Re: Car's physics and setting...

[Tomte]

- Once damperStrength is strong enough to result in 0.25 cycles to rest, it doesn't "dampen just a short bit and then take time to settle" like supsensionFriction, but rather just smoothly dampens at a steadily growing rate. I don't plan on figuring out this rate, as I imagine most of us don't want such behavior in a car. The exception could be if you were making a hovercraft of some sort.

Funny enough, all my cars have damperStrength set to more than 7500.

[C14ru5]

No need to worry, remember that this is data for 1m suspension travel.

[DonaemouS]

in quite addicted by your calculations, im not posting just for avoiding interruptions, but its time for me to understand one thing.

I suppose, these tables and your efforts will be really useful for supsentionfriction and damperstrenght. But what I missing is: you wrote about the time to rest of suspension based on travel and car mass. But It will be proportionally for all cars, or we need to use these table as just a starting point?

es:I have 2 BMW. They both have a suspension travel of 0,12 meters. But in the first one, the carmaker decided to have stiffer suspension, in the second one a suspension aimed for comfort. How I can extract this example from these tables?

[C14ru5]

But It will be proportionally for all cars, or we need to use these table as just a starting point?

It should be proportional, we just need to be sure about how it's proportional.

es:I have 2 BMW. They both have a suspension travel of 0,12 meters. But in the first one, the carmaker decided to have stiffer suspension, in the second one a suspension aimed for comfort. How I can extract this example from these tables?

To be honest, I don't know yet.

[C14ru5]

Different masses at damperStrength 1750, wheels.maxSuspension 1.0:

mass (m)	Cycles-to-rest (CTRd)
250 kg	0.5
350 kg	1
500 kg	1.5
700 kg	2
850 kg	3
1000 kg	4
1400 kg	6
2000 kg	8
2800 kg	12
4000 kg	16
5600 kg	24
8000 kg	24? 32?

Comments: The relationship seems to be linear for 1000 kg and above, but not for values below 1000 kg. I have no idea if this has to do with poor measurements on my side, or if there's a precision problem similar to what we saw with how low values for wheels.maxSuspension would give higher natural suspension frequencies than the formula.


Different wheels.maxSuspension at damperStrength 2500, mass 1000:

wheels.maxSuspension (smax)	Cycles-to-rest (CTRd)
0.0625 m	8?
0.125 m	8
0.25 m	6?
0.5 m	4
1.0 m	3
2 m	2

Comments: It's really hard to get data when it's on such a small scale, I had to use to use "allCams 1" and "finger-on-screen", and even then there was a lot of guessing. Let's look at some data at the other end of the scale, and see if we can spot a pattern...

Different wheels.maxSuspension at damperStrength 625, mass 1000:

wheels.maxSuspension (smax)	Cycles-to-rest (CTRd)
0.5 m	>16
1 m	12
2 m	8
4 m	6
8 m	4
16 m	3

Comments:
- These numbers look very clean, but I admit that I don't see the pattern. However, the behavior is very similar to the table above with the numbers belonging to the shorter suspension travel
- Mass and wheels.maxSuspension affect CTR in opposite ways. Hard to say if they're inversely proportional, though.


This work makes me wish that I had the code in front of me
It also wouldn't hurt to receive help from a mathematician.

[NoNameBrand]

It also wouldn't hurt to receive help from a mathematician.

log vs log might help you find a linear relationship...

I'm just at the wrong computer right now, will edit this later.

Different wheels.maxSuspension at damperStrength 625, mass 1000:

wheels.maxSuspension (smax)	log2 smax	Cycles-to-rest (CTRd)	log2
0.5 m	-1	>16
1 m	0	12
2 m	1	8
4 m	2	6
8 m	3	4
16 m	4	3

[NoNameBrand]

Zing! Spring constant!

At equilibrium, m * g = -k * x.

m = mass of the car * front-rear weight distribution / 2
x = maxSuspension / 2
k = spring constant, which redline calculates.

