Application of Simulation, specifically Lap Time simulators and sensitivities

[Bazanaius]

Hey All,

I've been thinking recently of all the various decisions that are made on FS cars at a conceptual stage, and the reasoning behind them. Often Rules-of-thumb, trend analysis from competitions, or copying from other teams generate numbers for wheelbase, track, CoG height, tyre choice, steering ratio, etc. etc. and having done all of the above myself (ssh!) I know that I dreaded being asked why I'd chosen them, as I'd have to respond with my most dreaded answer to any question ever (apart from 'ewww, no I will not, get out!'):

'It worked last year'.

I know that two of the answers to generating these nos. are testing and testing, along with iteration from previous years, but given the incredibly short development time of an FSAE car, and the hideously small amount of actual track time that most teams get, along with the high turnover of members, I wonder how much effort goes into simulation tools to try and predict at least an 'ideal' concept/setup for your cars?

I know there are teams out there that have been at this game a lot longer than ours, and who have far greater knowledge of the competition and what it takes to be successful, and so I thought I'd start a post to discuss what people have done and what methods you might use to exploit simulation tools to your advantage when it comes to both developing a good car, and answering those dreaded questions in Design Judging.

Of course, all simulations should be taken with a healthy dose of perspective in terms of what you're simulating, how, and what your results actually mean when compared to a real life car, but I see simulation as a vital tool in a world that is looking to cut cost and development times way beyond a cash-strapped FSAE team.

Since I've graduated I've been lucky enough to work where there is a lot data acquisition and testing going on, particularly through rigs and dynos, but it has always been used to feed simulation accuracy as well as diagnose performance.

I'm particularly interested in laptime simulations, as this is probably where I've had most experience and what I think offers greatest scope for establishing general relationships and sensitivities to what I'd call 'conceptual variables' like the numbers I mentioned above (compared to say FEA, or CFD).

Do people use laptime simulators, like lapSim, RaceSim, piSim or other products (or even written their own) to establish an 'ideal' performance of their car? Have people done much work in terms of varying key parameters of their car to see the effects? Could this be a useful tool for establishing which areas of a car need most attention, in terms of lowering weight or lowering CoG or moving roll centres or whatever you might be interested in?

Obviously as with all simulation garbage in = garbage out, and I imagine most simulations will be specific to a particular car (depending on the amount of work you want to put into your model); so where in the design loop do people use their simulations (i.e. do you review last years car and iterate on that, or at the start of the new car work with a simple model, or simply update the model as you go along)? And how much testing goes into providing accurate inputs to the model (beyond design values, one can use these models to establish sensitivies to installation stiffness, or chassis stiffness and give an informed answer when asked 'how stiff does your chassis need to be?' - this can then be backed up with bench testing, which is cheaper and can provide more detail than track testing)?

Woah, so a bit of a brain dump, but hopefully someone will decipher it and respond with their thoughts! ;-)

B

[Bazanaius]

Hey All,

I've been thinking recently of all the various decisions that are made on FS cars at a conceptual stage, and the reasoning behind them. Often Rules-of-thumb, trend analysis from competitions, or copying from other teams generate numbers for wheelbase, track, CoG height, tyre choice, steering ratio, etc. etc. and having done all of the above myself (ssh!) I know that I dreaded being asked why I'd chosen them, as I'd have to respond with my most dreaded answer to any question ever (apart from 'ewww, no I will not, get out!'):

'It worked last year'.

I know that two of the answers to generating these nos. are testing and testing, along with iteration from previous years, but given the incredibly short development time of an FSAE car, and the hideously small amount of actual track time that most teams get, along with the high turnover of members, I wonder how much effort goes into simulation tools to try and predict at least an 'ideal' concept/setup for your cars?

I know there are teams out there that have been at this game a lot longer than ours, and who have far greater knowledge of the competition and what it takes to be successful, and so I thought I'd start a post to discuss what people have done and what methods you might use to exploit simulation tools to your advantage when it comes to both developing a good car, and answering those dreaded questions in Design Judging.

Of course, all simulations should be taken with a healthy dose of perspective in terms of what you're simulating, how, and what your results actually mean when compared to a real life car, but I see simulation as a vital tool in a world that is looking to cut cost and development times way beyond a cash-strapped FSAE team.

