Using evolutionary algorithms to optimize wheel lacings

[thegunner]

http://master.matsemann.com/

not sure how much of the paceline demographic intersects with the hackernews demographic, but with the uptick in computational analysis and modeling to design things lately, it was only a matter of time before it leaked over to cycling.

interesting how expensive the modeling was - 300 generations with 10 individuals, so 3000 runs over 40 computers took 6 weeks.

[ergott]

I'll have to give that some further investigation.

[thegunner]

I'll have to give that some further investigation.

for the sake of innovation, i would love if one of these methodologies pumped out a new lacing that turns the 32h 3x lacing entirely obsolete, not because i don't believe in tried and true, i just want tech to succeed where human logic couldn't.

[witcombusa]

It won't replace my 32h, cross 3 wheels...:beer:

[ceolwulf]

Won't replace my 16 spoke radials ;)

[Keith A]

Looks interesting.

[RFC]

Anybody who mentions "evolutionary algorithms" will love this.

This model uses a genetic algorithm to evolve a two wheeled vehicle to move farther and farther down the track.

Fire it up and let it run. Mesmerizing!

http://rednuht.org/genetic_cars_2/

I use this at the beginning of my biotechnology law class just to get their attention.

[Kirk Pacenti]

I'll be looking into this as well.

Fwiw, our modeling for MTB wheels shows:

28 x 3x is adequately stiff while offering a worth while weight reduction.

32 x 3x is a little stiffer but (when viewed on a spec sheet) significantly heavier.

24 x 2x is much lighter but has a precipitous drop in stiffness.

ymmv.

Cheers,
KP

[thegunner]

i think the interesting thing that they're also looking into (although it's not them by design, it's the algorithm) is the abandoning of radial symmetry - some of their lacing patters are in triplets of 120 degree arcs akin to a tri-spoke.

kind of neat. i'm also impressed that the guy managed to turn something he loved into a legitimate thesis like this.

[zmudshark]

I saw this on Reddit this AM. FWIW, he says that 3X is generally the 'best'.

https://www.reddit.com/r/bicycling/comments/3p8hua/i_managed_to_make_my_thesis_in_computer_science/

[Tickdoc]

http://master.matsemann.com/

not sure how much of the paceline demographic intersects with the hackernews demographic, but with the uptick in computational analysis and modeling to design things lately, it was only a matter of time before it leaked over to cycling.

interesting how expensive the modeling was - 300 generations with 10 individuals, so 3000 runs over 40 computers took 6 weeks.

Very cool.

Makes me wonder what a skynet designed bicycle and wheel combo would look like....guess we'll have to wait til the machines take over to see.

[texbike]

It won't replace my 32h, cross 3 wheels...:beer:

Won't replace my 16 spoke radials ;)

Forget them computers! I'm sticking with my Rev Xs! :D

Texbike

[Keith A]

Those were the best :D

http://41.media.tumblr.com/tumblr_ljnjz2Zkzv1qj3lwto1_500.jpg

[thegunner]

half the spokes, half the drag (not really)

[oldpotatoe]

I'll be looking into this as well.

Fwiw, our modeling for MTB wheels shows:

28 x 3x is adequately stiff while offering a worth while weight reduction.

32 x 3x is a little stiffer but (when viewed on a spec sheet) significantly heavier.

24 x 2x is much lighter but has a precipitous drop in stiffness.

ymmv.

Cheers,
KP

Owing to just 4 more or 4 less spokes, this 'significant' weight differences?

IMHO 28-30 grams isn't significant on a multi hundred gram wheel.

[Kirk Pacenti]

Owing to just 4 more or 4 less spokes, this 'significant' weight differences?

IMHO 28-30 grams isn't significant on a multi hundred gram wheel.

Yes. When read on a spec sheet (as I said above) 30g is pretty significant. You know, like when people are comparing specs from one maker to the next. That amount of weight could be the difference between making a sale or not.

Fwiw, I've never weighed any of my own wheels, or bikes... So from a personal / real world pov, it doesn't really matter (at least not to me). However, when it comes to commerce it absolutely does matter.

again, ymmv.

Cheers,
KP

[oldpotatoe]

Yes. When read on a spec sheet (as I said above) 30g is pretty significant. You know, like when people are comparing specs from one maker to the next. That amount of weight could be the difference between making a sale or not.

Fwiw, I've never weighed any of my own wheels, or bikes... So from a personal / real world pov, it doesn't really matter (at least not to me). However, when it comes to commerce it absolutely does matter.

again, ymmv.

Cheers,
KP

Yup, unfortunately it does, 'matter' to sales lizards.

[MattTuck]

Perhaps the funniest part (so far) of the thesis.

However, having 32 well-tightened spokes exerting strong forces in multiple directions on the same object was something the physics engine did not particularly like. It was often unable to converge to a solution, instead it would start vibrating and eventually explode. (page 42)

[Kirk Pacenti]

Yup, unfortunately it does, 'matter' to sales lizards.

What was significant to me (with our models) is that the 28h x 3X wheel had adequate strength and stiffness for the job at hand. In this case, very hard AM/DH riding and racing.