Let's take a hypothetical car with 50/50 FR distribution that has a mass of 1000kg, and a suspension travel of 0.1m at each corner:

250kg * -9.81 m/s² = -k * 0.05m
k = 250kg * 9.81 m/s² / 0.05m = 49050 N/m

How this helps: we can calculate this value for each sample of C14ru5's data (need to know how far the test wheel was from the CoM), and reduce the number of unknowns to the damping coefficient, and try to figure out how that's calculated from the .car parameters.

[See Flat]

Im reading this with one eye and can get over how damn smart you guys are. Lucky thing I was cute when I was young!

[slow]

indeed....what seeflat said. it all looks like swahili to me. i wasnt that cute, so i had to fight my way out.

thanks all for the effort. i wouldnt be here if it wasnt for you. slow

[C14ru5]

need to know how far the test wheel was from the CoM

CoM was at {0,0,0}, which isn't relative to the ground, but to the resting height of the wheels at {1,0,1}, {-1,0,1}, {1,0,-1}, {-1,0,-1} (for instance, the wheels.pos height in the .car was 0.5 m for 1m travel)

I have one more test to complete, and that's combining positive and negative values of supsensionFriction and damperStrength to see if this affects response time and can be used to get around the stiffness/travel limit given in wheels.maxSuspension. I doubt I'll have any time to do it this weekend, but I'll try it sometime next week.

[DonaemouS]

(for instance, the wheels.pos height in the .car was 0.5 m for 1m travel)

Why for 1m travel? If you use a 0,125 suspension height, mean a 0,25 travel? (the value we use *2)

[C14ru5]

That's correct.

[NoNameBrand]

need to know how far the test wheel was from the CoM

CoM was at {0,0,0}, which isn't relative to the ground, but to the resting height of the wheels at {1,0,1}, {-1,0,1}, {1,0,-1}, {-1,0,-1} (for instance, the wheels.pos height in the .car was 0.5 m for 1m travel).

In game terms though, the centre of mass is always relative to the origin of the model, n'est ce pas? I can see that they're identical here.

So! Spring rate for any of your wheels should be easy to work out.

I want to do an example to see if this actually works, though.

The Corvette Z06 ( ) has spring rates of 531 lb/in and 782 lb/in (f/r). Converting to metric, we get 93219.6 N/m and 137283.8 N/m. ( To convert: metric = imperial * (1 / 0.0254m/in) * (1 / 2.2 lb/kg)* 9.81 m/s² )

The Vette has a curb weight of 1420kg, and a 50/50 weight distribution. I'm going to add 100kg for a driver, for 1520kg.

Each wheel is going to see a quarter of this, or 380kg.

Front wheels:
F = ma = -kx
380kg * -9.81m/s² = -93219.6 N/m * x
⇒ x = 380kg * -9.81m/s² / -93219.6 N/m
= 0.039989m
∴ wheels.maxSuspension = 0.07998

Rear wheels:
380kg * -9.81m/s² = -137283.8 N/m * x
⇒ x = 380kg * -9.81m/s² / -137283.8 N/m
= 0.027154m
∴ wheels.maxSuspension = 0.05431


...might explain why some of my cars drive so funny - bigger wheels at the back, so I made maxSuspension bigger - my springs are too loose and in the wrong balance.


Now, for the C6.R race car.
I have no idea what the real spring rate is. I want stiff suspension, which I think is accurate for such a beast.

I use a set of formulas for tuning race cars in Forza 2. I adjust the springs so that I have 1 lb/in for every lb at that corner (that part isn't very complicated).

This car has a weight of 2425 lbs and 50/50 weight distribution, so each corner sees 606.25 lbs, and therefore the spring rate is (dun dun dun) 606.25 lb/in.

That's 106430 N/m.

Onto Redline:
Mass is 1100kg, with another 100kg for a driver.
Each wheel sees 300kg.

300kg * -9.81m/s² = -106430 N/m * x
⇒ x = 300kg * -9.81m/s² / -106430 N/m
= 0.02765m
∴ wheels.maxSuspension = 0.0553


I'm eager to try this out.

[C14ru5]

Long body of text here, it's probably all due to me having a hard time turning off my 'master thesis mode'.

