation Folie > Vortrag > Autor Dokumentname > 23.11.2004 Road Map Spatial discretisation for flux computation Inviscid 1st order 2nd order viscous 1st or 2nd order Upwind Verfahren Central scheme + artificial dissipation Full viscous flux Gradient computation Thin layer approximation Folie > Vortrag > Autor Dokumentname > 23.11.2004 Central scheme Folie > Vortrag > Autor Dokumentname > 23.11.2004 Central scheme Basics e.g. the continuity equation ρt + uρx = 0 discretization of the continuity equation n " ρn+1 − ρ u ! n i n i =− ρi+1 − ρi−1 ∆t 2∆x second - order central scheme n ... time step, i ... spatial step Folie > Vortrag > Autor Dokumentname > 23.11.2004 Central scheme Basics Flux computation +&*( F! = density x-momentum y-momentum z-momentum energy F!f ace 012, /. $& $' & $ ! $ ($ & % $ !"# )( ' ($$ & $ ! $ &$ & % $ !"# -., , )& " 1 1 !! Fl + F!r − α̃ (w = !r − w ! l) 2 2 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Central scheme Basics What is α? α decides about the dissipation scheme scalar dissipation - α becomes the maximum eigenvalue matrix dissipation - α becomes a matrix with 3 eigenvalues Folie > Vortrag > Autor Dokumentname > 23.11.2004 Central scheme Basics What is α? α decides about the dissipation scheme scalar dissipation - α becomes the maximum eigenvalue matrix dissipation - α becomes a matrix with 3 eigenvalues How do we compute the difference? (w !r − w ! l) Pj5 Pj4 Pj6 Pj2 Pj0 ! D = (w !r − w ! l) Pj3 Pj7 Fi Pj8 Pj1 Pj9 dual control volume = "2 (!ur − !ul ) − "4 (L(!ur ) − L(!ul )) Folie > Vortrag > Autor Dokumentname > 23.11.2004 Scalar dissipation Parameter Parameter input Inviscid flux discretization type: Central Folie > Vortrag > Autor Dokumentname > 23.11.2004 Scalar dissipation Parameter Parameter input Inviscid flux discretization type: Central Central dissipation scheme: Scalar_dissipation 2nd order dissipation coefficient: 0.5 Inverse 4th order dissipation coefficient: 64 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Scalar dissipation Parameter Why do we need two parameters? 2nd order dissipation coefficient: 0.5 handles discontinuities Folie > Vortrag > Autor Dokumentname > 23.11.2004 Scalar dissipation Parameter -1.5 Z k2 = 0.25 k2 = 0.5 k2 = 0.8 -1 shock capturing low -0.5 cp M = 0.74 α = 2.0° Why do we need two parameters? 2nd order dissipation coefficient: 0.5 handles discontinuities 0 0.5 high 1 1.5 0 0.2 0.4 X 0.6 0.8 2nd order coefficient 1 = 0.25 4 1 = 0.5 2 1 = 0.66 1.5 1 = 0.8 1.25 1 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Scalar dissipation Parameter M = 0.74 α = 2.0° Why do we need two parameters? 2nd order dissipation coefficient: 0.5 handles discontinuities k2 = 0.25 k2 = 0.5 k2 = 0.8 shock capturing low high 2nd order coefficient 1 = 0.25 4 1 = 0.5 2 1 = 0.66 1.5 1 = 0.8 1.25 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Scalar dissipation Parameter dissipation high 0.1 z k4 k4 k4 k4 0.05 pl M = 0.74 α = 2.0° Why do we need two parameters? Inverse 4th order dissipation coefficient dissipation in smooth regions 64 → low 0.2 0.4 x 0.6 0.8 16 → 32 → = 16 = 32 = 64 = 128 0 0 Inverse 4th order coefficient 128 → 1 16 1 32 1 64 1 128 1 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Matrix dissipation Parameter Parameter input Inviscid flux discretization type: Central Central dissipation scheme: Matrix_dissipation 2nd order dissipation coefficient: 0.5 Inverse 4th order dissipation coefficient: 64 Matrix dissipation terms coefficient: 0.5 Minimum artificial dissipation for acoustic waves: 0.2 Minimum artificial dissipation for velocity: 0.2 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Matrix dissipation Parameter For robustness reasons we need extra parameters: Matrix dissipation terms coefficient: 0.5 Minimum artificial dissipation for acoustic waves: 0.2 Minimum artificial dissipation for velocity: 0.2 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Matrix dissipation Parameter For robustness reasons we need extra parameters: Matrix dissipation