The forces that act on the chassis build up with time. Similarly the flexing of the chassis builds up. In reality it doesn't do that in a linear way, because the chassis partly acts like a spring. A simplification that comes to mind is to assume a linear relation.
I didn't say that this simplification will yield all results that proper spring simulation would. My point is rather: How much would we miss if we would use this simplification?
The main results of bending would all be there. A different effect, loss of mechanical grip due to chassis oscilation, wouldn't. But the chassis oscilates at up to 50 Hz. That is above the framerate of many people.
That's where my question arises: How much would this simplification miss? How big is the influence of the oscilation around the static point?
That's the point that is beyond my knowledge and the main point why I created this thread.

Vain

Quote from Vain :

But the chassis oscilates at up to 50 Hz. That is above the framerate of many people.

But that is not related. LFS runs it's physics engine at 100Hz from what i remember, it has nothing to do
with the fps. I've read a few times how it's pointless to calculate anything over the refresh rate since it
would be 'lost' anyways. That's just not true. If LFS ran it's physics at an even higher frequency, the
effects would still be apparent even if the fps would remain at 50Hz.

Where there IS problems is with the online predicition code which sometimes can't keep up. One prime
example is the infamous collision bugs where contact is made in between 'frames' resulting in unpredictable
outcome at the following 'frame(s)'.

Then there's the issue of rounding off numbers, the higher the update frequency, the more accurate the
numbers and the smoother the result.

http://www.freebyte.com/cad/fea.htm#fea

Here you go. All they need to do is take one of these and incorporate it into LFS. Done.

Quote from Fonnybone :

I for one doubt it is realistic to think LFS should calculate every possible
forces acting upon the whole chassis. LFS has no physical chassis for
starters, basically what looks like anchor/pivot points held together in
a solid structure by an invisible 'chassis'. The simplest and most effective
way of simulating chassis flex would be to move the anchors/pivots
slightly according to forces. A simple box where each corner is at the
suspension anchor points would work well as a simplified structure. Having
the forces acting upon the box and deforming it would move the
suspension points accordingly which would simulate most of the side-
effect of chassis flex we are interested in anyways.

Bingo. That's how I'll do it for Kartsim.

LFS Tire physics run at 2000Hz and the general physics run at 100Hz.

Quote from Vain :

3.
That is because we don't use springs, aka rods, but bars. We simplify the car chassis to three bars of material. One from front to rear axis, one at the front axis, one at the rear axis, 3 in total. In elasto-mechanic theory a bar can do all deformations that a car chassis can do. And we know very well how that bar aka chassis behaves. The simplification is minor, but the change in CPU usage is major. 'Static' chassis flex (without swinging at up to 50 Hz) would utilize less CPU clocks than a single LFS tyre does.
Even without the effects of dynamic oscillation we get several nice effects:
Single seaters can bottom out under chassis flex, camber increases under high load, body roll increases due to chassis twisting.
If someone has knowledge about the effects of chassis oscillation, please share.

Vain

What about lateral twist etc. 3 bars is NOT enough. Also you shortcut is not that easy. All metal has spring like properties and this is what causes chassis flex so you need to calculate it as though it is springs. Because the springs are VERY rigid it means you have to run at high frequencies or the calculations will be wrong and your bars/springs will not act as you expect.

Think recording music. Sample at less that twice the highest frequency sound in the music and you will get distorted sound when you play your recording back. The lower the sample the worse the error.

Car chassis flex involves very high frequencies because its rigid. The more rigid the faster you need to calculate

Firstly we can leave out the bending and twisting of the suspension elements. If you measure some chassis flex under dynamic/static stress, probably 1% of it is caused by the actual suspension elements. (If they are road legal ).

Secondly we can leave out the deformation of the wheels (the metal part in the center ). Wheels can bend but the reason for it is usually some kind of accident, ie. in real life you accidetally drive against a curb and the wheel may bend.

Now we have a rigid suspension attached to a deformable frame. Where and how does the frame deform? Let's think the car as a very basic skeleton, consisting only from the axles connecting the wheels and a third axis connecting these axles (attachment1). So the body is thought to be a deformable structure made of three bars and wheels.

Then let's think the chassis is a box made from these same bars.

Now we have two models. Let's give it some force!

In the picture I have marked little arrowheads below the tires describing how they are "attached" to the ground. Basically it is meant to show that the car has 3 tires on a flat, even surface and the fourth is being pulled upwards (maybe a bump under the wheel). Now let's say that we lift some much above ground. (think it as a deformable frame with no suspension)

Now, I'll ask a question from all you who think that chassis flex can be modeled by using couple of bending and twisting bars? Do these two structures deform similarly?

