Tire force combining review

[NaN]

One of the things I wanted to do for quite some time has been to compare Beckman pre-combining with the Pacejka curve fit of the Hankook tire.

Caveat: I am using touring.tire data with Pacejka/Hankook combining parameters. Beckman pre-combining depends on max force slip value (Fx, Fy curve shape). So to have a more fair comparison I have to feed Beckman with the Hankook curves, as they differ somewhat from the touring.tire. This will be addressed in a followup post.

Pacejka curve fit ±0°-32° slip angle, 4° steps:




Beckman pre-combining ±0°-32° slip angle, 4° steps:



According to this preliminary comparison (see caveat) Beckman seems to overestimate the longitudinal force component (Fx) with increasing slip, while Pacejka is keeping Fy closer to maximum at lower slip angle values. This means with Pacejka we would have more side grip but less longitudinal grip at small slip angles but overall less grip (particularly in longitudinal direction) at high slip angles (which is probably what we want for aggressive steering, drifting).

[NaN]

Here a more direct comparison.

Beckman (blue) vs Pacejka (red) ±4°, ±8°, ±16° slip angle:

[NaN]

Here is the rerun using Hankook tire force curves with Beckman and Pacejka.

Beckman (blue) vs Pacejka (red) ±4°, ±8°, ±16°, ±24° slip angle:



The results are comparable to the previous run. Beckman doesn't get low slip angles right. The match is better at high angles. This might be explained with the Hankook tire curves being less pronounced (lower max grip) compared to the touring.tire.

[NaN]

And for completeness, here are the force curves of the touring and the hankook tire.

touring vs hankook


You might notice the large difference in side grip (blue curves), one of the reasons why our cars are hard to drift I guess.

I am wondering if we could get away with using Pacejka combing parameters from Hankook tire with all our tires. In the real world they will certainly vary from tire to tire. On the other side the single parameter set we've got is clearly better than what we use now (especially at low slip angles).

[NaN]

Here is what I'd like to move to.

touring.tire using simplified Pacejka combining (4 parameters instead of 15):


All three tire types (touring, sport, slick) will be using the same (hankook) combining parameters for now. They can be individually adjusted though.

I'll push tire editor updates when ready: https://github.com/logzero/vdrift-tools

I also would like to make some adjustments to the tire curves in general, to have a more clear difference between the tires (pushed a change to reduce side grip of touring.tire today).

[nomoo]

That's cool

[NaN]

Here is what I've been thinking about tire grip adjustments (350kg tire load and no downforce):

- Touring tires would max out at about 1.0g sustained acceleration. This matches the hankook tire.

- Sports tires would max out at 1.25g (25% more grip).

- Racing tires would go up to 1.5g (50% more grip) . I think this is about what F1 cars reach at low speeds / start.

[NaN]

First adjustments run (new vs old)...

Touring - Sports - Racing


The touring tire matches hnkook data. I've reduced Fy (lateral grip) for all tires and also reduced Fx falloff (inspired by hankook). The Fx falloff is something that might need testing. I am not too sure about it.

[NaN]

Testing shows a number of RWD cars, like the 360, to be almost undrivable (I am using a steering wheel). Although the 360 wasn't very drivable before either.

On the other side we have some very well balanced cars like the LE or GTV which are doing fine.

Now, the question is: Should I push the changes anyway? Identify and document the problematic cars, exclude them from test builds and releases until they are fixed (who would do the fixing)?

[arturo]

Testing shows a number of RWD cars, like the 360, to be almost undrivable (I am using a steering wheel). Although the 360 wasn't very drivable before either.

Hi NaN,

I've been having a look at the .car file of the 360... I've added up all the mass particles, including 40 liters of fuel, and the sum fell short of the stated wheight; front/rear distribution didn't match either; the implicit wheelbase and front track were also different than declared; finally, the brake bias was clearly wrong.

http://dfiles.eu/files/y6yn9fk1m

Now, you may disagree with the "no inertia" solution of putting all the mass in a single point, but at least the car is drivable (while still a lot of fun).

[NaN]

I've made some interesting observations while messing around with 360.

Replacing the powertrain (engine, transmission, clutch, diff) with the one from LE makes it much easier to handle, which is to be expected I guess. The car still has the tendency to oversteer, but gives you some control up to a certain maximum drift angle. This is something I need to look into more closely.

One general issue is that one is often using too much steering (especially without a steering wheel). I've been thinking about adding a steering assist option, as used by the AI currently, which would limit the steering angle to the ideal range.

It would also help to implement the effect of tire side forces on the steering due to caster angle (force feedback). Currently we are only using tire aligning torque, which is too weak and doesn't provide direct feedback about side grip maximum imho.

[NaN]

After taking a longer break from VDrift I went back to my old sim code, rewrote it from scratch to be able to run it at full simulation speed.

What I am observing with 360 is that the yaw torque (z-axis) from the front wheels pretty much always (8 - 60 deg slip angle) exceeds the rear wheels. This means the car will always tend to oversteer.

Code:

fz 8982.29 fx 119.966  fy -10776.9 sa -9.128   tz -16293.4
fz 5647.44 fx 146.752  fy -8397.29 sa -9.48577 tz -12997.5
fz 11643.4 fx -515.037 fy -12240   sa -11.3225 tz 13882.3
fz 8464.06 fx -468.251 fy -12412   sa -11.6335 tz 13274.6
...
fz 8398.46 fx 488.18  fy -10090.4 sa -27.2215 tz -15730.1
fz 5128.35 fx 513.325 fy -7386.57 sa -29.8089 tz -11015.8
fz 10880.7 fx 860.208 fy -10601.7 sa -33.3253 tz 10965.9
fz 8122.35 fx 888.05  fy -10530.3 sa -35.9443 tz 12305

What we see here is that the front to rear weight ratio can not offset the front to rear lever ratio. So even when the rear wheels have more grip due to higher load, the front torque is still larger.

I've been wondering why the real car doesn't have this issue. At some point, after going through 360.car params a couple of times, I've noticed the difference. The rear tires are 30% wider, 215 mm vs 275 mm. This is something we don't simulate in any way.

With the pacejka tire model, this means different tire params for the rear wheels, to fix the oversteer tendency. The question here would be, how much does the grip vary with tire width (contact patch size)?

[NaN]

I think, I'll add slip angle testing to -cartest, so that one also gets some general car handling info (under/oversteer tendency).

[NaN]

So here is what I've learned about the tire width and contact patch relationship. The area of the contact patch is determined by tire pressure and weight the tire has to carry. This means a narrow and a wide tire will have pretty much the same contact area.

For the 360 with 3000N front and 4000N rear tire load and 30psi pressure, we will get 215 x 67 mm^2 for the front, 275 x 70 mm^2 for the rear, or about 30% larger patch. The reason here is the higher weight on the rear, not the tire width. One could have used 215 mm rear tires which would have a 90 mm long patch.

So why using a wider tire? The obvious reason is that the deformation of the tire is smaller with a shorter patch, resulting in lower roll resistance and heat. The more complicated reason is that a contact patch consists of a grip and a slip zone. With a shorter contact patch the slip zone is smaller, which results in more grip and also less heat.

Now how do we apply this to vdrift tires? Ideally we would like to have a (probably simpler) tire model that takes tire shape and pressure into account and somewhat matches the pacejka curves, doesn't have to be too exact imho.

[NaN]

My first attempt at a physical tire model (very primitive atm).

Not quite there yet.