Quote from Juls :

Hm yeah my exemple with spring dampers was too simple. I was thinking something like take a simple model like a LUGRE stiction/friction and I figured it would not look simple at all for the reader

I wanted to say something like that: IMO if you really model the tyre/road action/reaction like a stiction/friction mechanism, even a very simplified one, you can avoid many pitfalls.
- some stiction/friction models are robust at speed 0
- they can immediately take into account hysteresis
- they seem scalable and very convenient for object oriented programming

Of course you have lots of experience I clearly don't have. But as a software developer, my feeling is that implementing such model instead of the very usual slip curve model could be a win win situation. Possibly richer model at the end, probably less problems adjusting it, more intuitive. Don't see the problems precisely in advance...but usually I am not too bad guessing which general direction is less crowded with problems than the other

Juls, you're intuition on this one is spot on. I've been doing things in a very similar manner for quite some time now and it indeed works very well

For any others here who have heard of "problems with Pacejka" when it comes to car sims, you've now seen exactly what developers run into and the dilemmas that pop up. Historically most sim developers have been completely unaware of these very issues. So when folks say "yes, Pacejka's fine, but you need to be very careful about combined grip" you now have an idea of what they're talking about

Pacejka has published solutions to this, so again I encourage anyone who wants to use Pacejka's magic model (he's written tons of other models too, this is just the only one that sim folks hear about), make sure you use one of his full, combined solutions that solves these problems or at least minimizes them.

From my personal experience as a sim developer, the dependence of a lot of tire model parameters on load is just about as important as the "flatness" of the slip angle curve beyond its peak. I currently use a simple variant of the brush model, and it's frustrating that most research papers don't really care about variations with load, or the effects are not described in detail.

On another note, one of the things I cannot really explain for myself from a physical point of view is why the longitudinal slip curve would have a peak and then fall when the lateral doesn't, or vice versa. From my understanding all of the contact patch is sliding at the long/lat peak, why would the long. force drop after that? Or, why wouldn't the lateral force drop? In my model the rubber friction coeff. drops slightly with sliding speed, this gives a reasonably small drop of lateral force, but also small, basicall no drop for the longitudinal force. If Todd would care to explain his take on this I would be eager to listen

Quote from JeffR :

Note that lateral peak force is related to the component of force perpendicular to the direction of travel.

Someone might choose to describe and model tyre forces this way, but it is not the standard method of doing it used by almost everyone.
See the figure on page 2 of http://zzyzxmotorsports.com/li ... uencing-tire-modeling.pdf . I think most of published tyre data is compatible with the coordinate system described in this document.
Lateral force is perpendicular to the direction of wheel heading, not the direction of wheel travel.

Quote :

laterforce relative to direction of travel

Quote from w126 :

Someone might choose to describe and model tyre forces this way, but it is not the standard method of doing it used by almost everyone.

I was confusing an email exchange I had with this thread (in the email the car chassis was the frame of reference, a non-conventional model). I deleted my previous message, since it would be confusing in this thread.

Everytime I see a discussion about tyre models, the main topic is tyre forces vs slip, let's call it tyre slip sensitivity. But what about tyre load sensitivity?

I don't know how it is with LFS, but tyre load sensitivity is strongly underestimated in all ISI/Simbin titles. Recently I could find some real world data showing tyre load sensitivity, how the peak friction coefficient decreases with increasing load. It is 10-20 times more sensitive than in ISI/Simbin titles.

For me the load sensitivity is very important to get accurate tyre behavior, as much as slip sensitivity.

When going sideways, the tyre slip sensitivity (grip slightly decreasing after peak) is responsible for the traction loss and wants to make your car spin passed the peak slip angle. It defines how punishing a tyre is. It is cause of unstability.

But the load sensitivity is very important too. It acts the opposite way as a stability factor and defines how easily you can recover at high slip angles. When you go sideways passed the peak slip angle, the lateral forces decrease (slip sensitivity). The car can't follow anymore the curve, the turning radius increases. Less cornering-> less load on outside tyres-> their grip increases pretty well thanks to the load sensitivity-> you can keep the car sideways without spinning or even recover.

IMO the balance between tyre slip sensitivity and tyre load sensitivity is very important for handling. The load sensitivity is strongly understimated in all ISI/Simbin titles (and no surprise these titles are well known for the punishing spins when you exceed a given slip angle). Another thing...the load sensitivity makes the car alive on track bumps. When you are cornering close from the limit and a bump makes your car pivot this is for me a typical consequence of tyre load sensitivity.