They do act differently under load. One being basically a simple bar structure, while the other is a 3d frame. But both are still quite simple, if you compare it to the real thing.

Btw. does FXO for example have tower strut brace or tower to firewall braces? You can't model even these by using some very plain model.

Your right in saying that the two structures will react in different ways. But for simplicity's sake surley the single backbone technique qill easily get the basic principles across.

Your box chassis has twenty points of rotation, 8 of which have 3 connections. Sure it will handle differently from the 6 point single backbone chassis (Which has been a technique used for many many years in chassis building. I renovated an old Triumph Spitfire which had a backbone chassis which we painted pink for some reason.) Your open box technique is more represtentative of modern spaceframe chassis.

But where do you stop? Monocoques? Steel, aluminium, carbon fibre, Titanium? Body shell involvement. Roll cages.

Is the simple chassis example good enough for all intents and purposes to get the general idea of Chassis flex without destroying the supposed behaviour of the cars and without complicating matters too much in regard to complexity, cpu load and the like.

With the three 'spring' chassis you will be able to get so many different handling characturistics just by adjusting the arbitrary figures on each spring and the rotational values on the connections. Surely as a simple starting point this would be more than adequate.

Quote from Woz :

Think recording music. Sample at less that twice the highest frequency sound in the music and you will get distorted sound when you play your recording back. The lower the sample the worse the error.

Car chassis flex involves very high frequencies because its rigid. The more rigid the faster you need to calculate

That's why we use low pass filtter with audio. I don't know any reason why it couldn't be used with chassis flex math also.

TBH, i think Suspension mounting flex is more important than chassis flex, are the mountings modelled?, the higher up the race cars you go (gti compared to BF1) the less mounting flex your going to get (probably neglagable on BF1) i bet on most road cars the wheel can move and inch or so backwards and forwards and the strut flexing making the camber/caster flex too, on rear trailing arms probably some in out movement aswell. All because of the rubber mountings

Troy

Yes, that's true, bushings used at pretty much any suspension connection
in real cars do affect handling/response a lot.

As for the chassis, i like the single backbone idea, although it is still
imcomplete. I've tested it for stress but it cannot simulate the upper
suspension mount flexing (the reason why many people fit strut-tower
braces...). I'd add a link at the each wheel going from the lower to the
upper suspension anchors. I made a quick 3D model and i'll make one to
test for stress.

Interesting. I didn't originally think about modelling moving suspension parts, but the extra-work seems to be minor.
There's still no one who can shed some light on the severity of the effects of chassis-oscillation to the car's dynamics?
The simple 3/7 bar model is capable of all types of chassis-flexing I can imagin and would show all major effects.

@Fonnybone:
Firstly, what software are you using? It looks like it could spare my pencil a lot of work. Secondly, the mounting of the central bar seems strange. Is it one single bar from front to rear axis or actually two bars mounted to a central "anchor point" in virtual space? The tension the software indicates reaches to this central box and is zero from there on. One single bar would spread the tension over it's complete lenght, even into the (in this view) rear axis, which would also (slightly) bend. Or am I just mistaken because I don't have experience with that software?

Vain

Quote from Vain :

@Fonnybone:
Firstly, what software are you using? It looks like it could spare my pencil a lot of work. Secondly, the mounting of the central bar seems strange. Is it one single bar from front to rear axis or actually two bars mounted to a central "anchor point" in virtual space? The tension the software indicates reaches to this central box and is zero from there on. One single bar would spread the tension over it's complete lenght, even into the (in this view) rear axis, which would also (slightly) bend. Or am I just mistaken because I don't have experience with that software?

Vain

The software is APM Structure3D. I'm always looking for a nice simple program to test stress on a simple
chassis, this is the only one i found so far and don't quite have the hang of modeling in it yet. As you've
noticed, i've modeled the central bar out of 2 pieces simply to provide a node to attach a movement
contraint. To apply a force and see the stress graph, i had to 'fix' a node. The added advantage is that
i can apply another force on the 2nd part for comparision or added info. The stress would indeed have
continued down the main backbone.

Quote from askoff :

That's why we use low pass filtter with audio. I don't know any reason why it couldn't be used with chassis flex math also.

So what you are saying is we should filter,reduce and cut out any high frequency oscilations in the chassis because this is what a low pass filter does. In a rigid body all the freqencies of ossilation are high due to spring rate.

