frictionless bearings ?

Gosh, why do you need one minute accuracy at 10,000 RPM?

Are you sure that shaft twist is the source of the error? I can think of a couple of ways to demonstrate if it is.

I can imagine a simple circuit that would apply a compensating time delay (less delay at high frequencies) that would compensate out most of the error you're seeing... IF you have a way to calibrate it.

If it is shaft twist, as it may well be, the concentric shaft thing sounds pretty good.

Tried washing the lube out of the bearings?

What's this for?

John

hmm to explain would be a bit more involved than i would like right now, and it would probably start a completly different discussion ...

its somewhat less than 2us I beleive, its averaged over quite some time with a dspic, however its too variable to be able to actually calibrate it, the error cuased by the time delay of the circuitry can be nulled/self cancelled.

adjusting the alignment of the bearings cuases a very wide swing of the error. as does lightly gripping the shaft. its the variability wich is the problem, particularly after reversing the rotation direction the error jumps about wildly. its only the change in difference of angle wich is of interest, rather than absolute position.

yes but unfortunatly my usual suppliers dont have anything like the right size tube etc that I can just use easily. might have to find a machine shop that can do it to tight tolerances but that sounds expensive to me.

magnetic bearings sound fun to play around with if only i had the time...

yes ... and they promptly destroyed themselves lol ! ball races also have moderate play in them unless they are preloaded, but preloading just seems to make the problem worse.

il give details when ive got it working...

Colin =^.^=

have you tried thrust bearings and a contact less tach ? the tach is just a U sender/Reciever LED hovered around a photo disc that is attached to the shaft.

..snip

thats sounds just like an optical encoder wich is what im using. theres little in the way of thrust on the shaft, I dont know that these type would be any less drag ? or if needle rollers would either.

Colin =^.^=

You're partly barking up the wrong tree-- if the shaft is changing speeds, the acceleration will result in some twisting forces, probably larger than the bearing friction.

You could reduce the friction somewhat by using two coencentric bearings, with the grease washed out of them and replaced by some very light oil or light silicone fluid.

You could also reduce the twisting by using a shaft with a better weight to stiffness ratio-- look at hollow carbon fiber arrows-- they're amazingly light and stiff at the same time.

Also note that you're asking for one part in 3600 accuracy-- even a thousandth of an inch of radial slop in the bearings is too much if the shaft is 1/3 inch. Most bearings have much more slop than that unless they're carefully pre-tensioned axially. Which leads to more friction.

the idea is to not use the shaft of the encoder but your own to support or what ever to hold what ever it is that's spinning that fast. i mean,i don't know what moving item you're trying to tach?

Hi, acceleration isnt an issue as the measurments are taken when it is at constant speed.

I had thought of concentric bearings, but would this work or would only one rotate ? a bit like parallelling power supplies lol, or capacitors in series.

yes a stiffer shaft would help, not sure if carbon fibre is actually stiffer than stainless steel wich im using, it certainly can take more load per weight, but the weight isnt an issue, maybe extra hardened steel if I could find some, or maybe diamond :o

the end bearings are carefully shimed to remove axial play, well ok, actualy the end blocks are just pushed in till its tight but not to tight. hovewer a few more bearings are needed to suport the rest of the shaft or it wobbles itself to destruction. The encoders are tiny so any play is magnified but the new faster encoders seem to come in tiny sizes, maybe I could roll my own fast large one, but I think this 'slop' would average out.

I seem to have made a slight error in thinking 1 minute was 1/60 of a degree doh! Im actually seeing an error of < 1:15,000, this is just with a home made type prototype, im an electronics engineer not a mechanical engineer and with limited machine tools ie a drill, and is the difference of going from one direction to the other, and averaged over a few minutes. however its not consistent enough to be able to calibrate out, although its much less going from one speed to another in the same direction.

Therefore I assume I would be able to reduce this by an order of magnitude with a better arangement and reduce the variation even more with a bit more precision etc. so a predictablity of 1:1,000,000 would be much better.

Colin =^.^=

well the encoder just fits onto the end of the shaft, its just an optical disc, but it needs to be 1M away from the spinning object, so the shaft has to be 1M long. as the title suggests it would be simple if I cld just use some frictionless beatings.

