Sie sind auf Seite 1von 41

FSAE Aerodynamics:

Using your time wisely


Riccardo M. Pagliarella
Riccardo@AeroONE.com.au

James Slaughter
js206@uowmail.edu.au

With special thanks to Joel Kennedy (Altair)


The Basics
 Know what principles drive good aero performance
 Know your compute resources
 Know your current and achievable skill levels
 Know what reasonable performance targets are
 Know what reasonable performance sensitivity is
 Know how to design tests for high-value outcomes,
and
 Design what you can actually build
The golden rule

There are no silver bullets in computational methods

Just understanding and better application


What is good aero?
 ‘Usability’ >> ‘big numbers’

 CoP does what you need

 Vehicle performance can be trimmed

 Performance has minimal critical sensitivities

 Changes have good sensitivities


Stuff you should care more about
 Consider making your designs more amenable to clean,
effective meshing

 Not using high-accuracy volume-based CFD methods for


every stage of your design process

 Being able to explain your methods, no matter what


they are

 Building your own HPC machines or learning to use cloud


resources

 Extracting significant value from result sets


Stuff you should care less about
 Copying other teams/cars. Most FSAE aero is
conceptually poor, even the winning cars. This is not an
insult – you’re students – be original.

 Copying F1 design or practices

 Headline numbers

 ‘Having to use’ high-resolution volume CFD methods

 High run rates


Know your compute resources
 RANS CFD on a full car in yaw typ. 40M+ elements. This is
super difficult to get right. Runs take a while even on a fast
machine, plus significant pre/post processing time.

 Reasonable results possible on 64GB RAM. 128GB preferred.


Dual-CPU Xeon class machine preferable. Needs cores &
memory bandwidth.

 Meshing typ. single-core activity. Needs speed = different PC.

 Visualization different again. Resources depend on package.

 Your aim is to build a successful workflow.


 Many hands make light work here
 Learn to multitask: watch your utility rates!
 Don’t be scared to go long on IT here (e.g. KVM, NFS, etc)
But we can’t afford that!
 Consider cloud/shared resources or sponsored hardware
 Ask your university about using shared cluster resources

 Consider second-hand, last-generation resources


 Consider making the problem computationally less
intensive without decreasing relevance (e.g. is panel
method in yaw more/less relevant than a straight-line
volume-method simulation at high resolution? Some of
the time? All of the time?)
 Never, ever confuse accuracy with resolution
 Remember – smart people designed successful racecars
on resources you don’t have – so can you
But we can’t even afford that!
 You can run panel method on Atom-class hardware.

 Go experimental, CFD is not a prerequisite. This does


not necessarily involve a full car.
Panel methods
 Non-volume based CFD method

 Useful for early concept analysis due to quick turn-around

 Some examples:
 PANAIR - Boeing and NASA
 Xfoil
 XFLR5
 NEWPAN

 Professionals use these for a first cut – just beware of


limitations around usefulness

 Other ‘first cut’ methods exist. Up to you to find them!


Do people overdo it?
 Short answer - Yes.

 Early stages of design do not require volume-based


RANS methods. It’s overkill. You’re wasting time.

 Once a concept is proved out, save big CFD for big


changes. Incremental work is best done experimentally.

 More runs ~= more gains.


 Smart runs == more gains.
Skills - am I awesome enough?
 Probably not! But this is a learning competition.

 Cars are hard

 If you’re not 3rd to final year mech./aerospace, you


should not be attempting 3D RANS volume methods
 Yes, you will get some results…
 …no, no one’s been successful in producing useful results…
 …and no, you will not the first to fluke this.

 This doesn’t mean ‘no CFD’ – it means ‘pick the right


computational methods for your skill level, and grow
your understanding with it’ – and you’ll be successful
Good performance (i)
 Increased Fy at the tyres - all of them, in the right way,
at the right time.

 Stability vs instability / CoP and CoG ‘dance well’

 Good transitional behaviours – EoS, in corner,


accelerating out, etc

 Thermals well managed, no silly drag

 Aerodynamic development activities that reflect the


above

 (What’s the elephant in the room here?)


Good performance (ii)
 Some of these outcomes drive aero packaging very
acutely – e.g. think about how to reliably achieve high
wing load with favourable CoP height?