Since I've graduated I've been lucky enough to work where there is a lot data acquisition and testing going on, particularly through rigs and dynos, but it has always been used to feed simulation accuracy as well as diagnose performance.

I'm particularly interested in laptime simulations, as this is probably where I've had most experience and what I think offers greatest scope for establishing general relationships and sensitivities to what I'd call 'conceptual variables' like the numbers I mentioned above (compared to say FEA, or CFD).

Do people use laptime simulators, like lapSim, RaceSim, piSim or other products (or even written their own) to establish an 'ideal' performance of their car? Have people done much work in terms of varying key parameters of their car to see the effects? Could this be a useful tool for establishing which areas of a car need most attention, in terms of lowering weight or lowering CoG or moving roll centres or whatever you might be interested in?

Obviously as with all simulation garbage in = garbage out, and I imagine most simulations will be specific to a particular car (depending on the amount of work you want to put into your model); so where in the design loop do people use their simulations (i.e. do you review last years car and iterate on that, or at the start of the new car work with a simple model, or simply update the model as you go along)? And how much testing goes into providing accurate inputs to the model (beyond design values, one can use these models to establish sensitivies to installation stiffness, or chassis stiffness and give an informed answer when asked 'how stiff does your chassis need to be?' - this can then be backed up with bench testing, which is cheaper and can provide more detail than track testing)?

Woah, so a bit of a brain dump, but hopefully someone will decipher it and respond with their thoughts! ;-)

B

[exFSAE]

Simulation is certainly used extensively in pro motorsport for chassis and setup development. Writing the actual sim code and getting outputs is easy.

Specific approaches, and what you pull out of it... is the 'good stuff' and generally held pretty tightly by those who use it.

[scott_rfr]

exFSAE I think that is exactly what this post is getting at. Maybe you could drop a few hints, not looking for answers here but hints are always nice. Bill Cobb is good at dropping a hint that will send you off an a tangent to try and figure out for some time....

Also I'm sure laptime sim are great but at fsae level basic steady-state skid pad works wonders to. Im sure that same thinking can get you pretty far at the pro level too.

Another thing you don't see discussed here to much is damper analysis. This is another area that is kept very tight lipped in the motorsports community. What we did here this year was write a 7 dof vibrations model and through some inputs and look at FRF (frequency response functions) for roll, pitch, heave and contact patch load. Run a DoE and come up with a new baseline. What do other do for damper analysis? Yea we all know linear/linear dirgressive/linear and .7 bla bla bla. What about other tools used to put a scientific approach on it? I feel like the 7dof and looking at FFT is just the start and there's a lot left in there to look at or try.

One thing I would be interested in that is semi off topic is what to do with limited tire data or no flat track data. There is an article in this months RCE about that exact thing.

Scott
Rutgers SAE

[BillCobb]

I resemble that remark..
Most Teams I've interacted with use sim way up front before any metal is cut. Car topologies are delt with parametrically: Jacobian factors are studied to optimize (minimize) interactions and problems with closed loop control. Open loop testing to validate the models is done as time permits and mule vehicles appear. This addresses driver training, too. Teaching a driver to diagnose a car's situations is 1/3 of the testing plan. There are a few who go all out and try to do it in ADAMS using templated models.

The GM Racing guys were getting a lot out of the full blown version of LapSim. They have produced track maps, tire testing schedules, datbases loaded with K&C data and powertrain maps. Carsim is another tool in use.

The successful record setting attempt by Corvette at the Nuerburgring was done in sim before it was run on the real car. I have the 'Ring track maps around here somewhere. The run in Germny was just a validation of the sim and Bingo: FTD.

There is a PhD thesis around which provides some very good insight into the science of computing and driving for laptime, has sims for it, and even the effects of mass distribution, etc. as published results.

Not sure where I'd post a copy. You can buy it for a few quid.

[Adambomb]

I've also been giving simulation a lot more thought lately. Of course, generally this tends to be more like "gee, I bet I could take my 10 DOF vehicle dynamics model, beef it up a bit more, then maybe make some sort of autopilot to drive it around a representative track." However, I have done this enough to know that it ain't that easy. For some reason I always seem to forget you can buy lap sims that at least have that much of the development wrung out by someone who knows a little better what they're doing.