From what I remember from older discussions, there's some sort of consensus about load sensitivity being botched in LFS and the reason why differentials don't behave quite the way you would expect them to do, i.e. locked diffs + ultrastiff suspensions work wonders, because the outer tyres can be overloaded like there's no tomorrow, and even with the inner suspended in the air, it can still exert more force than the two tyres both making contact with the ground and more realistic loads on them.

Tuning handling with spring stiffness and anti-roll bar changes the load repartition (and the way it changes during time), and this changes the cornering forces each tyre gives at a given moment.
It works even with a zero load sensitivity.

Don't know if I am clear.

Max lateral force = u*Load-loadSensitivity*Load*Load.

If you take loadSensitivity=0, you still have tyres reacting to load change....change the load repartition during time through suspension stiffness and anti-roll bar and you change car handling.

The load sensitivity is not vital for the basic handling of car, but it makes tyres more intersting. Without it, tyres have only one reason to give up or get back grip...slip angle, lateral force. With it you have two reasons to give up or get back grip, lateral force and load. One is unstable (slip), the other is stable. When you slide more and more the tyre gives up grip because of slip sensitivity, but get some grip back through load sensitivity.

IMO this is the reason why you can drift (slip angle up to 20 degrees or more ) with race tyres which reach peak lateral friction at 8 degrees or less.

It is obvious that tyre load sensitivity is how much the tyre friction coefficient changes when load increases. One time again you are confusing tyre load sensitivity with tyre friction for a given load. Sensitivity is the derivate of the friction coefficient according to load.

So when you take this line (-3.10e-6, 0.65, 14000.0)

Tyre sensitivity is the slope of this curve multiplied by friction coefficient, not the final value 0.65. Final value 0.65 is very probable for a real tyre...this is not the problem.

But if you look at the slope...tyre load sensitivity for a real tyre when load is zero is about -3.75e-5, ie 12 times this one above (first value is the initial slope).
And for a real tyre, even if the final value is close from the one above...0.55, the slope is almost constant, -3.75e-5 all the way from 0 to 12000N.

If you compare (-3.10e-6, 0.65, 14000.0) with (-3.75e-5, 0.55, 12000.0), then:
* at 0N, real tyre data gives 12 times more load sensitivity than this tyre.
* at half static load 1500N, real tyre data gives 3 times more load sensitivity than this tyre
* at static load 3000N real tyre data gives 2 times more load sensitivity than this tyre
* It is only at 8000N than this tyre load sensitivity becomes equal real tyre data, and later it increases a lot to give the 0.65 final value.

Looks like nothing, but this changes a lot the tyre behavior. I wrote 10-20 times higher because I was thinking about this line and the slope in zero...but it is definitely 2-3 times higher in normal driving conditions. Here sims and real world data are very different...why?
Yeah I am not such an idiot....surprise...

Sorry, are you talking about the same numbers? Using the data you provided and the equation Juls provided I'd agree that the load sensitivity was severely understated in GTR2 (although 20x more would be too much).

See attached table...

Edit: ah, Juls has posted again. Now I'm confused. Is there more to the load sensitivity in the ISI engine that just a linear drop?

Yes Bob...(-3.10e-6, 0.65, 14000.0)

It defines a multiplier applied to friction coefficient, initial value is 1.0 for load 0. First figure is slope in zero.
Second figure is final value for the multiplier, reached for load 14000N.

To rebuild the curve in a unique way, we assume the slope is zero in final value...we can make a 3rd degree polynom from that.
Other possibility is to make a 2nd degree polynom if you don't care about slope in last value.

Both curves rebuilt give very convex shapes. From real data and many publications, the curve is almost a straight line...slope should be constant...but rebuilt curves from sim data are very different...slope is 10 times lower than real data first, and only reaches real data value beyond load range a tyre will cover when used in normal racing condition (almost 3 times static load).

Comparing 2nd degree curve with straight line from real data gives bigger difference...12 times lower sensitivity at 0N, 5 times lower at half static load, 3 times lower at static load....curves cross each other at 11000N.

