Clutch modeling

Vehicle physics, suspension, tires etc

Clutch modeling

Postby ply » Sun Feb 17, 2013 9:54 pm

I wonder how clutch mechanics is modeled in the game? I mean “manual” clutch mode. Now it almost impossible to start moving a car with RPM values lower than 3000-4000, regardless how slowly I release the clutch pedal. I suppose in real car smooth take off should be possible at 1000-2000 RPM. For some cases it should be possible even at idle speed almost without using throttle only with proper clutch handling. Clutch pedal also has dead-zones at the limits, but not in the game. Is it now modeled as sharp step-wise attachment of spindle to the motor? It feels something like that.
ply
 
Posts: 37
Joined: Sun Feb 17, 2013 9:36 pm

Re: Clutch modeling

Postby Eugene Cojocar » Mon Feb 18, 2013 8:11 am

The clutch transfers torque from engine to wheels arcwise and vice versa from wheels to engine.
So, the current clutch torque = Max. Clutch Torque * clutch pedal.
Max. Clutch Torque can be found in the Vehicle Physics->Trans.
The dead zone is not implemented at the moment.
The moving also depend on the torque curve, please study the "Torque" and "Torque vs Throttle" curves in the Vehicle Physics->Engine.
Torque - defines max torque at the full throttle.
Torque vs Throttle - defines dependence on the throttle.
Your current torque = Torque * Torque vs Throttle (throttle).
Please make sure, that your torque at 2000-3000 RPM is enough to start moving.
You can watch it via telemetry (by key 'I').
XMR Developer
User avatar
Eugene Cojocar
 
Posts: 937
Joined: Thu Oct 18, 2012 2:49 pm

Re: Clutch modeling

Postby ply » Mon Feb 18, 2013 8:44 am

Oh, I see. So dependence is linear. I suppose it should be highly-nonlinear, at least at the position of the pedal where clutch discs just starting to grip each other. Maybe S-shaped curve (in this case dead zones will appear also). Ok, it's only my subjective thoughts, real experimental data is needed. I will try to find some publications on the problem.
ply
 
Posts: 37
Joined: Sun Feb 17, 2013 9:36 pm

Re: Clutch modeling

Postby Eugene Cojocar » Mon Feb 18, 2013 10:31 am

I'm looking forward to real data about this process.
Maybe, adding an additional curve to control this process is a good idea.
XMR Developer
User avatar
Eugene Cojocar
 
Posts: 937
Joined: Thu Oct 18, 2012 2:49 pm

Re: Clutch modeling

Postby ply » Tue Feb 19, 2013 12:07 am

1. Clutch pedal

First of all clutch pedal always have free-play area and also some non-functional space after clutch disengage. This is described in service manual for a car.
Image

Usually in a real car free-play distance is something about 1 inch, working range is about 4 inch and disengagement position has 2 more inches until floor.
Thus total range of pedal movement should be divided in zones something like 15% - 60% - 25%.

2. Torque curve

Next question is how transmitted torque is related to pedal displacement. I've found some articles about this, and most useful are:
http://dx.doi.org/10.1109/TMECH.2010.2047509
http://dx.doi.org/10.1016/j.finel.2010.08.007

Transmitted torque is related to load force by equation

T = n * μ * Req * F

T - torque;
n - number of pairs of contact surfaces (we have two because clutch disc has two sides, so n=2);
μ - friction coefficient;
Req - some equivalent radius of clutch disc;
F - normal component of load force applied to friction surfaces.

Load force F is dependent on clutch disc displacement, and controlled with the pedal.

Some typical curves look like these:

ImageImage

Displacement coordinates are measured relative to clutch disc, but we can assume they linearly dependent on pedal position.

