Linear Scales and Backlash

[MadTooler]

I wasn't sure which forum board to post this. Miscellaneous?

I have this older little Sherline Mill I am nearly finished retrofitting. It needs encoders of some flavor to close the KStep stepper loop. As a bit of a science experiment before I invest in the process on my larger machines, I am planning on installing Acu-Rite linear scales. Since this is such a small and slow machine, I am planning to use the scales to work around rotary encoders, backlash, and limit switches.

The scales have "Position-Trac" which will allow for easy pseudo absolute positioning after a short distance home cycle from almost anywhere along the scale. That will cover the encoding and the function of homing with limit switches (obviously not the hardware safety of limits, but I am watching soft limits).

The machine has 64,000 steps per inch on all axes. With a low tech brass anti backlash nut on acme screws, I can only get backlash on X & Y down to 0.001" while Z is at 0.003 (Z does not have an anti backlash nut). I believe the manual says to readjust when there is 0.008" backlash! With the scales, I plan to check the backlash at home cycle and fault if it is too far out.

With that much backlash, I need to work out some details. I need to determine if the large backlash coupled with linear scales will be stable or always suffer from over oscillations. If it can be stable, I need to decide between the 5 and 1 micron scales. Accuracy wise, 5 micron (0.0002") is better than I can expect with the machine and par with anticipated part work. I would only reach for the 1 micron scales if they would be more useful in keeping the control loop stable.

Any thoughts?

Thanks again.

[TomKerekes]

Hi MadTooler,

The scales have "Position-Trac" which will allow for easy pseudo absolute positioning after a short distance home cycle from almost anywhere along the scale. That will cover the encoding and the function of homing with limit switches

It isn't clear to me how the absolute position will be sent to KFLOP.

I need to decide between the 5 and 1 micron scales. Accuracy wise, 5 micron (0.0002") is better than I can expect with the machine and par with anticipated part work. I would only reach for the 1 micron scales if they would be more useful in keeping the control loop stable.

Higher resolution is better until you exceed KFLOP's count rate of 1 million counts/sec. But even at 1 micron that would equate to 1m/s. Besides position accuracy high resolution is most useful in controlling velocity. To control velocity well near instantaneous velocity measurement is required via delta position/delta time. Where the delta's are small and will have large quantization errors if the resolution is not high enough.

HTH

[Moray]

Something worth being aware of, is often with these scales the only difference between the 1 and 5 micron scales, are the reading heads which interpolate the additional 5 steps for the 1 micron heads. The 1 micron scales/heads are not actually any more accurate than the 5 micron.


I've just done a quick search on the Position-Trac system, and it appears the compatible scales output an additional channel R (differential output the same as A & B channels).
I suspect you may need to do a bit detective work to see how it functions. To re-establish position with their DROs, you have to move the axis negatively until it detects a grating, then move it positively up to 2 inches until it establishes accurate position. They say it's encrypted gratings, which I'm going to guess is they have gratings at varying intervals, so you move one direction to find a grating, then by moving in the opposite direction it either counts A/B until the next grating, at which point it knows where it is on the length of the scale, or there are batches of additional gratings which translate to a position.

A quick rig up to a logic analyser should reveal how they work. If you do want some suggestions for a suitable setup to test, just ask.

[MadTooler]

Thanks, Guys.

Moray, what scales did you use on your CMM?

Tom, Moray is correct in that there is an additional channel "R." The functionality is similar to what he noted. The distance between the R's vary so you can determine where on the scale you are. That is what I was referring to as pseudo absolute where with little movement it could be known vs. the super fancy and usually proprietary true absolutes or the scales that have battery power and memory to store the last known values after power down.

I have some great resources available to me to help with the specifics of the interfacing. I also am confident this is achievable with the dynomotion hardware. My concern for now is with the resolution and control loop stability with the potentially large backlash.

The scales are also available in 0.5 micron, but the cost is crazy considering the machine I am applying them. Even the cost of the 1 micron is more than the machine is worth, but if they are needed to keep things smooth, then I may do it.

I did understand your notes about resolution and velocity. To put a nail in it, with the KStep and steppers (with outputting steps it would have rpm and a good idea of potential resulting velocity), 64,000 steps per inch, and several thousandths or more backlash most of the time, could this system realistically achieve a smooth control loop with the coarser 5 micron scales? If not, would success with 1 micron be very likely, or is it better to go back to rotary encoders and simple backlash comp?

