I know it was a bit of a discussion when the BF1 first came out, but I figured it might be a topic worth discussing.

I was just watching the France GP and noticed they pointed out the Mercedes-Benz engine had broken the "20k barrier" for RPMs. It seems that every other manufacturer has troubles breaking this barrier aswell, and topping out above 20k optimally is not so common. Yet the BF1 in LFS is rather efficient for its power.

If you watch the way most of the cars in F1 perform when reaching high RPMs, they are just struggling ever so slowly to even creep near the 20k mark (more-so the 19k mark). On average they all shift around or below 19k I'd imagine.

It just seems odd that the BF1 has such a long reving engine in LFS versus reallife V8 F1 cars. Maybe something worth looking at? Or just the limitations of the virtual world?

its because engine damage isnt implemented fully yet.. so maybe if it was the rev limiter would come into play more ?

It's not about damage, it is about performance.

Getting similar RPM output from those V8 engines as with the old V10s is just not the same. If you watch telemetry and onboard statistics you can see that the engines just pretty much 'round off' their RPM curves around 19k. It isn't limited, and they really cannot go any higher as far as I know. Then take a spin in the BF1, ermmm, it has some different characteristics that seem a bit off.

but i've heard talk of teams/drivers allowing the car to run higher revs under certain situations like when trying to pass. yes its a power issue thats obvious ...but that also

Generating torque curves from an equation (as you do in a sim) makes it difficult to get any shape you want. For instance many manufacturers aim to get a very flat torque curve across a broad rpm spectrum from their engines. If using a polynomial equation (for example), it's impossible to get the same output value from a wide array of input values. Likewise after the engine passes peak power the power can only die off so fast.

...gear ratios?

Nah, what he means is that if you look at a torque curve for a real engine, it can be a broadly arbitrary shape with an irregular pattern to it.

The torque curve for LFS cars is generated from an equation using a few variables as input (those variables we play with in LFSTweak). This is a very flexible system as it allows quick and easy generation of new engine types, but the curves themselves are quite simple and mathematical in appearance. The current equation doesnt really allow for complex torque curves and so lacks some realism.

Other sims basically use a look-up table with a values of torque for each level of RPM, and the engine interpolates between them. Which is potentially more realistic, but isnt flexible or quick to alter.

Perhaps LFS could get the best of both worlds by keeping the equation driven system, but adding an optional 'offset' lookup table, so that values could be manually tweaked away from the equation results through the lookup table.

Then you could still use the equation system for flexibility and speed, and then in final tweaking, use the lookup table to offset the values nearer to realistic values, and add little details like where engines change cams and that sort of thing.

Quote from colcob :

Nah, what he means is that if you look at a torque curve for a real engine, it can be a broadly arbitrary shape with an irregular pattern to it.

Yeah, this would be obvious on an engine as tuned to a particular rev-range as an F1 engine, whose power probably drops off PDQ after its peak.

It would be pretty cool if Scawen could copy curves into LFS from similar real-life car counterparts.

Edit: I bet they still try to short-shift as much as they are comfortable doing to preserve the engine. The Mercedes engine Tweak mentioned blew up at the end of the race... I think DLR was used to wringing his engine out in practise sessions because it didn't have to last two races, or even one.

Sorry to be a geek, but how many degrees in the polynomial equation that LFS uses? You wouldn't have to go very far before some pretty arbitrary curves could be represented.

I know from a friend of mine that the Mercedes engine will do over 20k easily. It's just the endurance that's the problem.

That's a good suggestion Colcob. Not the most realisitic solution (in terms of just modelling the engine perfectly in the first place), but could be very useful as a stop gap.
Or perhaps just adding a few set of curves together would help?

Quote from jtr99 :

Sorry to be a geek, but how many degrees in the polynomial equation that LFS uses? You wouldn't have to go very far before some pretty arbitrary curves could be represented.

Scawen is the one to ask that one. It's not even necessarily strictly polynomial, could be some weird thing, although I have to say the torque curves back in S1 did look somewhat cubic or quartic.

Mercedes is taking allmost everything out of their engines at this time of the season, they only aim to win or to podium, and thats because all the titles are alleady been lost. So, unlike Renault or Ferrari they set their revlimiters higher because they just don't care if it blows up. Other thing in race what effects the performance is the fuel injection which can be limited, and thatoffcourse makes the revving to max more slow. And the Mercedes engine isn't that uber, I think I saw Ferrari and Renault taking 20k or over out of their engines in the qualifying session. BF1 in LFS revs what, about 21k?, I wouldn't say it couln't be possible in real car but they wouldn't take that much out in races because those engines need to last two weekends. So...LFS is the optimal situation and gives the optimal values, atleast until we get engine/gearbox/etc. damage.

