I
Panel Discussion:
Perspectives on the Future of CFD
FLUIDS 2000 Conference
Denver, Colorado
June 19-22, 2000
Dochan Kwak
Research Branch
Numerical Aerospace Simulation (NAS) Systems Division
NASA Ames Research Center
Moffett Field, California
dkwak @mail.arc.nasa.gov
(65O)6O4-674O
Topics for Discussion
I I
• What have we accomplished?
• Is CFD mature enough?
• Where can we go from here?
_=,,,,.,,_=, Progress to Date
• 2.,FO *a-; picr:eered the f!e!d )f flow simulation for
- Obtaining engineering solutions involving complex configurations
- Understanding physics (critical to mission success)
• CFC :-as progressed as compu_[r,g power has increased
- Numerical methods have been advanced as CPU and memory increases
- N-S solution of full configuration was a big goal in the 80s
- Complex configurations are routinely computed now
- DNS/LES are used to study turbulence
• As the computing resources changed to parallel and distributed platforms,
computer science aspects become ;mportant such as
Scalability (algorithmic & implementation)
- Portability, transparent codings e_c
Examples of Current Capability
• Algorithmic advances include
- Discrete models :
Various artificial dissipation models
Unified formulations, e.g. preconditioning
Unstructured methodology
Various gridding strategies
- Solution methods:
Explicit/Implicit
Preconditioning, dual-time
Multi-grid
bucce_,,.,, application of CFD ,'o engmeer:ng probgems
- High-lift configurations
Multiple bodies in relative motion
Components of propulsion system (both aero & space)
- Maneuvering vehicle
List goes on
I
/f Examples of Current Capability: OVERSET CFD Tools
J, tam_mu¢_ _
I
i
• E _,__,IFLE L_ HD!i'IG COr, iF'GUR;TION
22.4M mesh points
- 79 zones
- 201 C90 hours for convergence
(Lift within 2% of ex:pedment)
Small geometric variations have a major impact,
particularly near maximum lift
= Grid density study was performed
Accuracy of physical modeling needs further assessment
Sttmd Rog4r¢ _ Amei - .a_T/I_D High Lift
Examples of Current Capability:
OVERFLOW-MLP Performance
/m_m_w
II
System: 512 CPU Lomax (300 MHz Odgin 2000)
: }
soF.............................. :......................................... ! ....... ;_
4Sl"......... !.......... I,.......... _.......... _......... , ....... ;->_ ......... _..........
, : ', I ,,,, : :
, , , , j ' ,
I :
Steger '
(32):)it) _,, ! " .."_i
I ,,"" _.-._<-----!
= = Lomax_4b_t)
:
3o_......... T......... i........ :.'i" -;--,r : ! = ,
| : "''"" ' ' ' ' I
!
1_........._....,.:..,¥......_.._........_........
-7 '." ,,"' .,-'" I'q'---" steger(64Ibit) I I
/ ..p." .- " : : : : : I
II _,,,,_._-_-..-_ ...... : II CPU C9_ OVERFLO_/.mlt -- -- -f- -- i ---_
o_E-_
_-¢ -i-; -i- -. _W""m"_ st,,,,,._I_.,.-,, - - =', -I
0 128 256 384 512
Number of CPUs
Examples of Current Capability:
OVERFLOW-MLP Performance
• :Dngm 2000 i64 bit) performance is dramatically better than full C90
• OVERFLOW 16 CPUC90 = 4.6 GFLOP/s
• OVERFLOW 256 CPU O2K (250MHz) = 20.1 GFLOP/s
• OVERFLOW 512 CPU O2K (250MHz) = 37.0 GFLOP/s (cluster)
• OVERFLOW 512 CPU O2K (300MHz) = 60.0 GFLOP/s
• Str_king Performance, Cost Advantage of Steger_Lomax over C90
• OVERFLOW = 256 CPUs are 4.4x faster @ 4.5x Cheaper = 23x
• OVERFLOW = 512 CPUs are 13.0x faster @ 2.6x Cheaper = 33x
• Dramatic performance gains for small changes in code
• - 1000 lines of changes (<1% of total code)
Are we done with development?