FYI, my current steppers limit me to 25ipm (just barely shy of 1m/s ). I will upgrade those in the future, but I would not expect to run this little machine at too much higher speed even then.

Thanks again.

[TomKerekes]

Tom, Moray is correct in that there is an additional channel "R." The functionality is similar to what he noted. The distance between the R's vary so you can determine where on the scale you are. That is what I was referring to as pseudo absolute where with little movement it could be known vs. the super fancy and usually proprietary true absolutes or the scales that have battery power and memory to store the last known values after power down.

I have some great resources available to me to help with the specifics of the interfacing. I also am confident this is achievable with the dynomotion hardware.

Clever. That should be doable with just interfacing the index signal and searching for two of them.

My concern for now is with the resolution and control loop stability with the potentially large backlash.

The scales are also available in 0.5 micron, but the cost is crazy considering the machine I am applying them. Even the cost of the 1 micron is more than the machine is worth, but if they are needed to keep things smooth, then I may do it.

I did understand your notes about resolution and velocity. To put a nail in it, with the KStep and steppers (with outputting steps it would have rpm and a good idea of potential resulting velocity), 64,000 steps per inch, and several thousandths or more backlash most of the time, could this system realistically achieve a smooth control loop with the coarser 5 micron scales? If not, would success with 1 micron be very likely, or is it better to go back to rotary encoders and simple backlash comp?

It will probably be complicated to predict the closed loop performance because it would depend on many non-linear things like stiction, compliance, backlash, binding, disturbances, resonances, etc. Closing the loop with only the glass scales would likely be very difficult with backlash and such because of times where the glass scale don't provide any information on what the motor is doing. But your system will be more of a "dual loop" configuration with position control of the motors independent of the glass scales. So for example, your system is likely to be always made stable by moving the motors in a very slow controlled manner while taking out the backlash. The only way it would be unstable is if the stiction and compliance is so severe that no matter how slowly the motors are moved when the table finally breaks free from stiction it jumps a large distance.

So I think the issue will be more what kind of performance you obtain rather than whether the system will be stable or not. How fast errors can be corrected, and so limited in their extent.

I don't think you have really told us what your requirements are or your expectations.

5um = 0.2mils (thousands of an inch). So if you are expecting to make a controlled correction of 1 mil that is only 5 counts. A rough indication of how far to go, and very rough on rate. As a rule I like to see resolution at least 10X of my expected precision. So this would be borderline but possibly workable.

HTH

[Moray]

Moray, what scales did you use on your CMM?

I used the 1 micron own brand scales from Machine-Dro.co.uk.
They're the same scales as you can commonly get direct from China, but machine dro had them in stock, and I wanted them quickly at the time.

Only reason I went for the 1 micron, was they had both lengths I needed in stock, whereas they only had one of the 5 micron scales in stock, so I opted for keeping both axis the same resolution. Both resolutions use an identical grating (20 microns IIRC), and the heads decide on the resolution. I've since learnt you can buy the heads separately, so swapping between resolutions is possible.

[smrutibora]

Something worth monitoring, is regularly with these scales the main contrast between the 1 and 5 micron scales, are the perusing heads which introduce the extra 5 stages for the 1 micron heads. The 1 micron scales/heads are not in reality any more exact than the 5 micron.

I've recently completed a snappy inquiry on the Position-Trac framework, and it shows up the perfect scales yield an extra channel R (differential yield equivalent to An and B channels).

I speculate you may need to complete somewhat investigator work to perceive how it capacities. To restore position with their DROs, you need to move the pivot adversely until it identifies a grinding, at that point move it decidedly up to 2 crawls until it sets up precise position. They state it's scrambled gratings, which I'm going to figure is they have gratings at different interims, so you move one heading to locate a grinding, at that point by moving the other way it either checks A/B until the following grinding, so, all things considered it knows where it is on the length of the scale, or there are clumps of extra gratings which mean a position.

A snappy apparatus up to a rationale analyser ought to uncover how they work. In the event that you do need a few recommendations for an appropriate setup to test, simply inquire.