| Author | Message | ||
     Jlnance |
Hey Dr. G, feel like shooting the breeze about control theory? I talked to yet another person who demoed a Uly w/o the suspension set up properly. They of course felt like it handled poorly. So I started wondering if the bike could tell you the suspension wasn't adjusted properly. That got to be a fun thought experiment, beyond any questions of practicality. Here is what I came up with. What I want to do is determine the transfer function of bike, using both wheels as input sources, and the motion of the frame at the front and rear as outputs. You'd fit the transfer function to idealized model of the bike, locate the poles and zeros, and decide if they were well placed. I think it would require 4 sensors (accelerometers) to do this. One apiece on the unsprung parts of the front and rear wheel assemblies, and two on the frame, one front and one rear. If this was a linear system, the curve fitting would be a piece of cake. Something like the LMS adaptive filter algorithm would work. The shocks in the bike have different damping for compression and rebound, so it's not a linear system. I don't know how to fit that. I think it could still be done in one of two ways. Since the accelerometers would tell you whether the shocks were in rebound or compression, you could treat it as a piece wise linear system. You would derive 4 transfer functions for the bike, corresponding to the rebound/compression states of the front and rear shocks. The other alternative is to use something like state space averaging and derive a single linear transfer function around whatever large signal operating point the is at. I think the first way would work better, because it allows you to separate the compression and rebound settings, which would allow you to tell the user which knobs needed tweeking. Thoughts? | ||
| Seanp |
And some people claim that there's no intelligent discussion on this board...  | ||
| Thetable |
I can see this coming factory option on one of the new Bimmers, but never a Buell. Hell, on the Bimmers, all it would probably have to do is plug into the CANbus to make all the necessary suspension adjustments for you. | ||
| Court |
It's been done . . . . oh how the "bread box" is missed . . . what a great mounting place it was for transducers and instruments.  HINT: When you get good you can run along side in a min-van with the door open and film. | ||
| Dr_greg |
Well, I dunno. A motorcycle is a surprisingly complex dynamic system. Sounds like you're proposing to treat it as a planar system, e.g. no out-of-plane dynamics. You would have to assume that any modeling you did would apply to the spatial case (e.g. tossing it into a corner). Might or might not be the case... And, as you realize, as soon as you say "transfer function" you're stuck with linear. That might not be too bad, since probably a linearization about an operating point would be adequate for getting mode shapes and such. Anyway, with measurement of F/R sprung and unsprung motion one could try to fit a transfer function matrix. I've done a fair bit of system ID work myself (and have a class project doing that right now). We're just using simple least-squares stuff, but one could do it offline using, say, the MATLAB ID toolbox. Usually we like to use an input that excites all the dynamics of the system, e.g. a rider that is constantly braking, accelerating, etc. Who might that be? Mrs. Greg gave me a book on "motorcycle dynamics" for Christmas, but I haven't looked at it much yet. I'd love to have some students do something like that. BTW, back in 1971 I was all set to do my Ph.D. dissertation at Stanford on "motorcycle dynamics" under the legendary dynamicist Thomas R. Kane but then Vietnam got in the way. By the time I was back in school there was research money in robotics, so that's where I went. If I had stayed with the motorcycle dissertation, I'd probably be writing those motorcycle dynamics books now... --Doc P.S. Tom Kane's opinion was that we were going to have to include the rider dynamics as well as the motorcycle's... | ||
| Jlnance |
Tom Kane's opinion was that we were going to have to include the rider dynamics as well as the motorcycle's I was thinking about that yesterday. You could probably tell the rider he was holding the bars too tightly if you refined the system enough. I'm headed out for a ride. Your post will be good pondering material. | ||
| Froggy |
I would love to have some kind of setup like that on my bike. | ||
| Tootal |
Froggy, does your laptop ever leave your bike? | ||
| Froggy |
Yes! I usually don't take it on my commute to work, but usually have it if I am venturing more than a hour ride from home. | ||
| Jlnance |
Dr G, I thought about your comments. I think you are right, the side to side motion of the bike is probably important, as that is a major energy storage mechanism, and it involves the forks. I wonder if switching the frame accelerometers to 3 axis devices would be sufficient, or if you would need to measure how much the bars were turned as well (is there a term for that?) I know it's possible to measure all this if you plaster the bike with enough sensors. Part of the challenge is to see how few you can use and still get the results you need. | ||
| Dr_greg |
Turns out the "Motorcycle Dynamics" book has a chapter on "in-plane dynamics"...maybe I'll look at it soon and see what they say, esp. regarding application to out-of-plane dynamics. Re sensors: as you know, the term is "observable"... --Doc | ||
| Jlnance |
Observable. Goes with Controllable I believe. I've had 2 controls classes, classical and modern. Classical was fun. Modern, which is where I learned about observable and controllable, was mostly about doing unnatural things to matrices. |
