Hoover Craft using 3 PID controlled Air Jets

You are right, this sounds cool. Your are right in that the loops will interact to some extent. I would not use three independent controls. I would use a cascade control which told the three PID controllers what to do. I might also make sure that the ramping on the three was very similar and a bit slow.

Michael

Hoover is a brand of vacuum cleaner. I suppose you mean hover craft?

Well, a Hoover could do it, too. But the OP may have misread the assignment. Hovering is the easy part; a model airplane engine can levitate a heavy steel beer tray.

Getting the craft to accept nonuniform loading and hover level is almost as simple; four chambers instead of one, each fed through a common supply through a restriction that develops half the available pressure drop. The bottom can be mostly solid; you only have to feed the periphery of each chamber, and you can use a bristle brush strip to seal them from each other and the atmosphere. There used to be a good design paper on hydrostatic bearings from the NonFerrous Founder's Society that would be very useful here. At the pressures needed to lift a hovercraft, air is fairly close to incompressible, so the paper would be on point.

The more difficult problem is getting the hovercraft or beer tray to stay in one place, especially on a floor that's not quite level, or to navigate to places where you want it to go. For which I would use servo controlled air bleeds from the levitation chambers to produce small amounts of jet thrust. Using three jets would just drive you crazy with interaction; I'd use four, and just open them one at a time to go N,S,E,W.

Okay, there's one more degree of freedom that has to be controlled; rotation about a vertical axis, so you can point the payload or the pilot in a useful direction. For which you need four more bleed jets, controlled in pairs, arranged tangentially.

Sounds like fun.

-Mike-

This is less of a traditional hovercraft, and more of a VTOL-type of vehicle (at least according to my reading of his description). Therefore the stability would be entirely dependent on control of the air jets, and not so much on the air cushion principle.

In my opinion, this is a horribly complex project to complete in a few weeks. Note that once the platform tilts in any direction, the thrust vector will change and the PID loops need to account for that as well. You also need to be able to do this on purpose, in order to move in a given direction. Another gotcha: the distance sensors will also tilt with the platform, so you need to do 3-D geometry calculations on each sensor. There is a speed issue here, you'll need to take a lot of samples and do a lot of arithmetic in every cycle. If you aren't fast enough, the platform will flip over in midair or dive one corner into the ground. It's inherently unstable. Think Segway, but instead of two wheels there is just a ball, and the user wears rollerskates.

It may be possible to do if this thing only has to slowly rise up, hover, and go down. Probably will shimmy and drift to the side if the control doesn't take into account the thrust vectors in 3-D from each jet.

By intentionally angling the two front jets to oppose each other equally, and the rear jet to oppose the front two jets, some sort of bank-turning should be possible.

It is not an easy project.

If you consider the overall system, the dynamics of the three jets are clearly nonlinear coupled. However if you linearisize about small angles the three position/velocity state-pairs for each jet will be decoupled. If you make a stability analysis on the independently closed loop systems you will most likely be able to guarantee local stability, but not global stability.

This means you will archive stability as long as no external force make the platform wobble too much and the initial states is close to the linearization point.

In other words: 3 independent PID-controllers might work :) Good luck with the project!

Christian Gutvik MSc Engineering Cybernetics PhD canditate Medical Technology

============================================== Baromedical and Environmental Physiology Department of Circulation and Medical Imaging Norwegian University of Science and Technology

Hi Michael. I dind't realize my boss said Hoover and not Hover! there's goes a week of designing. ;-) (ever feel like your in a Dilbert cartoon? scary, ain't it? for holloween my manager should get a pointy-haired manager mask...)

i was thinking there's a role for cascade control here, but i'm not sure where the feedback comes from for the second loop. i could add sensors on the jets themselves, down-stream of the prop-valves, (otherwise, the height sensors are "independent") but not sure if that's neccessary/optimal. just what would those loops look like? some individual control seems necessary since each jet loop won't exactly act the same for a given setting (mech. tolerances), so levelness will probably have to be tuned in a bit. (there's a second measure of height/levelness that's entirely independent, but extremely slow, we intend to use that as the initial calibration baseline)

more details: this is a pretty stiff system, it's 12" square and floating 6microns above a flat surface (!)(or how high i tell it too...) the mechs think there will be a resonance at some 40Hz, which will be in my control BW. horizontal displacement is tightly stage controlled. how the control system will respond to sudden horizontal shifts will be interesting, and critical to performance too, but probably also faster then the pneumatics can address....

thanks all! michael c

Michael,

6 microns is not a lot of distance. I was anticipating something where the height was at least in the order of the size of the object. I have a feeling that this might change my answer a bit. I think I would use a master control for coarse control of the pressure, and then use individual controllers for trim control at each of the three jets. The latter would be limited in their action to a few percent of the valve opening. If there was some way to add some capacitance to the air supply, then the interaction via the air supply would be minimized. By letting the trim control send their position data back to the coarse control, it would be possible to slowly change the coarse control to get the trim valves at about the middle of their range. Note that the controls on the trim must have a frequency 3 to 10 times faster than the coarse controls.

I guess that would make a strange control. 3 PID controllers controlling a valve limited to moving perhaps from 40 to 60 percent open, attempting to get the right height. A coarse controller would then be modifying the overall pressure, trying to get the sum of the trim valve openings to 150 percent.

There are obviously other ways to do this. I'm not sure if that one is the best.

Michael

Six microns is way less than the peak to peak roughness of ordinary objects that might be 12 inches square. It's also way less than the deviation from true flat of ordinary objects, etc., even if they have a mirror finish.

Six microns over 12 inches also doesn't allow much space for 'banking' before the corners or edges contact, and therefore doesn't allow a very large lateral component of any vertical thrust vector that's available.

I suspect that maybe someone has microns and mils confused.

Even given that, the levitation sub- problem does not require control loops; hydrostatic bearing principles in a passive system would suffice. And may be overkill. Ordinary shop air at 100 psi introduced between two flat steel plates not larger than 2 feet square will levitate a ton or more, well enough to rotate it by hand.

-Mike-

thanks everyone, interesting stuff. no, the 6microns isn't a mistake. it's for a proprietary piece of equipment in the ATE industry. but it's also nearly the same order of magnitude of flatness as a hard disk's GMR head flying over your disks (that's 1mm on a side (less now) and flying at about 10nanometers! big molecules get in the way!)

in fact, we're open loop now and are trying to close it, for a variety of reasons (some of which might be worth it....) sub-micron dial-up control of height would be nice, not to mention terrain following, as it were.

i'll start with independent loops. i'm mainly concerned with how the sensor's response curve will change given minor degrees of tilt and what that does to stability. should be interesting.

michael c