Solid theory behind robot mastering

It depends. Now, I'm no expert on this, but as I understand it, one reason for the KUKA type of Mastering, as opposed to some of the competition, comes down to the different way the encoders/resolvers work. (warning: bloviation based on only partial understanding ahead!)

KUKA uses analog electromagnetic resolvers on each servo to measure the motion of the drive shaft. The only "smarts" keeping track of how many times the shaft has rotated is in the RDC card.

ABB, OTOH, uses absolute encoders. As I understand it, encoders report the actual position of the motor within one rotation at all times (that is, between 0 and 360deg). So the "smarts" in the SMB board have a little less work to do -- they only need to keep track of how many times the motor has done a complete rotation. Anything more precise, they can just read the raw encoder data directly from the motor. This means that, when "mastering" an ABB, one only needs to get to within +/- one motor rotation of "perfect zero", and the SMB can just read the encoder for any fine tuning.

What this means when comparing the two methods is that KUKAs require greater precision in their Mastering process, because they don't have the absolute encoder data to handle their "fine tuning". An ABB is more forgiving, because if the next person to master an axis is a bit off from the previous person, the mastering can be corrected for that "eyeball error" using the recorded fine-mastering encoder counts.

OTOH, this means that for an ABB, you have to have an offline record of the fine-mastering encoder counts, or else you can't easily re-master if, say, the SMB board gets fried or loses its memory. And you have to update that record if a servo ever gets replaced. Whereas KUKAs literally don't care -- swap the RDC, replace a servo, it's the same Mastering process (and tools) as any old Mastering.

So, your Mastering process (and record-keeping) is a choice between tradeoffs, depending in part on what kind of servo motor feedback you're using, and the people you expect to be doing re-mastering in the field -- basically, how idiot-proof is your mastering process? Your robot will be maintained by idiots at some point, if you sell more than a few of them -- it's essentially inevitable. One of the nice things about the KUKA approach is that it removes almost all ambiguity (unless someone really damages the EMD or the Mastering Gauge/Target). The downside is that it requires an EMD, dial gauge, or some real skill to do it by feel. The ABB version tends to be easier to deal with for day-to-day usage, but (IME) can be a bigger headache (and more vulnerable to making things worse, or losing the critical piece of paper) when something more serious goes wrong.

There's also safety issues -- Mastering is part of the safety-critical data chain (100x so if you're creating something like KUKA SafeOperation or ABB SafeMove), so your realtime position data and your Mastering have to be not just bulletproof and idiotproof, but you must also have some Safety-grade method of detecting if there's something wrong with it, and a Safety-grade response.

Quote from robotninja

After aligning the witness marks manually, the system is powered on, and the encoder readings are saved. Does this work?

I mean, it could. Depends on the fine details of your encoders and how you're tracking them. The most obvious pitfall would be if a motor got moved without power and somehow landed within the tolerance band of the encoder, but some N motor rotations off. That is, the encoder can't tell you if you're at 271.123deg on Rotation 1, or Rotation 300. And your encoder will have a tolerance band, because no servo holding is perfect, power on or off.

There's also backlash to consider -- the bane of robot accuracy, and the biggest reason most robot applications run on repeatability instead. Pushing joints around with the power off poses a very real possibility of "mastering" an axis on the opposite side of the backlash band from where that robot will operate, which can have a cumulative effect.

And that's just off the top of my head. Bottom line: there are reasons why the big robot brands' mastering procedures are as complicated as they are, and you need to address those reasons if you're making your own robot. Some of those reasons are legacy stuff left over from when computers were far less powerful and sensors were cruder, but a lot of them are in place from hard, painful lessons accumulated over decades.

If your team wants to make this work, they need to spend some serious time and thought asking "what's the worst that could happen?", and chasing down those rabbit holes. I've encountered robots and other machines were this wasn't done, and it was painfully obvious (my pain, usually, followed by sharply worded reports recommending blacklisting that supplier from our approved list).

it all boils down to how much downtime is tolerable...

with KUKA, there is a reason behind this... even in case of a disaster one can master all axes in 10minutes and all points in all programs will be fine.

with other robots you may not be so lucky. you may have to touchup points etc to get back to production. depending on number of programs and points that can take much longer, in some cases a week or two.

There is a sticker at the back of link 2 on ABB robots (black letters on silver), see the attached example photo. The sticker shows a number from 0 to 2*pi radians (0 to 360 degrees), which is the resolver offset, for each axis. Even for old ABB robots without a control system, the original calibration offsets should be available.

what happens when robot was maintained/repaired? for example when motor or wrist are swapped out

Quote from panic mode

what happens when robot was maintained/repaired? for example when motor or wrist are swapped out

True, then those axes must be recalibrated and the sticker values are no longer valid.

Quote from geir_hokarob

There is a sticker at the back of link 2 on ABB robots (black letters on silver),

On the last several ABBs I got "new from factory", those stickers were in a baggie in the controller, and had to be stuck somewhere by myself.