I know there's no replacement for physical testing, but simulation is better than little/no physical testing (which is where we are now). And physical testing is a lot of work (not to mention expensive), and always comes second to just plain getting the thing reliable and "right" from a seat of the pants first guess. Besides, once you get a simulation set up you can make so many iterations so quickly, at which point physical testing becomes more of validation than experimentation, meaning ideally you would end up doing less of it (and saving $100s on tires). I don't know, that's just where I see it going.

I am curious to what is being done elsewhere.

[BillCobb]

Here's the Laptime Thesis info:

"On Minimum Time Vehicle Manoevuring: The Theoretical Optimal Lap:, Casanova, D, PhD Thesis, School of Mechanical Engineering, Cranfield University. November 2000

ABSTRACT
This work is a research on the minimum time vehicle manoeuvring problem, with a
particular application to finding the minimum lap time for a Formula One racing car.
The proposed method allows to solve the general problem of evaluating the vehicle
lateral and longitudinal controls which yield the minimum time required to traverse a
lap of a circuit.
The minimum time vehicle manoeuvring problem is formulated as one of Optimal
Control and is solved using mathematical programming methods. Novel techniques are
employed to solve the resulting non-linear programming problem which allow to
achieve effective optimisation with satisfactory accuracy, robustness and computational
efficiency. Particularly, the proposed solution strategy is generally applicable to any
arbitrarily complex vehicle mathematical model.
Car and circuit models are set up, and the optimisation program is applied to
investigate the sensitivity of the vehicle performance with respect to vehicle design
parameters, such as the yaw moment of inertia, the total mass and the weight
distribution. Furthermore, the minimum time manoeuvring problem is solved for very
different vehicle configurations. The optimisation program accurately quantifies the
vehicle performance in terms of manoeuvre time, and the nature of the optimal solution
is shown to be always in excellent agreement with the dynamic properties of the vehicle
model.
A part of the work is devoted to the development of a strategy to obtain an initial
estimate of the racing line and of the vehicle lateral and longitudinal controls to be used
at the start of the optimisation. Two algorithms to compute the racing line using on
board measured data from the real car are presented. A new mathematical model for the
vehicle steering control is derived. The model uses multiple preview information of the
intended path. Its structure derives from linear optimal preview control theory, but it is
adapted to deal with non-linear vehicle operations arising from the inevitable tyre force
saturation in vigorous manoeuvring. The excellent path following capability of the
model is demonstrated by solving various path following tasks involving moderate
manoeuvring and racing speeds."

Howzzat?

[carbon_black]

A quick search on google scholar turns it up, on a site called Scientific Commons.

From the Scientific Commons site:
<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content"> ScientificCommons.org is a project of the University of St.Gallen (Switzerland) </div></BLOCKQUOTE>

So I presume they have the legal right to distribute it.

Thank for the tip Bill. This is definatley something I want to pursue (I've just finsihed my own tire modelling programs, so I'm keen to get stuck into some simulation stuff. Finding direction is hard though!).

[scott_rfr]

https://dspace.lib.cranfield.ac.uk/handle/1826/1091

Lots of others around there if you look.....

[exFSAE]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by scott_rfr:
exFSAE I think that is exactly what this post is getting at. Maybe you could drop a few hints, not looking for answers here but hints are always nice. Bill Cobb is good at dropping a hint that will send you off an a tangent to try and figure out for some time.... </div></BLOCKQUOTE>

It all comes down to what you want to get out of your project. Define your objective, and everything else falls into place. Sometimes "lap sim" software is the answer. Sometimes its not particularly valuable and is a waste of time. Even in racing.

Are you interested in absolute, or relative measures? If the former.. you're in for a surprise.

[Adambomb]

Yes, that is VERY helpful. Thanks again, Bill!

[Adambomb]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by exFSAE:
Are you interested in absolute, or relative measures? If the former.. you're in for a surprise. </div></BLOCKQUOTE>

For me definitely relative. If nothing else just because I already have witnessed enough variability (temperature dependency, etc.) in TTC data under high-dollar, well-controlled experiments run by really smart, experienced people that I have learned not to expect any real "absolutes." Not to mention the complete lack of repeatability when running a skidpad on, say, a beautiful concrete airstrip compared to the wavy, cracked, sandy parking lot we end up doing most of our on-campus testing at.

