Fast 3D Viscous Calculation Methods

Mark Drela
MIT Aero & Astro

Boeing/MIT Strategic Research Review

6 May 14
 Motivation
2D Inviscid+Integral-BL (IBL) methods have proven very effective
• Enormously faster than alternative Navier-Stokes for similar accuracy
• Can exploit any inviscid flow solver
• Compatible with inverse design methods
• Compatible with virtual displacements for linearized aeroelasticity

Navier Stokes Potential+BL (MSIS, TRANAIR-2D)

∼ 500 000 variables ∼ 5 000 variables
∼ 1 hr. runtime ∼ 1 sec. runtime

⇒ Present goal is to extend inviscid+integral-BL methods to 3D

1/23
 EIF/Defect Formulation
• Real flow decomposed into irrotational Equivalent Inviscid Flow q i
and rotational Viscous Defect ∆q, with . . .
– q i = q outside the rotational viscous layers
– q i · n̂ 6= 0 on solid walls, accounts for presence of viscous layer
(q i is not the solution to the inviscid problem!)
n n qi = q

EIF
Real q i (r )
Viscous Flow
q ( r) Viscous Defect
∆ q ( r) qi − q

Wall BC: q = 0 EIF Wall BC: q i . n 0

( EIF wall mass flux depends on integrated defect )

• Originally employed by LeBalleur with boundary layer methods,

but EIF/Defect formulation does not depend on BL approximations

2/23
 EIF/Defect Formulation
• Governing conservation laws for fluxes f split into EIF and Defect parts
f = f i − (f i −f )
replaced by
∇·f = 0 ⇒ ∇ · fi = 0 , ∇ · (f i −f ) = 0

• Inviscid-equation solutions are economical, well automated

• Defect solutions needed only on compact viscous-region domain

n ∆ .f d = 0 = 0
∆ . fi d

dA
∆ . fi − f d = 0
n
dA
d
d

Defect Control Volume

(viscous region only)

Control Volume

3/23
 EIF/Defect Formulation
Defect conservation law on differential volume spanning viscous layer
ZZZ ZZ "Z ! #
f ye
∇· (f i −f ) dV = ∇· f̃ i − f̃ dy dx dz
yw
ZZ
+ (f i −f )e · n̂e dAe
ZZ
− (f i −f )w · n̂w dAw = 0

where f̃ = fx x̂ + fz ẑ (in-plane flux)

f ∂() ∂()
∇() = ∂x x̂ + ∂z ẑ (in-plane gradient)

Differential Defect Control Volume

Defect Control Volume ne y, υ
d Ae ye( x,z )
n
... ...
...
... d x dz nw
n exit
... ... ... ... x,z , u,w

d Aw yw( x,z )

4/23
 Defect Equations
Defect equations on differential volumes spanning viscous layer
Z
f
(mass i − mass) dy → ∇·M − ρiw qiw· n̂w = 0
Z
(momi − mom) dy f
→ ∇· ¯ − ∇·M
J̄ f f
qiw − τw − (piw−pw)n̂w + ∇Π = 0
Z
(qi · momi −q · mom) dy f
→ ∇·E f
− ∇·M qi2w − ρiQ· ∇q
f 2
i − 2D = 0
Z
◦
f
(qi × momi −q ×mom)· ŷ dy → ∇·E f
− ∇·M qi2w ψiw − ρiQ◦ · ∇q
f 2 ◦
i . . . − 2D = 0

Conserved integral defects (fluxes), source terms

Z
M ≡ (ρiq i − ρq) dy Mass flux defect
Z
¯
J̄ ≡ (ρiq i qTi − ρq qT ) dy Momentum flux defect
Z
E ≡ (ρiq i qi2 − ρq q 2) dy K.E. flux defect
ρi qi
Z
E ≡ (ρiq i qi2ψi − ρq q 2ψ) dy
◦
Curvature flux defect ρq
Z
Q ≡ (q i − q) dy Volume flux defect ρ iw q iw . n w
Z
Π ≡ (pi − p) dy Pressure defect τw
Z =
D ≡ (τ̄¯ ·∇) · q dy Dissipation integral M , J, E , E ...
..