I wanted to figure out the sway bar equivalents and present them in this thread, but ended up discovering something peculiar about the Center of Gravity height. It's been taken for granted that this is measured from the ground (so if the model is at resting height with the wheels touching the ground at Y=0, all the numbers fit). In this post, I will argue that this view is a result of coincidence.

My interest in examining this issue came from my Ford Sierra plugin, where I had correct CoG data from a good source, but where I only could stabilize the car if I lowered the CoG height far below its real value. My initial hypothesis was that the suspension roll center (rotation on the z axis) was determined by the CoG, and was not in any way related to the position of the wheel axes. I further believed that if the distance between the CoG height and the height of the wheel axes became too big, the Redline engine got confused. I have not been able to reach any conclusion about the suspension roll center, so more testing is needed. However, I made a testing plugin where I exaggerated the dimensions, and suddenly I was presented with a puzzling problem conterning the CoG height - it doesn't seem to be measured from the ground!?

Early in this thread, I mentioned research that had shown most road cars having a CoG height that was 38-40% of the roof height (from now on: 'body height', as I don't want to confuse this with the height measured from the ground). Although there isn't any data available for racing cars, a thread at RaceSimCentral (link is dead at the moment) contains a discussion where it's believed that a value of 35% may be sensible in some cases, but in other cases it may even be higher than 40%. To make sense of these percentages, compare them with the front/rear weight distribution in a car. A muscle car with an oversized engine may have 35% weight on the rear wheels, and ye olde Porsche 911 may have 35% weight on the front wheels.* Therefore, a CoG height of 38-40% for most cars would seem reasonable. Many (but not all) cars in Redline have a CoG less than 0.1m above the wheel axes, which typically would equal about 20-30% of the body height. Still, the cars roll and flip through the air as if they have a natural and believable CoG.

*These values are measured at the wheel axes, and not at the body extremes, so my argument could be weakened if the weight distribution varies depending on the distance from the midpoint. (Unfortunately, I've forgotten the relevant physics to make any claims here)

Have a hack at this test car. Looking at the car model, most of its weight should come from the wheels, suspension frame, transmission and the engine (hence, the body height percentage theory must in this case yield to the total height percentage). Right now, the CoG is placed 0.15m below the wheel axes, and the car behaves the way I think it should. "Below the wheel axes, you say? That can't be right..." Now try raising the CoG to 0.1m above the wheel axes (still much lower than the engine and a large part of the transmission). We get a very different behavior, and the car may flip over if it lands at an angle after a jump. If the CoG height is raised to the point where the transmission splits, less than 30 degree roll causes a flip, even with the CoG at 40% of the body height (still below the midpoint!). I've prepared the different values inside the .car file, so you may just comment out the various lines during testing.

The behavior could be due to some unusual suspension settings for a Redline car (1-2 cycles-to-rest), and it could also be due to a false impression on my part. However, I strongly feel that the CoG height isn't measured from the ground, but from the position of the wheel axes at any given point in time. Try to spot the angle where the car flips over, if you want to understand what I mean.

A coincidence in standard vehicle proportions, combined with a collective assumption that a CoG close to the wheel axes is a safe bet, a lack of proper suspension testing, and a lack of real-world CoG comparison data has until now led us to believe that the CoG height is measured from the ground. Let me provide you with a numerical example: An unnamed car in Redline with total height 1.3m, road clearance 0.1m (thus, a body height of 1.2m), wheel radius 0.3m and CoG height 0.4m. If the CoG height is measured from the ground, we get: 0.4m/1.2m = 33% of the body height. However, if it in fact is measured from the height of the wheel axis, we get: (0.3m+0.4m)/1.2m = 58% of the body height. This means that the car would flip if it had somewhere around 30 degree roll. Thruxton GT3-contenders: Does this sound familiar? I believe that we've been able to live with this for so long, because many of the cars have too stiff sway bar values, and that some canonical plugin makers (like the man some refer to as 'Gerry') have set a rule of thumb that not all of us bother to look into.

I hereby invite plugin makers to test my CoG theory further, and to see if it has any truth to it, or if it's all just a false suspicion that I have.

[Tomte]

Now that is interesting. Looking at the .car file, I can't find anything obvious wrong with it, which concludes that you're onto something here.

