Drag Workshop Results Using CFL3D
Christopher RumseyRobert Biedron
NASALangley Research CenterHampton, VA
DLR-F4M=0.75Recref=3 million
Issues addressed
• Effect of grid (1-to-1 vs. overset)
• Comparison of 3 turbulence models
• Issue of transition for supposedly “fully turbulent” computations
• Effect of different versions of SA model
CFL3D V6.0
• Upwind, implicit 3-factor AF
• Finite volume, multigrid
• FDS (Roe)
• Globally 2nd order spatially accurate
• Multi-block capabilities, including 1-to-1, patched, and overset
• Parallel (MPI)
Grid convergence1-to-1 grid, M=0.75, CL=0.5
• If every-other-point grid is in asymptotic region for 2nd
order global spatial convergence (doubtful), then CD on fine grid is high by 30 counts!
• A finer-level grid of the same family is needed
N-2/3
CD
0 0.0002 0.0004 0.0006 0.00080.02
0.03
0.04
0.05
0.06
0.07
standard 1-to-1 grid(3,180,800 cells)
Effect of grid on surface pressuresα=0 deg, M=0.75, Re=3.e6
Drag Prediction Workshop standard grids
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.238
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.331
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.409
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.512
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.636
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.844
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
CFL3D, oversetCFL3D, 1-to-1exp
2y/B = 0.185
Detail
x/c
-Cp
0 0.2 0.4 0.6 0.8 1-0.8
-0.4
0
0.4
0.8
1.2
1.6
CFL3D, oversetCFL3D, 1-to-1exp
2y/B=0.331
Grid comparison near y=230 mm
Overset grid
X Y
Z
1-to-1 grid
3.2 million cells50c extent
3.7 million cells170c extent
Entire grid:
Effect of grid on forces & moments
α
CL
-4 -2 0 2 4 60
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SA, oversetexp
CM
CL
-0.16 -0.12 -0.08 -0.04 00
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SA, oversetexp
Effect of grid on forces & moments
CD
CL
0 0.02 0.04 0.06 0.08 0.10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SA, oversetexp
CD, CDp, CDv
CL
0 0.01 0.02 0.03 0.04 0.050
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SA, oversetCDpCDp
CDvCDv
Effect of turbulence model on surface pressuresα=2 deg, M=0.75, Re=3.e6
Drag Prediction Workshop 1-to-1 standard grid
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.844
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.636
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.512
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.409
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.331
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
2y/B=0.238
x/c
-Cp
0 0.2 0.4 0.6 0.8 1
SASSTEASM
2y/B = 0.185
Detail
x/c
-Cp
0 0.2 0.4 0.6 0.8 1-0.8
-0.4
0
0.4
0.8
1.2
1.6SASSTEASM
2y/B=0.636
Streamlines at alpha=2 deg
SA SST
EASM
Effect of turbulence model on forces & moments
α
CL
-4 -2 0 2 4 60
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SST, 1-to-1EASM, 1-to-1exp
CM
CL
-0.16 -0.12 -0.08 -0.04 00
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SST, 1-to-1EASM, 1-to-1exp
Effect of turbulence model on forces & moments
CD
CL
0 0.02 0.04 0.06 0.08 0.10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SST, 1-to-1EASM, 1-to-1exp
CD, CDp, CDv
CL
0 0.01 0.02 0.03 0.04 0.050
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SST, 1-to-1EASM, 1-to-1CDp
CDpCDp
CDvCDv
CDv
Actual “fully turbulent” transition locations for different turbulence models
alpha=0 deg
y, mm
x,m
m
100 200 300 400 500 600
300
400
500
600
700
wing outlineSASSTEASM
y, mm
x,m
m
350 400
500
510
520
530
540
550
560
570
580
590
600
wing outlineSASSTEASM
SA: 1.4%cSST: 2.8%cEASM: 5.7%c
At y=400 mm:
Effect of forcing SA transition to match “natural” transition of EASM
CD, CDp, CDv
