Lap/point/competition simulation

[StevenWebb]

This was originally a reply to another thread, but I feel it deserves its own voice


Originally posted by Demon Of Speed:

The question I have for people who have a "lap sim" what do you feel is the minimum level needed to be able to get enough accuracy when writing a simulator (I know it is subjective, but would like opinions). So I know I am way off topic, and I will fix the spelling and grammatical errors later.

my definition and opinions for varying degrees of simulation from less "accurate" to most "accurate".



Type-A (basic, constant everything particle model)

its very very easy in excell to simulate a particle with lateral and longitudinal acceleration limits, weight and a power output limit. then see how quickly and how much energy this particle will use as it travels around a series of lines and arcs which roughly look like the racing line of your most recent competitions fsae track.

this sort of simplicity is perfect for "should we go single or four cylinder?" or even "should we spend our money on tyres which will increase our coefficient by xx or a turbo which will increase power by xxkw.

Type-B (advanced particle model)

Its a bit harder than Type "A" (i.e. 2 weekends instead of 1 day)and may need to move to matlab or similiar to make some things easier (although excell is definately possible). This step may incorporate things like downforce and drag, weight transfer during acceleration, making the racing line change with overall car width and even incorporating multiple tracks and how the particles will go if the track changes drastically (i.e. will a particle in michigan do as well as the same particle in australasia?). you can can add functions into the simulation to show sensitivities and such or other things specific to your team/competition. competition strategies can also be looked at in this level

I think this gives the best effort vs reward and i think every successful team should make one(successful being finishing 80% of their endurances). This is the right sort of level for choosing a new concept (ie giant wings and undertray or super narrow single cylinder, etc)

Type-C (Transient Bicycle-ish model)
this is where stuff starts to get real heavy. i wouldn't suggest trying this in excell. I'm not going to comment much on this because i'm not at this stage yet (but will likely be in my final year thesis) but this is where you add in yaw acceleration, non circular racing lines, torque curves, f/r weight distribution, tyre load sensitivity, combined long/lat tyre coefficients, slip ratio, etc.

this is the sort of sim you use for concept refinement. gear ratios, weight distributions, track widths, maybe even weight transfer distribution and diff bias ratio. combine this with some basic tyre data and it could be really useful for the fine tuning of your cars overall geometry

Type-D (full 4 tyre transient model)
full transient model, all 4 tyres, iterations of different racing lines too see which is better, load variations over track bumps. full track surface modelled, springs dampers and unsprung weights, aero variation in pitch and roll etc etc etc. this needs a hell of a lot of data to get anywhere near where you want it to

In my opinion this type of sim is a waste of time. The accuracy demanded from a simulation like this is thrown out the window when you put someone behind the wheel or a chassis is welded half a mm off-square. This sort of sim is only useful if your formula would prohibit test time (and if it does, you're likely to just buy a simulation package anyway)

Type-E
this "sim" involves building every possible car and driving them on every single possible track. you need a very well trained driver (preferably robotic), a stopwatch, and a fuel measurement cylinder.

this is just referred to as "testing" by the majority of us. on track testing with your endurance drivers with a highly adjustable car will always be the best way to squeeze a last tenth out of the car.



The Accuracy required depends on what your trying to get out of it.
Generally the more substantial the variable, the less accurate the sim needs to be. (eg weight needs a very simple sim to see the effects. tire pressure need a very complex sim(or testing))

At the moment we are using a "Type b" sim which helped us define our concept for 2011 and help us with design choices. after seeing how the car drives i will work on a Type-C sim to refine certain aspects for 2012 and beyond.

some notes...

always compare apples with apples
think about your local competition and rules
comps are won by points gained over the competition, not how fast your car is
an adjustable car is always a better idea that an overly complicated sim.
a sim that says you will gain 1 point in autocross over another car will not help you finish endurance

I have obviously a lot to say on this topic so i'd like to see what sort of simulations other teams are using

[thewoundedsoldier]

Great post. I really like the way you put it when you say "always compare apples to apples". Building a simulation which truly and equitably compares a 150kg, single-cylinder car to a 180kg, 4-cylinder car is nearly impossible. This is because it needs to compare massive differences among the cars in a fair way. It's best to understand the input and output of every simulation you run, and limit your independent variables so that you can truly see what your dependent variable change means.