I haven't thought hard enough yet to prove or disprove your theory. But I'd like to remark on two things:

However, I strongly feel that the CoG height isn't measured from the ground, but from the position of the wheel axes at any given point in time.

1. What is the axle height? the straight line between two wheels? the monster truck shows very nicely that the 'virtual' axle, which connects the front or rear wheels at rest, will tilt considerably when cornering. I wonder from where the CoG would act on once the car leans over.
   
2. 2 of my cars, the Golf and the 190, have C0G's of about 0.55m (of what I assumed to be the ground, with the car's body placed so that the wheels would touch the ground at Y=0). Together with a wheel diameter of roughly 0.31m and a body height of about 1.25m, this would give (0.31+0.55)/1.25=69%. Now I have a hard time getting these cars to flip, swaybars or not. Scratch that, I do manage at Donington, quite easily actually So this is what you meant with stiff sway bars. Interesting.
   

I will see if I can look into this a bit more, but a quick test with an Atom with no swaybars and the CoG set to 0.1 (m above axle height?) lets it corner very very flat, no body roll noticeable.

[djpimley]

Before you get too intricate with those calculations, remember that gravity in Redline isn't even correct - IIRC it's about 80% Earth gravity.

[Pax-raider]

I know that jonas still checks the Ambrosia forum occasionally, so a PM might yield some useful knowledge from Redline's "centre of gravity". Worth a try anyway. If he can't remember how he did it, he should at least be able to poke around in the physics calculations to tell us what's there...

[NoNameBrand]

Jonas was on this forum for a few minutes this morning. I was hoping he'd pop into this thread.

[C14ru5]

No need for Jonas yet, I'm actually feeling pretty confident now, after doing some more tests

What is the axle height? the straight line between two wheels? the monster truck shows very nicely that the 'virtual' axle, which connects the front or rear wheels at rest, will tilt considerably when cornering. I wonder from where the CoG would act on once the car leans over.

You seem to have understood what I meant to say, and I share your uncertanties. Here's what I think is going on, and it's a solution that makes a bit more sense in a modeled world than in the real world:

Imagine that the car suspension in Redline is modeled as a virtual plane that is stretched between all the different wheel.pos values. This suspension plane may be mathematically defined by a vector positioned in the center of all (3 to 6) wheels. The position and angle of the vector is affected by different forces like gravity, momentum and resistance (supsensionFriction, etc.). The suspension vector also has some modifiers, which are vectors in each of the three dimensions defined by the CoG. In the case of the added X and Z vectors, these values remain fixed in relation to the suspension vector no matter how the virtual plane changes its height or angle. But since the suspension is continuously being re-calculated, it makes more sense to load the Y modifying vector from a position relative to the main suspension vector, than constantly having to translate the value from an absolute position in the model's coordinate system. In the versions of Racer earlier than 0.5.0, plugin makers had no choice but to position the model with the CoG placed at {0,0,0}, probably because of a similar approach to the physics model as the one in Redline. So by relating CoG directly to the suspension instead of the model, lots of code and processing power is spared, but the results are more confusing to us who write the car plugins.

(My guess is that the 'virtual plane' model mentioned above may be a reason why 2 wheel vehicles are impossible in Redline, because two points aren't enough values to form a virtual suspension plane. Just a thought.)

I will see if I can look into this a bit more, but a quick test with an Atom with no swaybars and the CoG set to 0.1 (m above axle height?) lets it corner very very flat, no body roll noticeable.

The "no swaybars approach" is probably not a bad idea during testing. I've now tested various CoG values on a couple of cars of mine, and they all get no body roll whatsoever when the Y CoG value is close to 0. Negative values produce inverse roll. When close to 0 and no body roll, the car doesn't respond normally. A good analogy is that it feels like you're trying to control a sliding piece of cardboard that you're sitting on across a bumpy surface, and your only way of steering is by leaning your body in a certain direction. It seems that the suspension roll center is placed exactly on this virtual plane between the wheels, no matter what the CoG values are. In normal road cars, the roll center lies close to the front wheel axle but maybe 0.2m above the rear wheel axle. In other words, it may be necessary to find a compromise between where the CoG lies and where the roll center lies.