In-plane flow angle variable: ψ (y) ≡ arctan(w/u)

5/23
 Attractive Features of Defect Formulation
• Viscous defect equations are 2D (only on surfaces, wakes)
• Fully exploits any inviscid solver
• Can represent small geometry changes, steady or unsteady
– Inverse or optimization design change without regridding
– Linearized aeroelasticity — static, forced dynamic, flutter
n

Virtual wall displacement

Viscous−displacement transpiration
Virtual−displacement transpiration

6/23
 Central Ideas of Present Work
• Assume EIF is represented by any fast inviscid formulation . . .
– Vortex Lattice (AVL)
– Panel (PANAIR, PMARC, QUADPAN, etc.)
– Transonic Small Disturbance Potential
– Full Potential (TRANAIR)
– Euler (CART-3D, etc.)

• Develop complementary “surface-only” method (IBL3)

to represent Viscous Defect for any EIF formulation
• Use modern sparse-matrix methods (ILU, GMRES . . . )
to solve overall EIF+Defect problem

7/23
 IBL3 Formulation
• Viscous defects represented by assumed profiles, parameterized by
δ A B Ψ (roughly equivalent to δ1∗ θ11 δ2∗ θ12)

• TS, CF-wave amplitudes and Reynolds stresses parameterized by

" # " #!
GC ψC equivalent to ln u′12 +u′22 , arctan ku′2k/ku′1k

• Enormous reduction in number of unknowns from RANS

Navier Stokes IBL3
j n

EIF
Real
Viscous Flow Potential solver
Typical required q i (r )
Navier−Stokes grid resolution
q (r) required for q i
grid resolution
Viscous Defects
(mass, mom., KE)

( ρ ρ q ρe ν ) j δ Ψ GC ψC
200 to 600 DOFs per surface point 6 DOFs per surface point

8/23
 IBL3 Equation Discretization
• Surface finite element discretization allows arbitrary geometry
• Geometry r, velocities q, defects M, J̄¯ . . . computed in global basis XYZ
• Defects put into local surface cartesian basis xyz for residual construction
u = q · x̂ , w = q · ẑ
Mx = M · x̂ , Mz = M · ẑ
Jxx = ¯ · x̂
x̂ · J̄ ¯ · ẑ
, Jxz = x̂ · J̄
Jzx = ¯ · x̂
ẑ · J̄ ¯ · ẑ
, Jzz = ẑ · J̄

n
u
υ y
Equation residual
u
n qi
u
υ w
n q
qi w x
u
υ
Local cartesian basis w
r z
Z w
Y
Common local cartesian basis applied
X to all nodes influencing residual

Global cartesian basis

9/23
 IBL3 Solution Approach
Finite-element discretization in local surface xyz basis
• Greatly simplified solution logic – no need to identify attachment lines, stag-
nation points
• Simple Dirichlet, Neumann PCs – no edge stencils needed
• Compatible with simultaneous solution with inviscid flow
ξ ξ
4 3
x ζ
ζ
1 2
z

N4
N3 Wi Wi
nS

N1 Wi
nS nS
N2 nS
+1 nS
−1 ξ
ζ −1
+1

10/23
 Test Case — Low-AR Wing

• 41 × 12 surface paneling
• 9 × 12 wake paneling (not shown)

• Coupled viscous/inviscid problem with 6528 DOFs

– 5280 viscous DOFs: δ A B Ψ GC ψC at each node
– 1000 inviscid DOFs: λ (panel source strength) on each panel

• Zero-flux viscous BCs on symmetry plane (equivalent to inviscid wall)

11/23
 Test Case — Low-AR Wing
EIF and wall streamlines for laminar flow at Re = 40 000

α = 0.0° EIF streamlines α = 4.0° EIF streamlines

Skin friction lines Skin friction lines

57% span

⇒ Separation lines are naturally captured within IBL3 surface grid

12/23
 Test Case — Low-AR Wing
Comparison of IBL3+panel solution at 57% span with 2D (XFOIL) solution
(2D α set to match local cℓ)