I think that engines can't really go higher than that and output usable torque is because of the most important factor in an internal combustion engine:

The fuel.

ive seen a vid clip of a F1 engine hitting upto 23k RPM on a test bed, cant remember which manufacturer it was tho..

Quote from jtr99 :

Sorry to be a geek, but how many degrees in the polynomial equation that LFS uses? You wouldn't have to go very far before some pretty arbitrary curves could be represented.

you will always have troubles with creating arbitrary curves out of polynomials ... youll either get very high order polynomials or oszillations or more likely both

Quote from Shotglass :

you will always have troubles with creating arbitrary curves out of polynomials ... youll either get very high order polynomials or oszillations or more likely both

I appreciate your input.

I guess it depends how far one wants to go down the road to arbitrariness. It seems to me that given the shape of typical power band curves, such as this one:



... a polynomial of say, degree 5 to 10, would be more than enough to get a decent fit. You might call a 10-degree polynomial "too high", and that would depend on your purposes, but we're not talking about a great deal of computational overhead to compute values here.

What am I missing?

its of course possible but as far as i can tell lfs works by taking some easy to understand values that express the shape of the curve and toy around with them until it looks the way you want it to
using higher order youll probably end up drawing the curves and then aproximating it untill you get a good fit
for develoment purposes and car balancing its probably better to use low orders that help you to express the curves behaviour in simple numbers which are easi to manipulate

Quote from Shotglass :

its of course possible but as far as i can tell lfs works by taking some easy to understand values that express the shape of the curve and toy around with them until it looks the way you want it to

Ah, I see the reason for me not getting it. You're talking about users being able to easily generate a desired power curve in an application like LFSTweak, I guess. Whereas I was thinking more about behind-the-scenes use of a polynomial as one possible way to model an arbitrary curve (presumably using some kind of least-squares technique to fit it to empirical data). Although having said that the model output might actually be precalculated for different RPM values and end up in a look-up table for faster computation in-game.

I completely agree that an order-10 polynomial would not be the easiest thing to play around with manually in order to get a desired curve.

Order 10 would be a nightmare. I've written code for a cubic and that nasty, I want to switch to quartic which frankly make my forehead ache just thinking about it. Order 10? No way.

I would imagine LFS would go down the simulating various bits of the engine, all with their own (simple) polynomials, which would end up with a realistic looking torque curve.

For an engine that has max revs (=redline) around 8k, the racing rpms are usually between 3500-7800 rpms. If you get this part realistic with simple polynomical functions there is no need to make the power curves etc. outside this any more complex as required. Getting 5hp/10Nm more or less at 2025.4 rpm doesn't matter as you should never run the car on those revs in a race

And you can use more than just one curve to get the right kind of "slope". But I think the derivates (1st and 2nd...) should be equal at the "continuing points" because otherwise a) it might cause problems later when drivetrain stress is introduced (stress peaks) and b) hmm, can't think anything else, but...

Remember, you need the derivates of those curves to calculate accelerations and turning forces (flywheel and drivetrain masses)...

Quote from Hyperactive :

And you can use more than just one curve to get the right kind of "slope". But I think the derivates (1st and 2nd...) should be equal at the "continuing points" because otherwise a) it might cause problems later when drivetrain stress is introduced (stress peaks) and b) hmm, can't think anything else, but...

in other words cubic splines ... which is presumably how isi does it if they save the curves as tables

Quote from Shotglass :

in other words cubic splines ... which is presumably how isi does it if they save the curves as tables

Haha, no, nothing so elegant. Its just a huge text lookup tables with values about every 50 rpm. Not sure what form of interpolation is used between the values though.

a 16bit integer for every 10 RPMs results in a 4KB table for an engine with 20K RPM range. even if you have 10 different cars racing, the entire datatable fits easily in most modern processors' L2 data cache.

Nothing to cry about.

IMHO? no, i'm not a hoe... but a 5th-6th order poly is enough to model the torque/power curve.

But i'd like to see every single piston firing be modeled...(IMNUTS)

Quote from shim :

ive seen a vid clip of a F1 engine hitting upto 23k RPM on a test bed, cant remember which manufacturer it was tho..

It was the Cosworth engine. Pretty much all of the F1 engines are capable of it, but only for short periods of time. It's not worth the risk of blowing and engine and getting a 10 place penalty in the next race. The drivers will use different rev limits at different times, you can often hear the engineers telling the driver that he's allowed to "turn the revs up" when he needs to catch someone, or when he's qualifying.

Quote from colcob :

Not sure what form of interpolation is used between the values though.

as i said ... probably splines