In music you have the wave you are recording so it can be filtered cheaply but still has a cost on sound. In LFS it would need to do all the calculations first and then apply your "filter" so that would just add more work.

That might be a good reason why it could not be used?

Although i understand the relation between harder/stronger materials and
frequencies, i'm having a hard time visualizing it. We aren't looking for a
huge amount of deflection but it does need to be precise else it'll make
the chassis jerk and possibly affect handling as it loads/unloads.

Quote from Fonnybone :

Although i understand the relation between harder/stronger materials and
frequencies, i'm having a hard time visualizing it. We aren't looking for a
huge amount of deflection but it does need to be precise else it'll make
the chassis jerk and possibly affect handling as it loads/unloads.

Think of it this way...

Draw a sine wave on grid paper where one cycle is 12 cells on the paper.

1) Mark dots in one colour at every point the wave crosses the virt grid lines then join those dots. (Sampled at 12 times frequency)
2) Mark dots in another colour and plot but only sample every 4 virt lines. (This is 4 times the frequency)
3) Now plot another set but sample every 8. (Twice the frequency)
4) Now plot every 12.

You will soon see that the nearer you get to the frequency the less of the wave you capture and the less possibility you have of getting the full extremes of the wave. When you sample at the frequency you just get a flatline.

Now turn on its head. If we are calculating the points the longer between the calculations the less detail and so more chance of incorrect values getting into the maths because the interpolation will introduce errors.

This is like the pinball crash issue where the collision detection does not fire quick enough and hence when the crash is detected the two surfaces are inside each other and so get interpreted as a far greater force than expected. If the maths had run faster the greater the chance of detecting the moment of impact and not past it so the better chance of getting the forces correct.

So, not doing this right could cause very strange and unexpected handling issues. I think the ROR developer said the rods calculation is done at 2000Hz and this is still not quiet fast enough which results in the body being less rigid than it should be.

Quote from Funnybear :

Funnybear:

Your box chassis has twenty points of rotation, 8 of which have 3 connections. Sure it will handle differently from the 6 point single backbone chassis (Which has been a technique used for many many years in chassis building. I renovated an old Triumph Spitfire which had a backbone chassis which we painted pink for some reason.) Your open box technique is more represtentative of modern spaceframe chassis.

But where do you stop? Monocoques? Steel, aluminium, carbon fibre, Titanium? Body shell involvement. Roll cages.

Is the simple chassis example good enough for all intents and purposes to get the general idea of Chassis flex without destroying the supposed behaviour of the cars and without complicating matters too much in regard to complexity, cpu load and the like.

Let me make one thing clear . If you use that very basic model as your main source of information, you will get unrealistic deformation, simply because the model isn't realistic. And the open box model I presented was just to show that even that could provide much more realistic deformation data than the very basic one. But true, yes, it looks like a "modern" spaceframe chassis

Quote from Fonnybone :

The software is APM Structure3D. I'm always looking for a nice simple program to test stress on a simple
chassis, this is the only one i found so far and don't quite have the hang of modeling in it yet. As you've
noticed, i've modeled the central bar out of 2 pieces simply to provide a node to attach a movement
contraint. To apply a force and see the stress graph, i had to 'fix' a node. The added advantage is that
i can apply another force on the 2nd part for comparision or added info. The stress would indeed have
continued down the main backbone.

I would be very careful when using the FEM simulations made by people who aren't familiar with them. There are some evil things which may make the results totally unrealistic if you aren't aware of them. A FEM program is not to be used to just create nicely colored 3d-models

And it really doesn't matter how complex the base model is where you get the deformation data under different conditions, the main thing is to make it compact and optimized in code-wise. You just need to know what are the biggest factors that cause most of the chassis flex.

The toughest challenge is still the frequensies

(yeah, forgot totally the bushings :shy

You may want to get as much accuracy out of any chassis flex simulation mate but untill it can be determined that LFS can actually cope and deliever a comprehensive and detailed ouput of chassis flex then surely the simple backbone method will give enough responses for any layman racer to get some benefit.

It's all well and good making yourself perfectly clear but if by implementing full chassis flex you take up a vast amount of resources then what is the point.

Like I have said already, go beyond the 'simple' get the idea across method where do you stop?

Can you design and engineer a fully working chassis for the FZR? Factor in all the different materials spring rates, work out the loading deflection accuratly and make it work using the minimum of resources.

Better off starting simple, the backbone, and working from there. Get the frequencies sorted and if all that works then you can add in more components. It's more than adequate to get the general point across.