While I am not a weight weenie, I am quite interested in optimizing designs. In other words, "stronger than strong enough" or "stiffer than stiff enough" offers very little value, imo. Put another way, a 28h x 3X wheel meets or exceeds our needs, using a 32h x 3X wheel only adds cost and weight with little appreciable benefit.

I suppose the same arguments were made when road wheels started transitioning from 36h x 4X to 32h x 3X...

Cheers,
KP

[ultraman6970]

Wonder which company will pay some dough to this guy to use his work to develop a super comfy and stiff wheel for 4000 bucks :p

[oldpotatoe]

What was significant to me (with our models) is that the 28h x 3X wheel had adequate strength and stiffness for the job at hand. In this case, very hard AM/DH riding and racing.

While I am not a weight weenie, I am quite interested in optimizing designs. In other words, "stronger than strong enough" or "stiffer than stiff enough" offers very little value, imo. Put another way, a 28h x 3X wheel meets or exceeds our needs, using a 32h x 3X wheel only adds cost and weight with little appreciable benefit.

I suppose the same arguments were made when road wheels started transitioning from 36h x 4X to 32h x 3X...

Cheers,
KP

I get that but with some larger flange hubs, spoke/flange overlap can be an issue. I look at it as '32 3 cross, exceeds our needs, 28h reduces the cost and weight a small amount and still meets the 'needs' of a wheel.

I'm not in the business of rim design, thankfully. Tullio Campagnolo once said, if he knew how hard it was to make rims he wouldn't have done it.

Jobst till his dying day said the 'optimum' wheelset was 40h rear, 32h front.

I'm just glad there are still choices for rims, hubs, spokes. Seems more now than ever..mostly from the poor performance of so many wheelsouttaboxes methinks.

Jack of all trades, master of none.

[Mark McM]

I had a quick read/skim through the thesis. The main aim of the thesis was to explore evolution simulation methods for optimizing design. The selected design was spoked bicycle wheels, but just about any mechanical structure design could have been used. As with any simulation, the results are only as good as the model. Unfortunately, the model of a bicycle wheels was too simplistic. It assumed a rigid rim, but 'rigid rim' as the previous modeling efforts have shown, the 'rigid rim' model is not simulate real wheel well, and gives invalid results. So, I don't think this particular paper can shed any light on how to design real wheels.

The thesis references a previous paper published by Williams wheels, which had shown that 3 cross lacing was the best spoke pattern. This paper used a more realistic wheel model, which considered the radial and lateral flex in the rim. However, I'm not sure we can conclude anything from this paper either, because it doesn't provide much information on the loading conditions. It does mention the value of the radial (weight) load, but it doesn't mention the torque load used, other than to say that it was "hard pedaling". The affect of drive loads on wheels is often over-estimated, so if the torque load applied to their model was unrealistically high, it skew the results.

[thegunner]

Unfortunately, the model of a bicycle wheels was too simplistic. It assumed a rigid rim, but 'rigid rim' as the previous modeling efforts have shown, the 'rigid rim' model is not simulate real wheel well, and gives invalid results. So, I don't think this particular paper can shed any light on how to design real wheels.

ha this is why the 32 3x wheel explodes.

but like you alluded to, it's more about the evolutionary algorithm backing it and not the mechanical engineering involved that's impressive. even an overly simplistic model took 40 CPUs over 6 weeks of computation time.

[marciero]

I had a quick read/skim through the thesis. The main aim of the thesis was to explore evolution simulation methods for optimizing design. The selected design was spoked bicycle wheels, but just about any mechanical structure design could have been used. As with any simulation, the results are only as good as the model. Unfortunately, the model of a bicycle wheels was too simplistic. It assumed a rigid rim, but 'rigid rim' as the previous modeling efforts have shown, the 'rigid rim' model is not simulate real wheel well, and gives invalid results. So, I don't think this particular paper can shed any light on how to design real wheels.

The thesis references a previous paper published by Williams wheels, which had shown that 3 cross lacing was the best spoke pattern. This paper used a more realistic wheel model, which considered the radial and lateral flex in the rim. However, I'm not sure we can conclude anything from this paper either, because it doesn't provide much information on the loading conditions. It does mention the value of the radial (weight) load, but it doesn't mention the torque load used, other than to say that it was "hard pedaling". The affect of drive loads on wheels is often over-estimated, so if the torque load applied to their model was unrealistically high, it skew the results.

So nice that someone read it! Indeed-good points. There are assumptions with any model used. Some other thoughts:

Genetic algorithms are iterative schemes that mimmic evolution by selecting out desirable characteristics at each iteration. The way the selections are made at each iteration, how performance is evaluated-probably by simulation, which again depends on model, exactly which performance criteria are evaluated, and other issues all conspire in the results. Finally, these types of algorithms are generally used in situations with many "locally optimal" solutions (like hilltops in a landscape with many hills) and are not guaranteed to converge to "global optimum" (the highest hilltop)even with ideal model assumptions and implementation.