I've come up with such empirical approximation for displacement-force dependence:
F(x):=Fmax*(1-(1-p*x)^γ)/(1-(1-p)^γ)
γ = 0.2
p = 0.95
Fmax = 5000 N
x range: 0.0 - 1.0
This is actually slightly modified gamma-correction curve from computer graphics. All parameters are very approximate.
I think it is a good starting point. It takes into account low rate at the beginning and has steep rate at the end of clutch working range.

Now we should define μ and Req coefficients. We could take these values:
Req = 89 mm
μ = 0.24

But the problem is friction coefficient μ is dependent on engine rotation speed.
Maximal torque will be transmitted only if motor is rotating fast enough to sustain “saturated” friction force at the clutch. If motor is rotating slower than some limit, friction force will be correspondingly smaller.

3. Slip speed

Friction coefficient μ is dependent on angular frequency of motor rotation. To be more exact, friction coefficient depend on difference between rotation speed at the input and output of the clutch. If engine rotates with angular frequency ωf, and clutch output (transmitted to the gear box) rotates at the frequency ωc, then friction coefficient should be function of angular frequency
ωfc = |ωf-ωc|.
Note the absolute value is taken. I am not sure how the “direction” of applied torque is implemented in the game. If the road move faster than the engine rotates, the car will naturally decelerate.

Fenomenological model of dependence μ(ωfc) is given in the first cited article, and I repeat it here with some simplifications.

μ(ωfc) = μs + (μd-μs) * tanh[g*ωfc*Req]

ωfc = |ωf-ωc|, rad/s
μs = 0.15 - static friction coefficient
μd = 0.24 - dynamic friction coefficient
g = 1.5 s/m
Req = 89 mm = 0.089 m

Resulting torque depending on clutch displacement and including frequency looks like this:
Image

We can notice, than friction coefficient μ(ωfc) saturate to its maximal value μd very fast. For rotation speed difference of 200 RPM it is almost at full value. So μ(ωfc) function is not actually relevant for the case we start moving the car from full stop, because idle speed is much higher (~800 RPM), and gear box is not rotating at all. But it became influential when speed of engine approaches the speed of road. And it has influence only if clutch pedal is depressed, because if the clutch is locked-up, the engine limits maximal torque, not the clutch.

We can conclude than most important part is to model pedal dead zones from section 1 and load force F(x) dependence from section 2. Dependence μ(ωfc) from section 3 is not so critical (can assume μ is constant) but could give more smooth transitions if implemented.
ply
 
Posts: 37
Joined: Sun Feb 17, 2013 9:36 pm

Re: Clutch modeling

Postby Eugene Cojocar » Tue Feb 19, 2013 8:11 am

Many thanks for this magnificent explanation of clutch characteristic.
It's very clearly for me after this explanation.
And of course, I will try to implement this clutch mechanism and hope it will make the clutch behavior more realistic.
XMR Developer
User avatar
Eugene Cojocar
 
Posts: 937
Joined: Thu Oct 18, 2012 2:49 pm

Re: Clutch modeling

Postby ply » Tue Feb 19, 2013 8:50 am

I tried to set "Max Clutch Torque" to small value of 100 N*m to simulate slow initial increase of F(x), and the car behaves much more smoothly. But I can't take good speed because torque is limited :)

Also I just thought that free-play zone of the pedal can be avoided, because usually maximal torque of the clutch is much higher than torque of the engine when pedal is not pressed. So when one starts pressing the pedal effect will only appear when we move along the F(x) curve down to the current torque of the engine.

Anyway, it would be great if all parameters will be adjustable.
ply
 
Posts: 37
Joined: Sun Feb 17, 2013 9:36 pm

Re: Clutch modeling

Postby Eugene Cojocar » Tue Feb 19, 2013 9:32 am

I think that an implementation with the adjustable values is more preferable.
It allows to get various behavior.
XMR Developer
User avatar
Eugene Cojocar
 
Posts: 937
Joined: Thu Oct 18, 2012 2:49 pm


Return to Physics

Who is online

Users browsing this forum: No registered users and 2 guests

cron