It'll give you everything you expect, twist, grounding out, and flex. The given values are adjusted for each vehicle to give appropriate reactions. I would be happy with that.

Quote from Funnybear :

You may want to get as much accuracy out of any chassis flex simulation mate but untill it can be determined that LFS can actually cope and deliever a comprehensive and detailed ouput of chassis flex then surely the simple backbone method will give enough responses for any layman racer to get some benefit.

LFS won't be able to give comphrehensive and detailed output of chassis flex. That goal would be simply too high. However LFS is about realism, not about effects. You won't get realism from a simple backbone model, you get effects.

Quote from Funnybear :

It's all well and good making yourself perfectly clear but if by implementing full chassis flex you take up a vast amount of resources then what is the point.

I am not suggesting that LFS should get "fully implemented chassis flex" as you said. I am suggesting that you need to have reliable info to make the in-game part work and act realistically. Optimization is the key here. Leaving out all the stuff that doesn't show is critical. But you need to know what to leave out. And things need to be in right proportions.

Quote from Funnybear :

Like I have said already, go beyond the 'simple' get the idea across method where do you stop?

Can you design and engineer a fully working chassis for the FZR? Factor in all the different materials spring rates, work out the loading deflection accuratly and make it work using the minimum of resources.

Better off starting simple, the backbone, and working from there. Get the frequencies sorted and if all that works then you can add in more components. It's more than adequate to get the general point across.

It'll give you everything you expect, twist, grounding out, and flex. The given values are adjusted for each vehicle to give appropriate reactions. I would be happy with that.

The numbers need to come from somewhere. Otherwise it will be just guessing. And guessing doesn't make good physics.

Btw. if you use that very basic ("backbone") model, how does it react when the car is cornering on max grip with constant speed (driving a circle with constant speed - static situation)? How do the outer front wheel change its camber angles? Now think any real car and say that they do even remotely the same thing.

The backbone-model, as it now seems to be named, can do all demanded deformations from the first post. That's why everyone agrees on it.
Secondly, we can't possibly reach "realism", LFS doesn't do that now and it won't in the future. Our undertray is a value that multiplies wind speed with a specific value to get downforce. Our skidmarks are simple functions of slip angle and load without any influence of the roughness of the tarmac. Our tyres have no internal structure. We're using a Euler-Integrator to get our spring-behaviour, which is totally off the real thing but behaves in a similar way and consumes far less resources than the simultion of the real thing. There are always places where you need to simplify away from realism so you still get the right effects from which the behaviour can stem.
A realtime simulator isn't a place to demand perfectionism. We're doing numerical mathematics, if you want to call it like that.

I had a look at that software, Fonnybone. I'm looking at that idea of including mounting-bending, but won't be able to do much on it today. Perhaps I can present something in th next few days. I need to run some tests to see what model reacts correctly to mounting-bending (though I'm not too confident).

Vain

Quote from Hyperactive :

I would be very careful when using the FEM simulations made by people who aren't familiar with them. There are some evil things which may make the results totally unrealistic if you aren't aware of them. A FEM program is not to be used to just create nicely colored 3d-models

I'm not sure what 'evil' you are talking about really. I studied in industrial
design and had software to test our models for deformation and such. I
think it's essential to have a basic understanding of how shapes behave
structurally. The idea is not to have actual values, but to compare
different designs using the same applied forces and see which ones deal
with them the best. I'm now into welding and plan on making a few chassis
and will be using similar software to test out various versions and/or
improve them.

As for the 3D mesh, i did that in Rhino for those interested...unfortunately,
i haven't found a Rhino plug-in that can test for stress so i had to use
another program to get an idea of the deformation under load.

If you look carefully at my 3D model, you will notice that each wheel is
restrained on every axis by the chassis making it possible to simulate flex
and rotation on all axis. Don't let the skinny backbone fool you, it can
react just like a full chassis under torsion and bending.

Quote from Hyperactive :

Btw. if you use that very basic ("backbone") model, how does it react when the car is cornering on max grip with constant speed (driving a circle with constant speed - static situation)? How do the outer front wheel change its camber angles? Now think any real car and say that they do even remotely the same thing.

Mate. I'm not really sure what your trying to bring to this discussion. The original topic was along the lines of 'would chassis flex be noticable enough to warrent inclusion into LFS and/or if so how difficult would it be and to what level of realism'

I think that everyone agrees that to simulate a full chassis for each car would be too resource intensive. It just can't be done at this moment in time.