Then there was a set of brand-new Fanucs that had gone to China first for "integration", before being shipped to the US. Everything inside every controller that wasn't wired in had been removed -- all papers (including the master encoder counts), spare fuses, mode-select keys....

Quote from robotninja

Do all robot companies have similar mastering devices? I saw quite a few patents of mastering devices from a google search. What about smaller robots like those from Universal Robots? How do they do their mastering?

Aside from KUKA, most of the commercial brands I'm familiar with use paint marks or some other kind of vernier marking, and rely on the absolute encoder and master encoder count to correct for "eyeball error".

Although Nachi robots (the last time I used them, many years ago) used a kit were you had to bolt a metal block onto each half of a joint, then jog the axis until you could get a slip-fit pin to pass through a machined hold in both blocks. And god help you if you managed to bend the pin or any of the blocks, especially since each joint required a different set of blocks....

Quote from robotninja

Does it make any difference if we were using resolvers instead of encoders?

It absolutely makes a difference to how your Mastering and data retention work. Does it make a difference to your Mastering process? That's a good question.

I'm not an expert in this area by any means, but in my experience, on any brand, a robot joint that's moved more than a tiny amount while the power is off is treated as a fault that, at minimum, requires the joint Mastering to be checked and verified. This is an attempto to deal with situations where a robot might have been hit by a forklift while powered down, someone might have replaced or altered a motor, or the encoder/resolver's mechanical connection to the motor drive shaft has become unreliable (I saw this happen catastrophically once -- talk about a GIANT KARATE CHOP).

So this goes back, again, to the "what's the worst that can happen?" question, and the reason it's so important to ask it, because the encoders/resolvers are part of a Safety-critical hardware chain.

Frankly, without an in-depth knowledge of what you're trying to do, it's hard to say. But the "just push it into Mastering position" feels like a cheap, lazy cop-out to me, by people who haven't really thought about what they're doing and how it can go badly wrong.

Quote from robotninja

Does it make any difference if we were using resolvers instead of encoders?

you need to think how they function.

for optical quadrature encoders limits include number of pulses per revolution that can be etched into the glass disc, how shatter proof the disc is, the speed of optical sensor, how the position is retained (in counter requiring battery power) etc.

optical encoders can be absolute too, then loss of battery power may not be much of a deal since encoder can still keep track of positions.

resolvers work as rotary transformer. primary side of transformer is excited by sine wave. there are two secondaries that are 90deg apart producing sin/cos voltage depending on position. the advantage here is no tiny windows in an optical disc, so this is shatter proof and works very well in extreme environments. there is no need for batteries as resolver can still tell if position changed. the resolution does not depend on something that is in a hostile place (motor that vibrates, gets hot) - this is determined by AD converter that can be placed in a safer place and is easier to access or replace. same servo motor can be used on KRC2 and KRC4 for example. but connecting it to KRC4 gets you higher resolution since that RDC has better ADCs.

one of advantages for such design is that robots can work in really REALLY harsh places - even in nuclear plant where nuclear radiation makes electronics go crazy. all you need to do is shield small non-moving box with RDC. you cannot do that with motors that use optical encoders.

but the best part is servicing... on KUKA robot one can simply power down, unplug any cable to any motor and do electrical checks (motor and resolver). when done plug it all back before powering up and voila... no calibration is needed. you can do it 100 times, any time... in comparison, those that work with robots that use encoders are walking on eggshells - "do not unplug that cable or battery!!!"

and even when calibration is lost, it is easy and quick to recover from. one does not need to spend a day with a laser tracker or CMM, or spend days teaching points.

How big will it be such that it needs to be disassembled for shipment rather than leaving it assembled?

Quote from robotninja

And the joints are also potentially going to have to be passed through a tight opening, and deployed in an area where the robot is not directly visible.

I... won't speculate. But that does put a very different spin on things.

You may want to examine the option of having actual absolute encoders on the joints, instead of on the motors. Something like a DRO for a mill or lathe, just wrapped around the joint. Not necessarily to replace the motor encoders/resolvers (getting high enough resolution could be an issue), but to provide a nearly-infallible cross-check.

The coarse version would be to add some kind of "homing switch" to each axis. So on reassembly, the first action would be to "home" out each axis, rather like most 3D printers do at the start of a print job. The potential downside there would be if the spatial constraints of the new location prevent a joint from reaching the Home position, but having a few uniquely identifiable Homing sensors around the joint arc could fix this. Have a semi-automated Homing routine that "sweeps" across the Homing sensor both ways, at high then low speed, to try and null out the backlash issues.

I suppose the "gold plated" option would be to have a laser tracker target nest mounted on the wrist, and on reassembly use a laser tracker to re-tune the DH parameters and Mastering for the "as-reassembled" condition.

Serious question: how precise and how accurate does this robot need to be when reassembled? Will anyone be within reach of the robot while it operates?