To summarize, I'll propose a method to find the proper CoG height:
- Determine the wanted CoG height by any means you wish, either by empirical data, by eye, or by calculating it as a percentage (often 38-40%) of the body height.
- Convert the absolute value you determined into a value relative to the plane between the wheels
- Test the car with little or no sway bars, in order to spot the point at which the body rotates around, both during normal driving and when the car is in a free tumble
- Raise or lower the CoG by a small amount to adjust for any offset roll center in the suspension
- Make adjustments to the friction parameters in the suspension, including sway bar settings, until the amplitude of the body roll is satisfactory

[Brook]

According to the wiki, shouldn't you recalculate inertia if you move the cog around?
If I increase the z inertia on your monster apparatus, things improve.

I've only just started to delve deeper into the .car file, so excuse me if I make no sense.

[C14ru5]

shouldn't you recalculate inertia if you move the cog around?

You should indeed. I didn't bother - instead I used the simplified formula that doesn't take CoG into the equation. Three reasons: 1) The CoG (usually) affects only a few tenths of the final calculated inertia 2) I didn't want to camouflage any behavior by "dumbing the vehicle down" with too high inertia settings 3) I wanted to test a lot of different settings for the CoG, and fiddling around by re-calculating the inertia takes too much time when the effects are only marginal.

The point of my experiment wasn't to fix the car, but rather to reveal its true CoG.

[NoNameBrand]

According to the wiki...

The wiki was written by us (the collective us, not just the posters in this thread) - if one of the fundamental assumptions we made turns out to be wrong, well, some things will need to be rewritten.

I still find it odd that everything else in the .car file is based on the model's origin. Interestingly, the external camera height is based on the assumption that y=0 is the same for every model. Notice the different camera heights/angles on different cars.

I hope to start reworking my cars this weekend, so it'll be interesting to try this in conjunction with the other ideas in the thread.

[C14ru5]

I still find it odd that everything else in the .car file is based on the model's origin. Interestingly, the external camera height is based on the assumption that y=0 is the same for every model. Notice the different camera heights/angles on different cars.

If we assume that there is in fact a two-world split (physics world and graphics world), then the reason that most of the other variables can be based on the model's origin is because those variables remain fixed in position. The variables are imported once from the graphics world into the physics world, and don't require any re-calculation. If you need another example, IIRC the air resistance values were vectors that were applied directly to the CoG, and not in relation to the model. Also, the fact that the wheels assume that the surface is placed beneath them (making driving upside down impossible), further suggests that there is a two-world split.

[slowDan]

Some interesting stuff here.

I do remember the illustrious thomm mentioning some problems with the CoG of the Lola way back in the day. Something about when it got below the axle height things started to go a bit weird. At the time I didn't think too hard about it and he found a value that worked. But it starts to add to the pattern now!

This would obviously be bad news for cars with a very low, possibly even below the axle, CoG wouldn't it? The simulation just doesn't work right if the CoG is down there. Obviously it can be bodged/fudged but it would be nice if it worked right.

[Tomte]

Maybe I forgot to say that when I moved the CoG of the Atom from 37 cm down to 10 cm above the axles (in the spirit of testing 'your' coordinate system), I didn't find the driving behavior natural at all.
I adjusted the suspension settings as well, the wheel travel settings should be close enough to the real world as well.

I understand what you want to say about the vector defined plane for the roll center. I just don't believe it works that way. Standard mechanic calculations allow us to use the CoG as a simplification of the car's mass spread over the dimensions of the vehicle. Using the distance to the wheels in 3 dimensions allows us to calculate the load and force at each wheel. The forces at the contact patch work against the forces applied by the tires. All this you can calculate without the need of the roll center.
I wouldn't wonder a bit if the caster angle of a wheel while cornering is not accounted for in Redline (I'm not talking about the wheels.tilt settings here). The suspension modelling is fairly crude in Redline, with the wheels moving only on the vertical plane and in parallel with the body roll. With this in mind, you wouldn't need to consider the roll center.

[Sponsored content]