"Can do it all" message was propagated in the past, but
CFD did not replace Wind Tunnel _ CFD was oversold!
Of course, we are not done and further research will create advances with
across the board benefits;
• Algorithm
Convergenceacceleration,Robustness,Errorestimation
Grid related issues, adaptlve grids ........
• Physica! modeling issues
Turbulence,Combustion, Multiphase,Spray, Plasma etc.
• Solution Procedures
Automation: CAD-Grid-Solution-Featureextraction
• App!icat!ons
Rapidturnaroundfor complexconfigurations
Designand productdevelopment - we stillneed trainedCFDers
=>Outsourcingmakes sense
However, sponsors are likely to view these as "incremental advances."
Where do we go from here?
- Tremendous information is available
- Single-handed code development is rapidly becoming outdated (CFD
discipline as defined in me past is disappearing)
- Problem solving environment is more collaborative
Requires software engineedng to mitigate risks:
Legacy software handling tools
Visualization
Data base handling tools
PROBLEM SOLVING ENVIRONMENTS
Examples of Potential Future (or Current) Challenges
• R,s.4 _;-essment
What are the risks of designing flow devices using CFD+IT tools?
Can we manage uncertainties?
Uncertainties can from many different sources:
e.g. methods, software engineering ...
• _-_e'e _ .i _-it oc. heur!st:.c _cde!
Ca, ce-efit from Scien_fic -- E"g,,'eer:_:_; aporoac': "or _-x,s_c,e
Compute transport properties to model real gas effect
LES to predict nozzle+jet noise, maximum lift of high-lift configuration
e.g. flow+structure+combustion
Can we use LES for wall-bounded flow, if we have lOOx faster
computer today?
Do we need to invest more in LES method?
or, take different approaches?
• CFD_-tT Tools
CFD for Information generation and control (a part of IT element)
e.g. Virtual Night
Example of Current Challenges:
Integrated Vehicle Modeling Environment (IVME)
Vo_Sw, Wind Tun_ orCFD NmroP_gh otrM g Lnb fw lqitM FIgN S_akli_l am[ Virtual Ratll y
to _ahatt VtMck C_wtrol
[ A_I_,/or C_f_tMr_s_a Codrol/klt_facG [Mtlgolom
Co_¢pe & _mlllNg _ailts
L Sp=l.llcAem h_mla
Karen OLr_y-Bu_e_ _V_A Ames
Example: Real Gas Effect Model
ELECTRONIC STATE FOR N2
Euler :
- Do not require knowledge of internal
internal molecular structures and
interrnolecular potentials
' Excited Stale
Navier-Stokes :
- Molecules are structureless
"4
"L !,_ - Transport properties are based on a
single intermotecular potential
- Collisions are assumed to be elastic
1'
i' 3
° Non-equilibrium flow equations:
2
- EOS for each species is based on equil
distributions over many internal states
t
- Reaction rates account for ground states
I i
1J _,1 16 | I _.1 l_
- Empirical intermolecular potential is used
Example: Real Gas Effect Model
• Proposed Approach
ELECTRONIC STATE FOR N ATOM
Based on more accurate solution of known
microscopic equations, develop better
:=_ OOUIl,11"i ¢UA _l'rC*r$
, i rJ • r • P • ! macroscopic equations:
Dedve micro eqs and constitutive eqs
:t ""-'" " from Bloltzman eq (inelastic collision)
Obtain state-to-state rates and product-
state distribution functions
.,i • rl44
_Provide macro properties to be used in
CFD codes
•tmpact
_l -J---
',I _'wl 'g i The results are more accurate physics-
based representation of macroscopic
L2 ox_ properties
Applicable
reentry
(from current
to high-speed
/ RLV in descend
curve fitting)
planetary
Example of Data Base Management Tool:
Data Compression Using Multi-resolution
CROSS FLOW VELOCITY
• _'ting -:_; ,,':ca× Vaiida tion
NACAO012, A R---0.75, Re=4.6x106, o=10o
INS3D Code, 2.5M Grid (115xl 8.9xl 15) "' ,',. :', %=4
IMAGES BEFORE AND RECONSTRUCTED
FROM COMPRESSED DATA ARE I .,[ / : _"
INDISTINGUISHABLE ] ",', . .!