As for the original question, I would add that accuracy in any simulation is not the be-all-end-all as long as you understand the inaccuracy. Know the difference between qualitative data and quantitative data.

I always love threads about simulation. Let's hear it, people!

[exFSAE]

I would say you're missing some things between B and C.

[Charlie]

Originally posted by thewoundedsoldier:
Great post. I really like the way you put it when you say "always compare apples to apples". Building a simulation which truly and equitably compares a 150kg, single-cylinder car to a 180kg, 4-cylinder car is nearly impossible. This is because it needs to compare massive differences among the cars in a fair way.

I couldn't disagree more. These are not 'massive' differences in cars. As a whole the FSAE car configured either way is fundamentally a very similar machine.

If a 20% increase in mass and/or a 20% difference in horsepower in a simulation is 'near impossible', then god help the simulation world because they are damn near useless.

[BillCobb]

C.5 would be a 4 wheel nonlinear tire model with a quasi steady state solution technique. It's Bosch Lapsim, and work very well. Its the ONLY one worth developing because any racing machine requiring heavy transient analysis needs a new driver.

I posted a Matlab Sim on TTC which runs ISO procedures appropriate for open loop testing. That keeps your bumpkin driver from obfuscating preliminary design settings. If a car is NFG without a driver, adding a driver (even a Great driver) sure isn't gonna fix it). Ain't that right Mario?

Get the Math right first then add complexity in small steps. Back it up with theory. Cornering first, then add braking and launch. Pick appropriate procedures, especially ones that you can run in a parking lot. Walk first, then trot, then canter, then gallop. Otherwise, drill a lot of holes, buy a good air compressor and quit when you run out of time.

[Thrainer]

Originally posted by Charlie:
...
If a 20% increase in mass and/or a 20% difference in horsepower in a simulation is 'near impossible', then god help the simulation world because they are damn near useless.

I'm not convinced that the crucial differences between the two concepts are just mass and power. Since both concepts compete on a similar level, the performance difference doesn't seem to be extremely huge. I reckon that effects like different tire warming and wear, car width and handling might tip the scales, but are "nearly impossible" to simulate.

Derived from this, I'm wondering what amount of points difference you guys see in a competition simulation between a ~150 kg single and a ~200 kg four? Or generally, what points difference makes you opt for a certain design (aero, turbo, E85 etc.)?

Regards,
Thomas

[Charlie]

Originally posted by Thrainer:
<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by Charlie:
...
If a 20% increase in mass and/or a 20% difference in horsepower in a simulation is 'near impossible', then god help the simulation world because they are damn near useless.

I'm not convinced that the crucial differences between the two concepts are just mass and power. Since both concepts compete on a similar level, the performance difference doesn't seem to be extremely huge. I reckon that effects like different tire warming and wear, car width and handling might tip the scales, but are "nearly impossible" to simulate. </div></BLOCKQUOTE>

Any simulation has limitations and it is easy to bring up them and say that it makes the simulation less accurate. However, it is impossible to prove that any of these relatively small details actually have an impact, without some serious testing. In my opinion a basic simulation will show you 90% of what you need to know in terms of car configuration.

The fact that the concepts 'compete on a similar level' is irrelevant. Team knowledge, preparation, testing, and execution has been proven to be vastly more important than car configuration in FSAE. It doesn't prove one way or another that a specific concept is best if it is competitive.

[flavorPacket]

Originally posted by BillCobb:
C.5 would be a 4 wheel nonlinear tire model with a quasi steady state solution technique. It's Bosch Lapsim, and work very well. Its the ONLY one worth developing because any racing machine requiring heavy transient analysis needs a new driver.

This is splitting hairs, but a vehicle that 'looks good' in a couple of open loop ISO procedures is not necessarily the best for amateur FSAE drivers who essentially keep the car in transients all the time.

Another thing to consider is how your parameter sweeps are confounding your results. When your wheel base changes 10 inches, are you keeping the yaw inertia the same (etc)? Some of these overlooked parameters actually display higher sensitivity than the 'important parameters' for certain vehicle configurations.

