Hall rotary encoder tempco--wisdom?

Can you add mechanical gain, instead? Is the motion always "one way" (or, can it be made to be so?)

It's not a backlash issue--it really is the encoder tempco. (Happens both forward and reverse.)

The next rev will have gears and a preload spring, or possibly a sine bar drive with a shaft encoder on the lead screw, but we need to get this one working to make a clinical trial date.

Thanks

Phil Hobbs

Maybe that shouldn't be surprising; Hall effect sensing starts with (usually) a few dozen microvolts, and thermal and other drifts aren't swamped by that signal.

I'm not pleased with most shaft encoder designs, in fact. My preference is for stepper/loaded geartrain with an optointerrupter so I can get a step 0 indication. That has faults too (but one COULD imagine optical lever variants); I just recalibrated every measurement run against a standard.

Have you considered ovenizing the shaft encoder, yet? It sounds icky, but it might be a good quick fix.

Why not an optical encoder, with lots of lines?

Yup, that's underway, as I mentioned, thanks.

As soon as I found out that the encoder used Hall sensors, which are notoriously drifty, I fished out the cold spray and had at it.

Cheers

Phil Hobbs

No. My point was that (additional) mechanical gain can be a win -- but only if it isn't offset by the loss of precision that backlash in the geartrain can introduce.

Good luck!

Too big and expensive. This one is only about 12 mm long by 12 mm diameter, and costs about $15.

It's in what looks like an LC plastic housing, which may well be the origin of much of the problem. Do you folks ever use these things?

Cheers

Phil Hobbs

Could be that a Hall encoder doesn't have a lot of actual lines, and they are interpolating. The Hall TC could mess that up.

the datasheet says mechanical Angular Accuracy

Internally it's a Hall version of a synchro or maybe a resolver. The differential-transformer ones are wonderful, but much bigger and more expensive.

Cheers

Phil Hobbs

It actually drifts 0.7 degrees with a 15 degree temperature increase, so I sort of doubt that last spec.

The PWM is sampled with a high speed capture input.

Cheers

Phil Hobbs

Yeah, it looks like maybe two analog elements, not a bunch of lines like an optical encoder. I can see how the case, the magnets, the Hall sensors would all have bad TCs.

It's, basically, interpolating 2 bits up to 12.

Are you ratiometric on frequency? The spec says that the frequency can drift a bunch with temperature.

Ah, good catch, thanks. I asked the SW guy, who said he's just measuring the pulse width! (All in LabView, of course.)

One of the many reasons that I hate LabView is that it's ridiculously hard to debug compared with procedural code--it's as bad as debugging a spreadsheet.

We'll probably ovenize it anyway--it's cheap insurance.

Cheers

Phil

Den onsdag den 8. januar 2014 23.24.03 UTC+1 skrev John Larkin:

it like a resolver, trig on two voltages

I wonder if it works better at certain angles such as multiple of 45deg

-Lasse

That could explain a lot of the error.

As Lasse pointed out, there may be angles that are better than others. Since you're not moving much, maybe you can rotate things to a sweet spot.

I'd guess that multiples of 90 degrees might be good. I have a 51% chance of being right about that.

The quoted spec on duty cycle error vs temperature is about 10 times better than we're measuring, assuming that the drift is a straight line, and that "0.9 degrees" means +-0.45 degrees. If it's quadratic, the slope will be four times worse than that, of course, so the improvement may not be enough. We'll certainly fix the S/W, but temperature control the encoder as well. If we get lucky, we can leave the temperature control off the other five pre-production units.

Belt-and-suspenders forever!

Cheers

Phil Hobbs

It's not supposed to budge out of spec'd performance limits over -40oC to +125oC operating temperature range.

ibly some ethological insight:

ometer. The SNR is pretty good, way over 60 dB (which is good for a SWIR s pectrometer).

f sort of swimming around a bit, i.e. the gross shape stays roughly the sa me, and the small scale noise is low, but there are small systematic variat ions on scales of 1/10 to 1/2 of the scan range, where multiple spectra don 't quite line up with each other.

nstable with temperature, something like 3 arc min per kelvin. Since the g rating only has to rotate about 8 degrees for the whole measurement, that w inds up being, like, 3 nm/K, which is _horrible_.

le on the medium scale motion, although it didn't do small steps very repea tably. It used a pot for the encoder, of course.

oder, mostly by thermally grounding the leads to a controlled plate,

ody and the shaft.

bodge in a pot if we have to, but it'll limit the instrument lifetime fair ly severely.

ter, and costs about $15.

HP used to do reasonably compact and tolerably cheap optical encoders and o ptical encoder kits, aimed at front panel controls.

Farnell has three like this from Avago