⇒ 3D and 2D separation locations match reasonably well (76% vs 79%)

13/23
 Test Case — Low-AR Wing

Wall streamlines in leading edge region

⇒ Attachment lines are naturally captured in IBL3 surface grid

14/23
Combined transition-prediction / turbulence-lag treatment
Reynolds shear stress coefficients:
−u′1v ′ −u′2v ′
C τ1 ≡ , C τ2 ≡
qi qi
Reynolds stress magnitude, angle variables:
 
2 2 1/2
GC ≡ ln Cτ = ln Cτ1 +Cτ2
C τ2
ψC ≡ arctan
C τ1
Governing equations:
∂GC f
+ qc · ∇GC = fG
∂t
∂ψC f
+ qc · ∇ψ C = fψ
∂t

Source functions fG, fψ model:

– laminar TS,CF wave growth for Cτ < Cτcrit
– turbulence evolution, lags for Cτ > Cτcrit

15/23
 Torpedo Test Case

“Torpedo” over a wall boundary layer

"Torpedo"

Outflow edge

Inflow edge Computational grid

Side View
Inflow edge
q
Leading edge Source
"Torpedo"

16/23
 Torpedo Test Case
1
N= 2 ln (Cτ /Cτ∞ ) Wall friction lines
1
Ncrit=6
8 laminar

7
6 0.8
5
4
3 0.6
2
1
0
0.4

1 0.2
0.9
0.8
0.7
0.6
0.5
0.4 0.9 1
0.3
0.2 0.5 0.6 0.7 0.8 0
0.1 0.4 0 0.2 0.4 0.6 0.8 1
0 0 0.1 0.2 0.3

17/23
 Unsteady Extension
General IBL3 formulation includes . . .
• Unsteady terms, allowing . . .
– Robust time-marching startup, ∆t → ∞ recovers steady solution
– Time-domain unsteady (nonlinear)
– Frequency-domain unsteady (linearized)
• Artificial dissipation
– Necessary to stabilize FEM discretization of hyperbolic IBL equations
– Captures converging-characteristic “shocks” (separation lines)
– Conservative — can only redistribute momentum defect (drag)

 
∂M ∂m f  ¯ ¯ · ∇M
f  f
− qiw + ∇ ·  J̄ − Vǫ h̄  − qiw ∇ · M − τ w = 0

∂t ∂t

18/23
 Torpedo Over Wall – Time-Ramp Test Case
Streamwise shape parameter
t=0.1
t=0.4
3.2 t=1.6
t=5.0 3.2
3
3
2.8
2.8
H1 2.6
2.6
H1
2.4
2.4

2.2
2.2

2 2

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1 1

0.8
0.6
0.4 1
0.8
0.2 0.6
0.4
0.2

19/23
 Time-Ramp Test Case – Torpedo Over Wall
Streamwise momentum thickness

0.003
0.003

0.0025
0.0025

0.002
θ11 0.002

0.0015
0.0015
θ11
0.001 0.001

0.0005 0.0005

0 0

1
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
0.8
0.6
0.4 1
0.8
0.2 0.6
0.4
0.2

20/23
 IBL3 Work Done to Date
• Development of FE discretization on general surface grids
• Full unsteady implementation
• Laminar closure method
• Envelope-eN 3D transition prediction method
• Development of turbulent closure method
• Combined laminar/transition/turbulent method
• IBL3 1.01 Routine Package, Documentation
• Strong coupling with simple inviscid solver (done)
• Paper: “Three-Dimensional Integral Boundary Layer Formulation for General
Configurations”, 21st AIAA CFD Conference, San Diego, June 2013.

21/23
 IBL3 Work Underway or Planned

• Continue testing, calibration

• Better wake treatment
• Implementation into TSD code (Drela, Sato)
• Implementation into TRANAIR (by Boeing)

22/23
 People

• MIT:
– Mark Drela (IBL3, EIF coupling)
– PhD student David Moro (VGNS)
– Bob Haimes (geometry)

• Boeing:
– Dave Young & TRANAIR group
– Sho Sato (TSD code)

23/23