_,. !_
:7
X/C,= 1.5 "_ t _
TOTAL VELOCITY
--N
id t I _ ;
Compression Ratio :
40 (Pressure), 45 (Pressure & Velocities)
Error:. 7.93x10 "2(Max Residual), 2x10 "_(L z)
Comput_on by ;=¢.df=r ]_te=-M_t=ti ,,a ._1 o:I z n'Jl_,r
Data Corn Fmicm by D_h_ Lee
Where do we go from here?
• integrated solution for assessing the total system pedo,m'_ance, life cycle
and sa._ety can very well be the next challenge
e.g. Need a more complete picture of entire design space not just one design
Some challenges specific to CFD are:
- Physics-based simulation for more predctive capability
- Integrated analysis
e.g. multi-discipline, performance for entire flight envelope
- IT tools can be used to integrate CFD, experiments and flight tests
e.g. virtual flight
Requires : Many simulations which will be put into data base, and
data base management tools, query tools to extract desired info
• Validation ts an issue
Example: Impact of Real Gas Effect Model
Typical RLV Descent Trajectory for Aerodynamics Analyses
I11 : IV V _- Vl
Trajectory Segment
÷
Non-Continuum
T mjectory_, *_
__ Region
I i !
,? i
o I =
i
X
Continuum
Region " i
, =
' i
-_>_-,', _-"---_, Aerothermody namics : ii
1- Aerodynamics e
:
: h ;
Reacting Gas
Perfect Ga, -'_" _,,_- Equilib. i =
Equilibrium "_ Non-Equilibrium
:
|
i
I* "j i _ ' ,i Coupled Radiation
I
t
I
J
i , I , I 1 ' i
0 4 8 12 16 20 24 28
M ~ Mach Number
What are some of Target Problems?
] IIIII
• Bottom _ine for researc_ iS "moc, ey"
• We can _arget some of the dnso!ved challenges in flow devices
- Compressor rotational stall
- Turbepump system in rocket engine
- Jet noise
- Maximum lift of high-lift system
Rotor-based propulsion system
• T-ere are a w:de range of cha;ieng{ng applications ,n nomaerr)space
Climate prediction
Flow-related problems in human body; e.g. heart, lung, hemodynamics ....
- Automobile
- Naval hydrodynamics
Chemical engineering
Example of Target Problems:
Rotor-Based Propulsion System (Army AFDD)
ii i|
_SSUeS:
High Cycle Fatigue: Unsteady loads associated
with rotor-based propulsion systems are the
primary driver of high cycle fatigue of system
components.
Whirl-Flutter: Interaction between structural
dynamics of wing and rotational motion and
vortical flow of propulsion system can lead to
catastrophic structural failure.
V-22 TUhrotor • Potential !mpacl:
High Cycle Fatigue: High fidelity simulation
and analysis capability for aero-elastic effects for
propulsion systems.
Whirl.Flutter: Confu'm existing theory or define
improved design standards.
Bob Mc.a.kir_ Army Ab"DD¢ Na_ Ames
Where do we go from here?
i i
We need the next level of BIG CFD goals.
Where do we go from here?
Potential Topics
• Tough Problems:
Physics-Based Scientific Computing + CFD
• Big Impact on Aerospace Engineering :
for Developing 3rd Gen RLV
_/__,,_ r,_ Strategy
