ALTERNATIVE SOURCE FOR MAGNETIC SHAFT ENCODERS

This does not relate directly to Darwin but more to Tommelise and perhaps Mendel if Mendel goes for shaft encoded DC motors instead of steppers.

Heretofore we've had only a single source for magnetic shaft encoder chips, viz, Austria Microsystems. That has been worrisome. Now there is a second source of very similar chips out of Slovenia at roughly the same price (~US10).

[www.rls.si]



They're also surface mounts.

Interesting. The specs and capabilities of those chips appear to exceed those of the Austria Microsystems hardware. They also have a US distributor. It makes me feel like they're a more established player.

I think that the distributor is the established player rather than the manufacturer.

Hmm, just a thought. The ics originally tested on
[reprap.org]

Is the quadrature pulsed output the only option? Wouldn't it be easy to miss a pulse or two and thus accumulate a substantial error over time? It seems this has to be complememted with some kind of zero-passage detector to work out of the lab bench?

The slovenian ones (just skummed briefly over the data sheets) seem to have an absolute binary position output, it should probably have its zero where the magnets south-north border passes over a certain point, right?

Of course reading a string of bits could introduce lag insted at fast speeds but at least you always get an accurate reading at stand-still.

mimarob Wrote:
-------------------------------------------------------
> Wouldn't it be easy to miss a pulse or two and
> thus accumulate a substantial error over time?

Yeah, that's pretty easy to do.

> It
> seems this has to be complememted with some kind
> of zero-passage detector to work out of the lab
> bench?

I reset the positioning system after printing each layer.

You can avoid missing pulses by either polling faster than the inputs change with a timer interrupt or have one or both of the quadrature pins on an input that can generate interrupts on both edges. I have used both techniques in the past and had no problems with absolute accuracy, it's just standard engineering. Whenever you see one input change you look at the state of the other to determine whether it is forward or backwards.

One nice feature of quadrature is that if you get a glitch on one of the lines it cancels itself out because it has two opposite edges. For example, if the shaft stops right on the edge of a transition then one of the inputs may dither up and down. That just causes your count to go up and down by one because of the logic of quadrature. If you just count pulses from one of the signals then you are in trouble if this happens, which it will.

Edited 1 time(s). Last edit at 08/07/2007 12:59PM by nophead.

---

[www.hydraraptor.blogspot.com]

I looked again at the data sheet for the AS5035, it seems to have an "index" pin as well, it will fire at the north-south passage to enable recovery more often then when using the end-stops. I guess the endstops have to be with precision as well when using the present technique ;-)

If you have an index then the end stop only needs to have a precision of less than one turns travel.

When you see the index you round your count to the nearest whole revolution. When you see the end stop you take the remainder of the count divided by a revolution.

Once you have seen the index once and the end stop once you should always know you position absolutely to the resolution of the encoder provide your software does not miss any transitions, which if it well designed it will not. I.e. you should have no cumulative errors in the system. You should always know you absolute position up to the limits of the slack in the thread etc.

---

[www.hydraraptor.blogspot.com]

Quadrature isn't your only option on any of these chips.

Quadrature is very good for being a generic interface to incremental encoders. Its also great for those high speed positioning systems, especially with semi-dedicated hardware. However, all the various encoders I've seen listed here have an absolute output function. No worries over skipped codes, you can deal with a simple numeric rollover check, SPI/SSI means you clock it at your choice of speeds, and theres more widespread hardware support for those who want a little helping hand in their PICs. It'd be relatively simple in code and hardware. Single turn absolute may also be useful for situations (how about power loss ride-through and restart where you left off without rehoming?).

For example, string them back to the main Tommelise board and someone could easily read all data for 3 data lines + n chip selects (so 6 for the current incarnation).

Austria microsystems trades speed for accuracy in this mode, but exactly how fast are you spinning these little shafts?

I've heard of an entirely different method of doing what a shaft encoder does, but optically instead of magnetically. The technology, to my knowledge, has so far only been used to in model train engines.

The divice uses a par code printed on a wheel, and some kind of cheap LED/Photocell arangement to take readings from the bar code, which a microcontroller uses to determine the position of the wheel. Problem is, it's patented.

I was thinking that something similar (not to mention simpler) could be done using a method that old 5

To pull the topic even further off course, I had a think about the whole position encoding exercise and noted the following:

1) Any solution which only counts turns of the screw will be susceptible to backlash / tension / compression effects caused by the forces required to move the driven load. Ideally, the position sensing should be independent of the leadscrew assembly.

2) I then considered having a separate feedback belt / wire attached to a rotary encoder (be it optical or magnetic - doesn't really make a huge difference...)

3) As a further refinement, a really nice approach would be to have a linear position encoder, which simply watches this belt go past.

4) To get high resoultion, one can use a strip of acetate (as used for overhead projectors), and print a series of stripes on it with a nice 600DPI printer. (Yes, yes, we've now gone to optical sensors, which is off topic...)

5) Obviously we then need a nice set of finely spaced photodiodes to pick up this pattern. These things are called 'linear photodiode arrays'.

6) Some searching around found the Texas Advanced Optical Systems' TSL3301, which is a 102 pixel diode array with inbuilt ADC and 300 DPI pixel spacing. The datasheet talks about a 'maximum pixel clock' of 1MHz, which suggests that the whole array could be read at just below 10kSamples / sec - nice and fast.

7) If we print a Gray code pattern onto the acetate ribbon, we have plenty of pixels and can read off ABSOLUTE position.

8) An A4 sheet of acetate is easy to obtain, and I'm sure one can find A3 if they look hard enough. This gives a position indicator strip up to 400mm long, with a 300DPI resolution (i.e. 85 um). The sensor chip costs about $15 USD.

So, does this catch anyone's attention?

Sure! Build one and make it work! That will really get some attention.

BTW, the 256 bit magnetic rotary enconder chips are down to about $3-4, now.

Hi Jono,

... i played around with web-cams and encoding positions with open-source image-recognition.

The normally used optics allows absolute resolutions down to 5 microns, if you revert the objective.

For another approach i tried 3D-measuring with a single camera-chip without lenses, a LED and two metallic sieve-sheets - with interpreting the moir

Eh, anyone know a bit about inductive sensors?

[www.newall.com]

I've heard excellent results from these guys and the basics are fairly simple. Ball bearings + plastic tube + electronics + plastic reader head (notice how this tends to follow Darwin here...) and we might be able to get a good position reading out of this.

Not sure if this helps or not, some HP printers have a fringe strip and pickup, also would optical mice, or the camera in them be any good, I think I saw something on the arduino website, you would not need a fringe even with this, just a flat strip, that is if the resolution is good enough.

---

Random Precision