terms coefficient: 1.0 Minimum artificial dissipation for acoustic waves: 0.2 Minimum artificial dissipation for velocity: 0.2 α̃("ul , "ur , λ) = kCM · T |λ|T −1 |λ| = diag(|λ1 |, |λ2 |, |λ3 |) |λ1 | = max(|vn + Af |), δa (|vn | + Af ) |λ2 | = max(|vn − Af |), δa (|vn | + Af ) |λ3 | = max(|vn |), δv (|vn | + Af ) Folie > Vortrag > Autor Dokumentname > 23.11.2004 Matrix dissipation Parameter Parameter input With the matrix dissipation - change from matrix to scalar !!! Minimum artificial dissipation for acoustic waves: 0.2 Minimum artificial dissipation for velocity: 0.2 Acoustic Velocity Scalar Matrix 0.0 0.0 0% 100% 0.2 0.2 20% 80% 0.7 0.3 70% 30% 1.0 1.0 100% 0% Folie > Vortrag > Autor Dokumentname > 23.11.2004 Scalar & Matrix dissipation What is the main difference between both dissipation schemes ? scalar dissipation: very stable very less parameters needed matrix dissipation: scales with all three eigenvalues of the flux jacobian less dissipative than scalar dissipation mainly seen in the boundary layer region needs experience for the extra parameters to run a stable computation Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Basics e.g. the continuity equation ρt + uρx = 0 discretization of the continuity equation n " ρn+1 − ρ u ! n i n i =− ρi − ρi−1 ∆t ∆x first - order upwind scheme n ... time step, i ... spatial step Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Basics Flux computation ! F = density x-momentum y-momentum z-momentum energy F!f ace +&*( 012, /. $& $' & $ ! $ ($ & % $ # " ! )( ' ($$ & $ ! $ $ & & $% # " ! , . , )& " 1# # 1 !! ! # Fl + Fr − Ā (w = !r, w ! l , !nl,r )# (w !r − w ! l) 2 2 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Basics 2nd ( higher order) discretization explanatory for ρ: ρf ace ! " 1 # = ρnode + ψ ∇ρnode #li − #li−1 2 +&*( 012, /. &$& $' $ ! &$($ % $ !"# )( ' $&($$ ! $ %&$& $ # !" , ,-. )& Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Basics 2nd ( higher order) discretization explanatory for ρ: A ρf ace ! " 1 # = ρnode + ψ ∇ρnode #li − #li−1 2 B A) We need the gradients from the state variables! +&*( B) We need a limiter for discontinuities! 012, /. &$& $' $ ! &$($ % $ !"# )( ' second - order upwind scheme $&($$ ! $ %&$& $ # !" , ,-. )& Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter How do we control this features through the parameterfile? Considerations have to be done for: First order upwind Second order upwind Gradient computation Limiters used Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter Parameter input Inviscid flux discretization type: Upwind Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter Parameter input Inviscid flux discretization type: Upwind Upwind flux: Roe Van_Leer AUSMDV AUSMP AUSM_Van_Leer EFM MAPS+ Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter Roe: Roe, P.L., Approximate Riemann Solvers Parametric Vectors and Differences Schemes. JCP, Vol. 43, 1981 AUSM (Advective upstream splitting mehtod) : Liou, M.S.; Steffen, C.J., A new flux splitting scheme. JCP, Vol. 107, 1993 AUSMP: Wada, Y.; Liou, M.S., A flux splitting scheme with highresolution and robustness for discontinuities. AIAA 94-0083, 1994 EFM (Equilibrium flux method) : Pullin, D.I., Direct simulation methods for compressible inviscid ideal gas flow, JCP, Vol. 34, 1980 MAPS+ (Mach number based Advection Pressure Splitting): Rossow, C.C., A flux splitting scheme for compressible and incompressible flows. JCP, Vol. 164, 2000 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter Parameter input Inviscid flux discretization type: Upwind Upwind flux: Order of upwind flux (1-2): 2 Order of additional euqations (1-2): 2 Roe Van_Leer AUSMDV AUSMP AUSM_Van_Leer EFM MAPS+ Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter Parameter input Inviscid flux discretization type: Upwind Upwind flux: Order of upwind flux (1-2): 2 Order of additional equations (1-2): 2 Reconstruction of gradients: Green_Gauss Least_square Roe Van_Leer AUSMDV AUSMP AUSM_Van_Leer EFM MAPS+ Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter Reconstruction of gradients: Green_Gauss n ! 