 Know when to stop designing and start optimizing -


have clear goals, or you will never finish

 Know how to appraise gains – a challenge:


 Hypothetical car 1: CD=0.8, CL=-1.8, %Fr=40%
 Hypothetical car 2: CD=0.9, CL=-1.9, %Fr=38%
 Which one is better? (Can you read results ‘on the slope’?)
 What are the mitigating factors in your analysis?
“On the slope”
-2.40

-2.35

-2.30

-2.25
CL/CD

-2.20

-2.15 • Use wing flaps to trim CoP


• Slope limited by wing stall or
other behaviours – up to you to
-2.10 understand these
• Higher slope = gains
-2.05 • Strength depends on validity of
headline data!
-2.00
35% 36% 37% 38% 39% 40% 41% 42% 43% 44% 45%
%Fr
Hypothetical car 1 Hypothetical car 2

CD CL CL/CD %Fr
Hypothetical car 1 0.8 -1.8 -2.250 40.0%
Hypothetical car 2 0.9 -1.9 -2.111 38.0%
*This is not real data – that much is up to you!
What Not To Say… (i)
 Not knowing what’s y+/inflation layers are, or suggesting they
were too hard to use properly

 Straight line runs only

 Not knowing what reasonable lift/drag numbers are

 Not knowing anything about your CoP


 And not knowing the ‘trick question’ backwards!

 Not knowing why you chose a turbulence model or numerical


method (esp. beyond what someone ‘told you to do’)

 Poor domain boundaries/setup

 Not thinking aero is important!


What Not To Say… (ii)
 CoP significantly higher than CoG
 Ahem, much pitching moment then?

 Only analysis was done using RANS techniques


 What are the pitfalls/limitations of RANS techniques?

 Lack of understanding of transient behaviours

 Resolution in results exceeds accuracy of methods

 No appreciation of thermal management


 Thermal loads are a major cause of attrition in SAE
 Drag due to cooling appendages is also significant
component of overall vehicle drag
Performance sensitivity (i)
 A top F1 needs many, many counts of downforce to
notice a difference or aerodynamic improvement – your
drivers are not top F1 drivers, you will need more

 Your error margins probably afford less accuracy in


results from small changes than CFD resolution allows

 Make big, glorious changes – try new things! Small


changes for fixing critical flow structures only
Performance sensitivity (ii)
 Know what vehicle attitude does to performance –
some deltas are very significant (e.g. FW)
 Does limiting/constraining this improve aero outcomes?
 Big no-no’s?

 On-car changes must give big performance deltas to


be at all useful
A good test looks like… (i)
 More than a single test per package with smart results
aggregation

 Sensitive stuff captured, not so sensitive stuff not

 Meshed properly. This a thousand times over. This can


mean ‘design for meshing’.

 Careful choice of turbulence model, domains, etc.

 Understanding of limitations

 Robust visualizations

 Relative results to other cases, clear future directions


A good test looks like… (ii)
 Think about what the test model needs to represent!
 Vehicle attitude
 Representative shapes & surfaces == representative flow
structures
 What’s the driver doing

 Otherwise garbage in == garbage out

 Consider putting aside 10 minutes a night to keep a


project diary of your experiences, gains and learnings
 Makes your design presentation super easy
 Makes learning and moving forwards much easier
But can you build it?
 If you can’t dance on it, it’s not built!

 Practice overestimating your resource requirements

 Consider a staged release schedule – designing it all


then releasing it all to build is a silly idea

 Consider a modular component design

 Typical aero SF=3-4 (no, really)


 Don’t need to have it designed to appraise loads – use
coarse best practices
 There is no ‘too stiff’ at this level
Even RANS has limitations
 RANS simulations only resolve some of the flow
physics – i.e. Large length scale flow phenomena
 Do not treat RANS results as absolutes!

 In reality large array of length and time scales that


interact in complex manner.

 For an example of a “real” flow field see:


NASA and Exa. SC17 Landing Gear Simulation
Vortex Ring Collision – University of Melbourne
Plan, plan, plan… (i)
 Assume:
 January start (you need holidays)
 4th year knowledge – you have taken 3 core fluids subjects and
either 1 core CFD (or CFD within fluids subjects)
 In lieu of this – trade 6 months preceding intensive part-time study in
computational methods (overload, electives, online learning) and 2
core fluids subjects - FSAE is not a place to learn what y+ is or how
to setup domains – see that you can confidently repeat what you’ve
learned in class before trying a car.
 Basic proficiency in at least one CFD package
 Student proficiency in at least one CAD package
 Basic proficiency and knowledge of limitations in at least one ‘short
cut’ method (VWT, vortex lattice, 2D, etc)
 Proficiency in theory of wing design, converging-diverging sections,
bluff bodies, ground effect
 Proficiency means a credit average or greater in relevant subjects
Plan, plan, plan… (ii)
 Extra material you will need to study up on in preceding
months:
 Basic scripting proficiency (C/Python depending on package)
 Gaining confidence in convergence criteria / knowing where to
‘deep dive’ to increase accuracy at minimal resource spend
 Vortex flows
 Critical point method (Chong & Perry)
 Handling porous surfaces in CFD (HEX design)
 Aero/kinematic interactions, intermediate vehicle simulation
(collaborative)
 Design of Experiments
 Literature review of relevant turbulence models, existing
solutions and performance – don’t confuse this with copying
 Get your compute resources sorted before January
Plan, plan, plan… (iii)
 Key drop-outs here:
 You can build a winning aero design with no previous FSAE
knowledge (previous designs can even be limiting)
 Work diligently on getting to a point where your gaps are
mostly ‘how do I do this’ not ‘what does it mean’
 Identify and exploit accessible resources and reasonable
stretch targets
 Theory first, tools second
 First-cut tools important: ‘80% of the design on 20% effort’
 Aero requires a broad skill base for success. Pace yourself.
 Pack humility for the journey – be prepared to learn and
evolve – your designs should change with knowledge growth
 Short cuts on knowledge limit your growth, your design’s
potential and ultimately your career progress – consider why
you’re here
How do I start?
 Get your CAD and mesh it!
 Watch it fail horribly
 Get your vehicle attitude in a representative state
 2D before 3D straight line, 3D straight line before yaw (think of
it all like a pen license)

 Then:
 Optimise your CAD for meshing
 Pick a turbulence model intelligently from lit review
 Work with a supervisor on better meshing, domain,
visualisation etc towards reliable convergence + workflow
 With hard, dedicated work and reliable compute you’ll be
ready to start 3D RANS development in 6-12 weeks
 Repeat for key scenarios. Summarise. This is your baseline.
What if I’m 2nd/1st year
 DO NOT ATTEMPT 3D RANS METHODS – there has not
been a single attempt in the history of FSAE to get 3D
RANS ‘right’ with this knowledge level, and you will not
be the first
 Doesn’t mean you can’t produce a winning design
though, focus there – its not a race to use complex tools!

 2D RANS, vortex lattice method permissible if 2nd-year

 ‘Set & forget’ CFD methods permissible (e.g. VWT) if


coupled with strong understanding of limitations

 Shift skill set from complex tools to developing strong


fundamentals – exploit those – and push experimental
Remember
 Despite advice here many here will persist otherwise.
Consider:
 It’s a competition, welcome to motorsport – whilst you’re
busy wasting time, someone else is busy making gains
 Your challenges are not new - “Those who don’t learn
from history are doomed to repeat it” – your solutions can
be new, so focus there
 “If you want to run fast in this, lift a lot – slowly”:
 Studying one aero package configuration per week in sufficient
detail to provide useful understanding is breakneck speed for
FSAE
 Development speed without depth is wasted effort
Stuff you should read
 “Road Vehicle Aerodynamics” (T. Schuetz) – this is the
‘god book’ that has most of what you need – latest (5th)
ed has a good motorsports section, other sections useful
 “Ali/Wings” (Benzing)
 “Race Car Aerodynamics: Designing for Speed” (Katz),
“Competition Car Aerodynamics” (McBeath) – not as
essential as the previous two
 Critical point method papers (Chong & Perry)
 Basics on aero shapes, BL methods (e.g. Hoerner,
Schlichting)
 …And as many good theses in the space as possible!
Draft, not-definitive schedule (i)
January Commence work on reliable 3D RANS. Commence concept scheming.
Commence and conclude trials with ‘short-cut’ software (SCSW). CAD
concepts, run through SCSW in isolation.
February Continue work on reliable 3D RANS. Develop loose performance
specifications and key metrics. Identify key intended flow structures
in concepts. Continue with SCSW.
March Conclude work on reliable 3D RANS – finalise methods and workflow.
Start 3D RANS development. Develop alternate whole-vehicle
concepts at concept scheme level.
April Finalise performance targets within 25%. Develop wing package in 2D
or VLM; use understanding of CoP movement to drive development.
Incorporate best results in 3D RANS.
May Update RANS model to incorporate more practical vehicle details:
mountings, powertrain, etc. Aim to release most complex aero parts
to manufacture (e.g. sidepods). Know approx aero-structural loads.
June Development continues. Aim to release wing package to
manufacturing. No more SCSW from end of month.
Draft, not-definitive schedule (i)
July Aim to release a complete car of parts by end of month – aside from
large items (sidepods, wing elements, floor main section) these need
not be final versions. RANS 3D continues to optimise flow phenomena.
August Testing, validate key flow phenomena experimentally. If found
lacking, see to understand rationale and reconcile w/CFD. 3D RANS
continues based on experimental observations w/limited envelope.
September Testing. 3D RANS continues based on experimental observations
w/limited envelope.