What seems to be possible is a very simplistic model that will 'simulate' what we are looking for. A given amount of flex, rotation and longitudinal bending. Sure it's not a true facsimile of a racing car's chassis but as pointed out above there isn't much in LFS that is made like a real world car. What is important is that the effect is implemented in a manner that reflects real life. No car in LFS has a real world conterpart (apart from thingy and the F1 wotsit), so nobody knows how these vehicles should react per se. But what can be noticed is if they feel like they should. If a simple backbone chassis model gets across the feeling that is required then a simple backbone model is perfectly adequate for our needs here.

It's all well and good to be able to say that LFS models a realistic chassis, but if you can't tell the difference between that chassis and a simplistic version then why waste the resources.

And in response to your last point, how does the tyre change it's camber angle now whilst it's just attached to a point in space? Just connect that point to a chassis model, ask the suspension to take inputs from both directions and bam, a working chassis / suspension interaction system. A very simplistic view I know, but one that shouldn't need to much work to get to operate properly. (He say's).

As the non techy here and possible a grounding influence, I would not be looking for ultra realism in chassis simulation, it's fine just to know that there is a system in place working in the best way it can to add to my online driving experiance. We have Chassis Flex in real life, LFS has chassis flex, it can't by definition operate in the same way, but it's there and I can play a better game for it. It's not nessesarily what LFS delievers that makes it realistic, it's the processes and the way it delivers the goods that make it what it is.

It would just add to the whole amazing world of LFS. Imagine having to adjust all your setups because you found out that the reason you can't get the BF1 to go around that high speed corner without bouncing 10 foot sideways is because the chassis is bottoming out. It adds a whole new area to take into account.


Cool.

Quote from Woz :

You will soon see that the nearer you get to the frequency the less of the wave you capture and the less possibility you have of getting the full extremes of the wave. When you sample at the frequency you just get a flatline.

Now turn on its head. If we are calculating the points the longer between the calculations the less detail and so more chance of incorrect values getting into the maths because the interpolation will introduce errors.
(...)
So, not doing this right could cause very strange and unexpected handling issues. I think the ROR developer said the rods calculation is done at 2000Hz and this is still not quiet fast enough which results in the body being less rigid than it should be.

Ya, again, i understand the frequency's relation with accurate values.
I also used the online prediction code as an analogy.

We are interested in simulating a relatively small movement, where even
an 'extreme' isn't much, at a fixed frequency, the LFS physics engine's
refresh rate. Both require a high-frequency to get accurate values.

The bottomline is the chassis must be included in an existing, complex,
yet relatively stable, structure.

Would the chassis be stiffer that anything already in LFS requiring the
physics engine to use a higher frequency to keep the whole car together ?

Quote from Fonnybone :

Would the chassis be stiffer that anything already in LFS requiring the
physics engine to use a higher frequency to keep the whole car together ?

Yes, the frequencies are much higher in the body than say the susspension etc. As I said RoR runs at 2000Hz and he has stated that is not enough.

I would prefer no chassis flex than one that does not behave right and hence make the cars do strange things.

Prob more milage in doing flex around the suspension mountings TBH than the chassis ATM

Quote from Fonnybone :

I'm not sure what 'evil' you are talking about really. I studied in industrial
design and had software to test our models for deformation and such. I
think it's essential to have a basic understanding of how shapes behave
structurally. The idea is not to have actual values, but to compare
different designs using the same applied forces and see which ones deal
with them the best. I'm now into welding and plan on making a few chassis
and will be using similar software to test out various versions and/or
improve them.

The evil is the details. It is more pronounced when analysing 3d-models using 3d elements. Basically you get wrong kind and size of stresses and deformations on from model.

Quote from Funnybear :

Mate. I'm not really sure what your trying to bring to this discussion. The original topic was along the lines of 'would chassis flex be noticable enough to warrent inclusion into LFS and/or if so how difficult would it be and to what level of realism'

For the last time I say this: to get the chassis flex to act even remotely like the real thing you need to have a good base to make your model to in-game. You won't get valid information from that "too basic" model. With little more effort you can double or triple the calculation times and get better results. Even the base model is way too complicated to be the chassis flex part inside the game. And how a re you planning to get the suspension joint deformation data from that model? The model has attachments to the tires, not suspension joints...?

Btw. the original topic was: "Physics discussion - Chassis Flex - How complex is it?".

I am not trying to be tricky or knowledgeable or anything. I think chassis flex needs to be simulated in LFS. But not at all costs, and it should be on the par with other "simulational objects" in LFS.