1 1 ! ∇φ = (φi + φ0 ) !ni Si V i 2 Least_square φx ! = φy = x ∼ ∇φ = R−1 QT b φz Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Error of the reconstruction on hybrid meshes φ=x+y Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Error of the reconstruction on hybrid meshes φ=x+y Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter Reconstruction of gradients inside structured part of a hybrid mesh Automatically with a MUSCL scheme lines found during the preprocessing step hl h pll hr pr pl prr face h pr pl face hl hr prr pll h pl pr face Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Parameter Parameter input Inviscid flux discretization type: Upwind Upwind flux: Order of upwind flux (1-2): 2 Order of additional equations (1-2): 2 Reconstruction of gradients: Green_Gauss Least_square Limiter freezing convergence: 0 Mach number limit for limiter: 0 Venkatakrishnan limiter constant: 1 Lowest pressure for 2nd order: 0 Roe Van_Leer AUSMDV AUSMP AUSM_Van_Leer EFM MAPS+ Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Limiters Limiter freezing convergence: Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Limiters Mach number limit for limiter: Detail ! Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Limiters Venkatakrishnan limiter constant Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Limiters Venkatakrishnan limiter constant -1.5 -1.5 M = 0.74 α = 2.0° -1 -1 0 -0.5 Z 0.5 cp cp -0.5 Venkata 0 Venkata 0.5 Venkata 1 0 Venkata 2 1 0 0.2 0.4 0.5 X 0.6 0.8 Z Venkata 0 1 Venkata 0.5 Venkata 1 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Upwind scheme Stability Lowest pressure for 2nd order: 0.001 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Viscous flux computation Folie > Vortrag > Autor Dokumentname > 23.11.2004 Viscous flux computation Basics Consider friction of the fluid F! = density x-momentum y-momentum z-momentum energy F !inviscid + F!viscid ! =F + n × T ransportequations (n Equation T urbulencemodel) Folie > Vortrag > Autor Dokumentname > 23.11.2004 Viscous flux computation Basics Consider friction of the fluid F! = density x-momentum y-momentum z-momentum energy F !inviscid + F!viscid ! =F + n × T ransportequations (n Equation T urbulencemodel) Folie > Vortrag > Autor Dokumentname > 23.11.2004 Viscous flux computation Basics Viscous part effective for momentum and energy equation momentum D"v = ∇P + ρf" Dt 0 σxx - p 0 + τyx p τzx ρ p P = − 0 0 0 p 0 ρf" ... volumef orces τxy σyy - p τyz τxz τyz σzz - p σ ... normal stress How do we model P ? τ ... shear stress ! " Newtonian fluid ∂vi ∂vj 2 pij = −pδij + µ + − δij ∇ ∂xj ∂xi 3 δi,j ... Kronecker tensor Folie > Vortrag > Autor Dokumentname > 23.11.2004 Viscous flux computation Parameter Parameter input Viscous flux type TSL/Full (0/1): 1 Full viscous Face Pi Pj ! " 1 ! j + ∇w ! i ! = ∇w ∇w 2 Folie > Vortrag > Autor Dokumentname > 23.11.2004 Viscous flux computation Parameter Parameter input Viscous flux type TSL/Full (0/1): 0 (Full) Thin Shear Layer Face Pi Pj wj − wi ! ∇w ≈ ∆x Folie > Vortrag > Autor Dokumentname > 23.11.2004 Viscous flux computation Parameter some comments on Thin Shear Layer approximation: more stable almost same solution maybe less accurate High-Lift computation (SAE, SA) error was negligible viscous flux jacobian depends on p0 and p1 only some comments on Full Viscous approximation exact formulation complex limiting of gradients Folie > Vortrag > Autor Dokumentname > 23.11.2004 Viscous flux computation Parameter Parameter input Turbulent convection scheme for central RANS scheme: Central AUSMDV Roe ConsVarAveragedRoe default settings: laminar and one equation turbulence models: Central probably less stable but more accurate all other turbulence models: Roe more stable but less accurate Folie > Vortrag > Autor Dokumentname > 23.11.2004 Thank you for your attention. Have fun with TAU Folie > Vortrag > Autor Dokumentname > 23.11.2004