October Final parts released to manufacturing. All CFD from here is to explore
configuration changes only. Testing.

November Testing.

December Final testing/breaking/finalising build. Competition. Holiday, you’ll


have earned it!
Example Formula 1 workflow (i)
Epiphany.
performance No Yes Build best
Abort concept Promising? parametric into
requirement,
upgrade
plagiarism etc

No
Driver
vibe

Concept
Yes
Decent
scheming learnings? Test at 20-80
attitudes in wind Try at test
tunnel per session DAQ,
configuration process

No

Yes
Design, build Promising?
CAD Promising?
scale test parts

No
Yes
Significant
Experience
Parametric 3D higher-order Shelve it Race it
evolutions in CAD RANS sim, yaw
Example Formula 1 workflow (ii)
 Won’t work for your team!
 Not enough experience to guarantee processes
 Insufficient net resources
 Insufficient availability of resources at critical times
 Relies on driver sensitivities you won’t have

 Typically less than 1% of parts evaluated


experimentally make it to the race car
Suboptimal FSAE process (i)
Epiphany, External
problem, performance Packaging
targets
plagiarism etc

“Standard” Ambiguous Accepted


Single, ‘big’ RANS metrics, insecure
CAD visualizations and Design to build
CFD run(s) process
force data

Argued against

Can it be
Shelve it
No modified to fit?
Yes

No time!

Iterate design
Define Functional
towards Validate flow
performance checkout
See you in intent and structures Build/mount
envelope tests
December improvement
Suboptimal FSAE process (ii)
 Very typical

 Poor concept evolution, gating of resource commitment

 Fails to characterise critical dynamic behaviours

 Weak decision-making processes

 Insubstantial validation

 Time compression tends towards poor build quality


Draft better FSAE process (i) BASELINE
External DATA +
performance LIT
Research targets REVIEW
Known data-
literature: driven
theoretical + challenges
industrial
Packaging

Draw workable
concept on paper,
Discuss within Does it make Yes Is there a quick Yes
Epiphany complete with key way to an ‘80%
flow structures and team sense? answer’?
pressure fields

Argued against BS!


Change run
CAD
parameters, fix
Shelve it
No

No Did the runs 3D RANS runs per key Iterations


vehicle attitudes Evolve CAD from Yes
check out Still promising? through quick
(+pre/post learnings
perfectly? processing) sims

Knowledge
Vehicle Knowledge of method
dynamics of RANS limitations
input limitations
Draft better FSAE process (ii)
Vehicle
again if feasible

Detailed dynamics, Fluids


visualisations lap sim theory
Redesign, try

data

Looks like it
Were all Understand
Yes Process headline
idealized flow targets and Bold CAD 3D RANS as
numbers – obtain
structures opportunities for evolution before
benchmark
present? improvement

Aero loads, Design for


target
deflections manufacture
Vehicle performance
targets not met?
+ Concept
Push Yes Is there more time to design? exhausted
experimental More gains?
+
test parts Are your remaining ideas
big, bold and good
Work on a for 3D RANS?
different aero
system (watch
interactions)
GO ON… Yes
Draft better FSAE process (iii)
 Then do all the experimental work as previous

 Emphasis on thoroughness over speed

 Emphasis on decision clarity over ambiguity

 Emphasis on judicious use of resources over use of


specific resources

 This is not a complete, right or wrong way


 What you do is up to you
 You need to justify your workflow – this is just as
important as explaining the work it produces
 Better workflow >> more work - every time
Best car ever seen for aero?
 In design: “too much” design concept
 Developed by some engineers with 5+ years away from uni
 An exercise to see how far rules could be exploited
 Quite radically different but the performance targets
aren’t dissimilar
 You will never see it (don’t ask)

 In practice: a very conventional-looking car with great


principles giving very usable downforce
 Student with excellent command of CFD methods,
turbulence models, etc
 Strong aero/kinematic mix gave high real-world
performance
Final thoughts
 Focus on right tools, best workflow – and copy no one
here, it’s what’s right and best for YOUR TEAM

 We didn’t even talk about experimental, validation,


thermal and more

 Many hands make light work. This is typically the work


of a team of people – unrealistic to expect a single
student to deliver from concept to built.

 Be inclusive in design. Encourage alternate lines of


thought.

 Have fun, think lots, understand more, design bold

Das könnte Ihnen auch gefallen