Colin =^.^=

In fact, we are using an experimental defusing pump at work that has some new type of magnetic bearings in it. it operates up to 50K RPM's, it has started bearings that retract once the load cells on the 4 points detect even pressure which indicates the rotor is spinning in center with the magnetic bearings.. It actually is designed to spin up faster than that, we were ask not to let it go any faster for now. It has a special controller for it to run this operation. and i can tell you that it works very nicely and has been working nicely for some time now. I think the company that made it either product a different model for the original market or it may had been a closed market that it was originally designed for. We use it to maintain a vacuum down in the 10e-8 to 10e-9 torr.

Dare I ask if they are very expensive ?

Colin =^.^=

could you have an encoder, say, 20cm before the bearing and an encoder after the bearing measure angle with both and calibrate an error to subtract depending on the difference in the two measuments?

-Lasse

Make concentric shafts. Outer a pipe and a rod running in bearings inside. Drive both at one end. The encoder is on the rod at the other end.

Assuming you have the power you don't really care about friction in the outer bearings. The inner bearings run a zero speed so you don't have to worry much about wear and lubrication.

The mechanical equivalent of a guard ring.

You may have problems with 'siction' on the inner bearings but I imagine very much less than you have with the current set up.

get

one

stiffer

actualy

degree

not

going

more

(Thought a min' was 1/60th degree?.) What about inreasing shaft diameter?. Stiffness goes up as the fourth power of diameter, 10mm shafting would give that order of magnitude reduction in angle of twist. Or maybe go with 10mm but use tube. You'll lose some of that fourth power but still result in a big gain in torsional stiffness.

hmm so did I maybe I was right after all, someone seemed to think 1 minute was 1/3600 of a circle.

hmm 4th power thats interesting, my hole is 16 mm wich limits the size bearing I can get in there. the bearings with small id to od ratio tend to be a bit flimsy.

and then theres the problem of how to interface to an encoder that has a 2mm hole. I ground this one down to 2mm in situ. (with scrap bearings ofc) how convenient those dremmel cuttoff discs have a 2mm hole in them.

weight isnt so much of an issue, as long as it doesnt make for to much wobbling about, I had it described to me as "whirl mode" where it resonates at a certain speed and vibrates like heck. not sure i understand the exact mode of corelation between resonance and speed.

If i could find the right place to get some tubes and long enough shafts just the right size I would go for the concentric shafts idea.

another problem I am contending with is when the thing rotates in the other direction the signals are on the opposite edge, thus any diference between rise time and fall time cuases an error, although the digital signals are ~5ns the optical system may have different rise and fal times. however if it stays consitent I can null it out.

Colin =^.^=

Yes thanks, this was one of the options I considerd in my first post. just a problem finding parts to construct it out of.

yes guard ring thats the comparison I was looking for.

a bit of inevitable vibration should get over that.

Colin =^.^=

Brass tubing of many sizes can be had at the hobby store; sizes allow concentric configuration in a number of steps,depending onsize needed.

I wonder if it is viable to set up a second (identical) shaft, for run-time calibration. Second optical encoder on the end, as now, driven by a dc motor at the far end, with a third optical encoder directly on the motor.

Servo the motor so that the calibration shaft runs at the same speed as the working shaft, as measured by the difference between the optical encoders on their ends.

Then use the difference between the end and motor encoders on the calibration shaft as a measure of the error on the working shaft.

Verify the scheme by building two calibration assemblies and compare the outputs from their motor encoders.

You might be missing a magic thing called a shaft coupler. This will take care of forces on the bearings caused by a) shaft misalignments, and b) variations in shaft length (as the shaft twists).

yes, there used to be a realy good model shop nearby, with all sorts of things like that now the only one just sells complete RC models and stuff.

I can find 300mm lengths quite easily, in fact I have some at hand, but 1M lengths is difficult. although I found a place to sell 5M lengths but with a significant min order quantity lol.

Ive been googling like mad but so many false leads or links to lists of links wich dont realy help. If anyone knows somewhere online in the UK that would be nice.

Colin =^.^=