[BillCobb]

Steady state gains and cornering compliances plus mass distribution define the transient properties. You should know analytically what those properties ought to be from Systems Engineering fundamentals. That includes non-linear behavior. Only 6 parameters define a car: inertia to mass ratio (k2/ab), front cornering compliance, rear cornering compliance, steering gain, wheelbase, and speed. A couple a very nonlinear. Include steering effort and steering torque gradient and you have a completely represented car. Then you put a driver in it and add some Ax characteristics. The Published World knows what a 'driveable car' car is (and what it isn't). No need to say its beyond most teams. All it takes is ONE team to figure it out. They get the job offers.

[cwatson]

7 parameters.

What about downforce? http://fsae.com/groupee_common/emoticons/icon_wink.gif

[BillCobb]

All downforce does is modify the cornering compliances via higher tire loads and suspension position dependent parameters. There can be some yaw angle dependent aero moments, but nobody is demanding 100% simulation coverage. Some tunnel data would give you the function's magnitude vs. polar angle. Just add it to the moment equations.

[flavorPacket]

Originally posted by BillCobb:
Steady state gains and cornering compliances plus mass distribution define the transient properties. You should know analytically what those properties ought to be from Systems Engineering fundamentals. That includes non-linear behavior. Only 6 parameters define a car: inertia to mass ratio (k2/ab), front cornering compliance, rear cornering compliance, steering gain, wheelbase, and speed. A couple a very nonlinear. Include steering effort and steering torque gradient and you have a completely represented car. Then you put a driver in it and add some Ax characteristics. The Published World knows what a 'driveable car' car is (and what it isn't). No need to say its beyond most teams. All it takes is ONE team to figure it out. They get the job offers.

I think we're saying the same thing. Yes, those parameters are necessary. They are also largely unknown to FSAE teams. I don't know of more than a handful of teams who actually know their vehicle's yaw inertia. That number drops even further when including steering gain, compliances, etc. Hence my original point, which is that FSAE lapsims are likely to produce trends that are not in line with theory or practice if the sim writer isn't very careful.

[thewoundedsoldier]

Originally posted by Charlie:
Team knowledge, preparation, testing, and execution has been proven to be vastly more important than car configuration in FSAE. It doesn't prove one way or another that a specific concept is best if it is competitive.

I think you've come full circle to agree with my point. I was not suggesting that the two vehicle types should be represented only by the differences in mass and power. On the contrary, I am pointing out how different the cars are from the standpoint of intangibles like reliability, servicability, etc. Can't really model those, at least not fairly (i.e. not without bias).

Much better to focus on the tangible, and limit the variables you seek to affect. This is not the same as "over-simplifying" your model.

[Charlie]

Originally posted by thewoundedsoldier:
<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by Charlie:
Team knowledge, preparation, testing, and execution has been proven to be vastly more important than car configuration in FSAE. It doesn't prove one way or another that a specific concept is best if it is competitive.

I think you've come full circle to agree with my point. I was not suggesting that the two vehicle types should be represented only by the differences in mass and power. On the contrary, I am pointing out how different the cars are from the standpoint of intangibles like reliability, servicability, etc. Can't really model those, at least not fairly (i.e. not without bias).

Much better to focus on the tangible, and limit the variables you seek to affect. This is not the same as "over-simplifying" your model. </div></BLOCKQUOTE>

I disagree that I agree with your point. http://fsae.com/groupee_common/emoticons/icon_wink.gif

Reliability and serviceability? Those don't have a place in a simulation discussion. The decision on what is possible in those realms is made well before simulation starts.

The differences you make in car configuration can be and have been all designed properly and improperly. A small single car can be designed to be easy to service and god knows that a four cylinder can can be made a huge chore.

Comparing a car configuration such as engine choice can very simply and effectively be determined with lap simulation. Like flavorPacket said, the simulation needs to be understood well but it is not something ANYWHERE near impossible.

[Kevin Hayward]

I seem to find that often when people start to doubt the accuracy of their simulations it is because they disagree with the results.

Charlie is right on the money saying that sims can be used to adequately assess the performance difference between concepts such as fours and singles. There was no mention of lap sims being the only measure however, which is equally valid.