CL
0 0.01 0.02 0.03 0.04 0.050
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SA (trans), 1-to-1CDpCDp
CDvCDv
Effect of SA version on transition• 2 versions of SA in wide use:
– SA (Ia): “official” version in Aerospatiale Journal– SA+fv3: “unofficial” version resulting from a Spalart
e-mail in early 90’s
• SA (Ia) transitions very near L.E.– Typically 1 – 2 % c for alpha=0 deg case
• SA+fv3 delays transition for low Re (1-10 million)– 7 – 8 % c or more for alpha=0 deg case (OVERFLOW
results similar)– Seems to show more sensitivity to grid & free stream
turbulence level chosen
Effect of SA version on forces
CD, CDp, CDv
CL
0 0.01 0.02 0.03 0.04 0.050
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SA + fv3, 1-to-1CDp
CDpCDv
CDv
α
CL
-4 -2 0 2 4 60
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
SA, 1-to-1SA + fv3, 1-to-1exp
Summary
• Grid issues– Official 1-to-1 grid too coarse to resolve
pressures (L.E. & shock under-resolved)– Nonetheless, global forces & moments similar
to those using better quality overset grid; at alpha=0:
• CD: ~8 count difference (2.4 %)
– Family of grids (2 or 3 for each type) needed for grid sensitivity study
Summary, cont’d
• Turbulence model comparison (1-to-1 grid)– SST & EASM give lower CD than SA by <20
counts (8.7 % difference at –3 deg, 3.3 % difference at +2 deg)
– Primarily due to lower friction drag
Summary, cont’d
• “Fully turbulent” is misnomer– All turbulence models “transition” on their own
– At low Re (order 1-10 million), transition is not at the leading edge! E.g., for alpha=0:
• SA: 1-2 %c typical
• SST: 2-5 %c typical
• EASM: 2-7 %c typical
– Effect is small: forcing SA to transition at EASM location changes CD by <1 count (0.1 %)
Summary, cont’d
• Two versions of SA are known to be present in today’s U.S. production codes– SA + fv3 (unofficial version) widely used, can
delay transition significantly for low Re (order 1-10 million) compared to official SA (Ia)
– Effect for alpha=0:• Delta CD=1.4 counts (0.4 %)
Summary of effects at alpha=0
grid turb model trans loc SA version
effe
ct,i
np
erce
nt
0
1
2
3
4
5
6
L
D
M
L
D
M L DM L
D
M
L=liftD=dragM=moment
Conclusions• Good quality grid a MUST
– It is possible to miss details in Cp yet do reasonably well on forces and moments
– Right answer for wrong reasons? – only a grid study using a family of grids will tell
• SA, SST, EASM turbulence models give verysimilar results for this case (but still ~20 count drag difference)
• CFD transition location should always be checked• Better version control and consistency checks are
needed for turbulence model coding
Comments on EASM
• Nonlinear explicit algebraic stress model, k-omega form (AIAA 2000-4323)
• More robust than earlier versions of EASM
• Roughly 40 % more expensive than SA
• As good as SA and SST for aerodynamic thin-shear flows, but better for flows where nonlinear and curvature effects are important
• Validation on-going
Recommendations for future workshops• Give out family of successively finer grids for a required
grid study– Grid study needed for CFD validation of this type– Some participants do not have 3D grid generation capability– “Official” grids ensure consistency– For wing body: 7 million, 3 million, 1.5 million cells? (structured)– Structured grids should be multigriddable
• Include surface Cps as part of required results– Integrated quantities hide things that could be helpful in evaluation
• More fixed-alpha cases and fewer fixed-CL cases– Fixed alpha cases are easier to run & better for comparing code-to-code
• To ensure transition location is not a cause of variability:– Force transition at specified locations (harder), or…– Include high Re (order 50 million) fully turbulent case (easier)