Mostly just want to see, in an "ideal" scenario, how a track width of X compares to Y, how much aero could/could not help, etc.

One other thing I like about lap sims is it takes the driver out of the loop...which I imagine is important when you are hard pressed for really good, smooth, consistent drivers. It also removes other variables like the amount of sand and broken glass on the parking lot, etc. http://fsae.com/groupee_common/emoticons/icon_biggrin.gif

[Chris M]

This PhD thesis might interest you (and it's free!). It's about simulating the performance of an FSAE car to maximise competition points (rather than laptime). The first couple of chapters are about genetic algorithms, the last few are about applying it an FSAE car.

http://theses.library.uwa.edu.au/adt-WU2009.0018/

It was written after the author had left the team, so the simulations he constructed aren't actively used by UWA.

Chris

[Crispy]

I used Interlibrary Loan to get the Casanova PhD thesis, I think most university libraries should have access to this service (awesome for getting papers).

[Bazanaius]

@scott - spot on, I thought I'd start the thread so people can discuss what they've done, etc. etc. in an FSAE context.

The thesis from BillCobb looks interesting, certainly a good starting point if people are looking to write their own sims. Before I graduated I did some work with RaceSim in my final year project, but I also know people who effectively wrote their own from scratch to similar effect. The playoff is time available and the complexity of the model that is required, vs a much deeper understanding of what your model is (and isn't) telling you.

I cannot agree more on the 'absolute vs relative' point - you'll never get absolute numbers that will correlate with the track performance from a lapsim, but I think it's a very useful tool for establishing sensitivities to various car (and even track) parameters. This should arm the FSAE designer with plenty of ammo when it comes to justifying their decisions, and assigning resources to development projects.

[ben]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by BillCobb:
I resemble that remark..
Most Teams I've interacted with use sim way up front before any metal is cut. Car topologies are delt with parametrically: Jacobian factors are studied to optimize (minimize) interactions and problems with closed loop control. Open loop testing to validate the models is done as time permits and mule vehicles appear. This addresses driver training, too. Teaching a driver to diagnose a car's situations is 1/3 of the testing plan. There are a few who go all out and try to do it in ADAMS using templated models.

The GM Racing guys were getting a lot out of the full blown version of LapSim. They have produced track maps, tire testing schedules, datbases loaded with K&C data and powertrain maps. Carsim is another tool in use.

The successful record setting attempt by Corvette at the Nuerburgring was done in sim before it was run on the real car. I have the 'Ring track maps around here somewhere. The run in Germny was just a validation of the sim and Bingo: FTD.

There is a PhD thesis around which provides some very good insight into the science of computing and driving for laptime, has sims for it, and even the effects of mass distribution, etc. as published results.

Not sure where I'd post a copy. You can buy it for a few quid. </div></BLOCKQUOTE>

https://dspace.lib.cranfield.ac.uk/handle/1826/1091

[flavorPacket]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by Chris M:
This PhD thesis might interest you (and it's free!). It's about simulating the performance of an FSAE car to maximise competition points (rather than laptime). </div></BLOCKQUOTE>

We have worked extensively on a similar project. We developed our own LTS models and designed a series of sims that correlate with the dynamic score at competition. Then we run DOEs to determine the parameters we believe the model can be trusted to simulate with good fidelity. I'd like to echo BillCobb in recommending running open loop sims. You can really lose yourself in the 'chicken or egg' debate if you only run closed loop sims.

For the things that the model cannot handle (nonlinear/hysteretic dampers and tires), we try to use simpler models and go after what we think are good metrics. Scott, FRFs are definitely the way to go, but the quality of your drive file (your input) and your analysis tools plays a huge role in what you see.

In my current job, testing is all but banned, so we rely almost entirely on simulation. Rigs for cars, rigs for car parts, models of cars, models of rigs, etc.

For FSAE it's not worth it to get into such depth. The marginal benefit of improving your tire load variation by 5% is small compared to the marginal benefit of getting the car built one week sooner by not spending time building and correlating some model. That's the hard thing to simulate...