It is incorrect to assume that effects such as tyre warming and wear, car width and handling are nearly impossbile to simulate. The tyre warming and wear are quite easy to incorporate using vehicle testing data. This is the only one of three effects that were mentioned that might throw a curve ball when it comes to comparing singles with fours. There is no law of manufacturing that states a single has to be narrow and a four has to be wide.

I don't see the point in running closed loop full transient models for the decisions you are trying to make in the design of these vehicles. Rather some open loop transient sims and a decent quasi steady state lap sim is all that is really needed. Bill's advice is worth following.

It is worth having an idea of the error value of the sims. If there is less than 10 points difference between concepts it is likely that it wont really matter when you get to the track.

Kev

[thewoundedsoldier]

Originally posted by Charlie:
I disagree that I agree with your point. http://fsae.com/groupee_common/emoticons/icon_wink.gif


Hehe, I would never try to impose my opinions on another! Maybe...

How about if instead I say that my original post agrees with you? Is that better?

I never tried to say that you can't use simulation to model complicated systems. I also never said that engine choice can not be "simply and effectively determined with lap simulation". I just want to point out that there are always variables which will not be present, no matter how hard you try. I find it more useful, therefore, to focus in on the variables I am interested in and understand their relationships within the simulation. I use simulation to gain an understanding of what is going on in a qualitative way, rather than use it to quantify a pre-determined design. To each his own.

My intention was not to warp the thread to be about particular semantics, so I apologize for the on-and-on.

Kevin, if my simulations EVER spit me out results that agree with my expectations, then i know for a fact there is some inaccuracy somewhere http://fsae.com/groupee_common/emoticons/icon_biggrin.gif

[Crispy]

Originally posted by BillCobb:
C.5 would be a 4 wheel nonlinear tire model with a quasi steady state solution technique. It's Bosch Lapsim, and work very well. Its the ONLY one worth developing because any racing machine requiring heavy transient analysis needs a new driver.


Does anyone know how, or if, the Bosch Lapsim takes into account the yaw degree of freedom (yaw moments and accelerations)? In their online manual, when referring to mass moment of inertia, they state "Their influence is very limited and one should not draw any conclusions out of the influence of these parameters on the lap time". This indicates to me that the yaw degree of freedom is not simulated.

My understanding is that "quasi steady state" implies only lateral and longitudinal accelerations are accounted for when generating velocity profiles. I'm not sure what a model would be referred to if if also included the yaw degree of freedom, but is still not fully transient (no pitch or roll transients or suspension vibrations). Perhaps quasi transient?

[murpia]

Originally posted by Crispy:
<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by BillCobb:
C.5 would be a 4 wheel nonlinear tire model with a quasi steady state solution technique. It's Bosch Lapsim, and work very well. Its the ONLY one worth developing because any racing machine requiring heavy transient analysis needs a new driver.


Does anyone know how, or if, the Bosch Lapsim takes into account the yaw degree of freedom (yaw moments and accelerations)? In their online manual, when referring to mass moment of inertia, they state "Their influence is very limited and one should not draw any conclusions out of the influence of these parameters on the lap time". This indicates to me that the yaw degree of freedom is not simulated.

My understanding is that "quasi steady state" implies only lateral and longitudinal accelerations are accounted for when generating velocity profiles. I'm not sure what a model would be referred to if if also included the yaw degree of freedom, but is still not fully transient (no pitch or roll transients or suspension vibrations). Perhaps quasi transient? </div></BLOCKQUOTE>
From watching Bosch Lapsim at work (only for a short period, though) my conclusion is that it must 'stabilise' the car from timestep to timestep. In physical terms imagine a car with a rotational yaw damper connected to the road by a slotcar shoe on the racing line. (The line is pre-programmed and smoothed by Bosch Lapsim).

That makes the simulation easier and faster, but minimises the effect of yaw inertia changes.

Note: I admire Bosch for releasing a free lapsim, and this should not be interpreted as a criticism.

Regards, Ian

[BillCobb]

Lateral acceleration has components from speed, yaw velocity and sideslip acceleration. That's all you need. Yaw acceleration is not needed or used in LapSim et al. The yaw inertia to mass ratio is a key ingredient to setting steady state performance factors, though. Yaw inertia separately is necessary for transient response.