Tyre load sensitivity (last chart in the document) for two unknown tyres, showing a straight line and a very important grip at low load.
http://www.optimumg.com/Optimu ... chTips/TireComparison.pdf

Quote from Juls :

Tyre load sensitivity (last chart in the document) for two unknown tyres, showing a straight line and a very important grip at low load.
http://www.optimumg.com/Optimu ... chTips/TireComparison.pdf

I don't know if you're aware, but that's not measured data, just the load sensitivity representation of the Pacejka '96 model:

Quote :

It is often useful to fit a tire model to the data –
even if we won’t do any simulation work. Comparing
two tire models instead of directly comparing the
raw data allows us to more easily see differences – the
model removes the experimental noise and test hysteresis
inherent in the raw data. Fitting a tire model
is a relatively simple task using OptimumT’s Model
Fitting Tool. For this example, we will fit Pacejka
Magic Formula ’96 models for both tires.

So that's "just another model".

Uh, sorry you are right, picked the wrong document

Here is the doc with actual measures. I uploaded it because I can't find any link to it anymore.
http://www.megaupload.com/?d=KDME226O

Load sensitivity (figure 4.6, 4.10,4.14, 4.18...E.2, E.3, E.5, E.8, E.9, E.12, E.13, E.16, E.17) is almost linear, with high friction coeff for low loads....never a convex shape like in ISI-based sims tyre files (assuming the three coeffs are what they are said to be in the comments).

I dare to hope they have more than 2 points of measure and don't call "measure" model-fitted data

BTW this document has other valuable things, like the speed sensitivity and the stiffness...pretty hard to find.
Amazing how tyre grip falls quickly with velocity...ISI-based sims use less dramatic values there too.

Quote from Juls :

Uh, sorry you are right, picked the wrong document

Here is the doc with actual measures. I uploaded it because I can't find any link to it anymore.
http://www.megaupload.com/?d=KDME226O

Load sensitivity (figure 4.6, 4.10,4.14, 4.18...E.2, E.3, E.5, E.8, E.9, E.12, E.13, E.16, E.17) is almost linear, with high friction coeff for low loads....never a convex shape like in ISI-based sims tyre files (assuming the three coeffs are what they are said to be in the comments).

I dare to hope they have more than 2 points of measure and don't call "measure" model-fitted data

BTW this document has other valuable things, like the speed sensitivity and the stiffness...pretty hard to find.
Amazing how tyre grip falls quickly with velocity...ISI-based sims use less dramatic values there too.

I haven't looked at the doc yet, so I may have wrong assumptions, but the case of _sliding_grip_ of rubber being a function of the sliding velocity is mentioned in a lot of publications. This does not mean that the max. grip varies similarly, although it seems logical to assume _some_ speed dependence there as well.

Actually two characteristics of rubber friction (rubber block, not a full tyre) appear often in publications, and with mostly the same formulae.
One is the speed dependence of the sliding friction (~1/(1+k*speed)), and the other is the load sensitivity of friction, which is *not* linear, but more like the graphs we see from LFS's model (~1/load^0.x).

While on this topic, I have seen 3 kinds of load sensitivity formulas for tyres, one as mentioned above, the other two being exponential decay and linear decay. It's true that I've not seen the rF type curve mentioned anywhere, but the load sensitivity curve points in RBR (although the scale is not clear there) don't seem to fit a "simple" mathematical formula either, and they are also supposed to be from measured data..

It is not the friction of rubber for different sliding velocities, it is a measure of max lateral/longitudinal friction coefficient of the tyre for different velocities of the vehicle. Not so easy to find....most benchmarks are performed at constant velocity.

You can see in this document the load sensitivity of friction is more linear than expected (from a rubber block), at least for the lateral friction. Quite logical if we think the tyre contact patch is always between stiction and friction.
Even for very large slip angle, as the tyre is rolling, the rubber from the contact patch is "renewed" all the time and this "new" rubber, before fully sliding, first sticks and deforms.

As a result, it seems to me the behavior of a tyre (not locked) is always more stable, more progressive than the rubber block. Don't know if I am clear. In fact I know I am not

Just ran across something that reminded me of this thread. Regarding the shapes that the poster was coming up with in Excel, what I was trying to get at is the end result should look something like this:

http://white-smoke.wetpaint.com/page/Combined+forces

There shouldn't be any valleys and so on in the middle regions. This is probably old news by now, but wanted to post it anyway for others in the future.

After years and years being away from this sim/game, I came back to see what is going on.

Now where I’m browsing through the forums, I found this thread.
Quite an interesting discussion going on here.

And nice to see some familiar faces still in here, like Todd.
(So you are a computer nerd ?, ahm,… Yes ) Still remember ?

Anyway, I will follow this a bit and look what you clever guys come to agree with.

Rene

"Game nerd" were her exact words.

Yes, and I like horsey food. Good to see you, Rene