[scott_rfr]

flavorPacket the reason you give of getting the car done is an example of some of the ways we are doing things on a 2 year plan right now. I won't be around next year but hopefully the guys follow through. We spent a lot of time getting what we think is a good base down after last years car. The plan is to keep pretty much the same car next year and we even have some parts made for 2011! The down side is we have basically 1 month of testing this year. But this allows for the build cycle to start much later and end earlier next year. Giving the most amount of time to simulate and test with this current base model. Seeing as most of the parts will remain the same we can test lots of setups and start to get some validation out of the models.

So what are some other metrics that are used for evaluating simulation outputs? We have beat to death a lot of points on this forum already.

Scott

[exFSAE]

<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by scott_rfr:
So what are some other metrics that are used for evaluating simulation outputs? We have beat to death a lot of points on this forum already.

Scott </div></BLOCKQUOTE>

Unfortunately, metrics that 'work' in pro motorsport aren't generally shared privately, much less publicly. Plus, everyone has their own opinion and experience on what may be most relevant in a given circumstance. Some things are more basic and obvious than others.

Again.. just start at your objective and work it back.

Where are you going to focus your effort? Skidpad? Accel? Auto-x? Are the objectives for each mutually exclusive?

Let's say its auto-x. How would you characterize the track? Where are you going to gain the most time? Hairpins? Sweepers? Straights? Slaloms? For any of those.. what's gonna give you the most speed? What do you want to maximize? What do you want to minimize? How do you put numbers to those parameters?

While some parts of vehicle dynamics are well-established... others are still very much open to interpretation.

[Silente]

For some things, if you want to simulate you can use simple stuff, not necessary a whole lap simulation.

For others things i think the whole lap sim is the only way to go. Working on some simulation softwares I have seen one of the biggest problems in obtaining useful results is how you simulate the driver.

Normally in steady state sim you have a driver who is always at the car full potential, but sometimes you can set something like an aggressivity, which change his beaviour above all in brakings and this can have an influence on the results.

On the other hand, what do you think of simulations where the driver is actually the guy who drives the car also in the sim? Like driving simulations such as GTR2, Rfactor etc but obviuosly thought at an higher level of simulation. For what i know and for what media say this is the way a lot of F1 constructors have choosen in the last years...

[BillCobb]

I believe the simulators have a role to play, but not to train a driver. In fact, you can screw up a driver really well by exposing them to a sim and then shoving them into a car on a real track.

This is because the sim transient responses are still lacking fidelity. Not because the computer is too slow, but because the visual queues are not in proper phase. Much of this is becauee the screen response is still too slow.

From my direct experience, a very high performance racing car can be driven by a novice in a sim and they will become quite good at operating it. They will shine on the score sheet because they believe that's what an Cup car or F1 car is really like. But, when you put a Buick driver in the loop with a Buick sim, their smile fades pretty fast. Sure the car goes to the right when they turn right, and the amount of response is also right (even when turning left), there will be steering torque feedback and that can be approximately the correct magnitude. But, they will all agree its not a Buick. The visual response, and the steering torque/effort response is wrong with the screen queues. Much of this has to do with how the model is excited by means of steer angle inputs. In real life, a vehilce is operated in closed loop with a human driver by a moment input. The steer moment is reacted in a sequence of events involving the steering mechanism (including power assist reaction dynamics), tire mechanics transient response, and feedback up to the driver. The vehicle goes someplace but not to a specified path, but on a trajectory resulting from the moment balance reaction.

No better way to see this than to put small steers into a car while restrained laterally on a chassis dyno. By computing an equivalent front lateral force from the restraint(s), you can then not only see the response develop, but watch the sequence(s). I have some video of this around here somewhere from past experiments. When an angle signal is used, the response are very different than those produced by a moment input. And, the reaction forces have different phase response.

This is a good science project for those of you looking for an R&D thesis. For those of you who believe you have experienced a 'real' F1 car by driving a sim, the real car will actually be much easier to drive. But, have them throw a Buick in there and share your disappointment. You would expect the Buick to be easier to sim because the transient response are so much slower. In fact, the results will be more disappointing to those who have real driving experience.