Can a PNP machine be accurate with belt drive and stepper motors?

[TankSparks]

I don't understand how it can be if the head moves about 60mm per revolution, and the stepper has 200 steps per revolution
 
Neoden 4 literature.... X-Y repeatability +/-0.02MM   really?   Looks like they are using 16 microsteps to get that number.  I'm not convinced that mcirosteps gives increased accuracy.  Am I wrong? 

[Doctorandus_P]

If you put a DC current through a stepper motor, it will just stay put in one position, but if you then apply an external torque to that motor, it will rotate a little bit, and when the rotor is rotated for a single full step, then the motor delivers it's maximum torque. And this also works the other way around. When a motor is loaded to way less then it's rated torque, then you can get more resolution by applying microstepping. So you can use the full motor torque for fast acceleration and when the guide rails are smooth you can gain two or 3 bit of extra resolution during the final placement when forces are low.

But there are other things to consider too. I do not know what Neoden does. but the bigger laser cutters often have some extra pulleys and a small timingbelt to get a 2:1 or 3:1 speed reduction, which also enhances resolution. 3 phase stepper motors often have 300steps/revolution, and two phase 400steps/revolution motors also exist. With closed loop motors, you can also get more accuracy, but stretching of the belts is also a real issue. Getting a real 20um repeatability with timingbelts that are a meter long (there and back again) is doubtful. Heck, you probably already have such deviations if the aluminium frame warms up a bit from a motor getting hot.

[thm_w]

Steppers are available in 400 step/rev as well as 200 step. As well as gear ratios mentioned above. You'd have to look at how they've actually implemented the motion control. Of course yes, most manufacturers stretch the spec and just tell you the lowest micro step resolution.

In this case they also mention "Once the coordinates are established, the semi-closed-loop stepper motors are able to repeat these locations to 20μm accuracy without further need for this camera". No idea what that means, if there is angular sensing on the stepper or not. But that can also help, especially if you were to drive the stepper with FOC (eg TMC4671).

[NorthGuy]

Microstepping works. Half-steps are very accurate. When you drive a stepper with 32 microsteps, you can move it one microstep, and you can see it move with dial indicator. It may not be very linear between two half-steps. Also, the driver generates current with switching, the current may not be very accurate and will have some ripple as well.

I would worry more about the belt. I doubt it gives 0.02 mm accuracy. Not anywhere close.

0402 is 0.5 mm wide, and the smallest reasonable IC pitch is 0.4 mm. So 0.02 mm wouldn't matter much.

[SMTech]

Microstepping works. Half-steps are very accurate. When you drive a stepper with 32 microsteps, you can move it one microstep, and you can see it move with dial indicator. It may not be very linear between two half-steps. Also, the driver generates current with switching, the current may not be very accurate and will have some ripple as well.

I would worry more about the belt. I doubt it gives 0.02 mm accuracy. Not anywhere close.

0402 is 0.5 mm wide, and the smallest reasonable IC pitch is 0.4 mm. So 0.02 mm wouldn't matter much.

0.4mm pitch infers a lead width of ~0.25mm or even less, your target accuracy will be 1/2 of that or better.

You will notice all the Neoden class machines mix and mash accuracy of the robot with that of the placement. Commercial machine will clearly differentiate between those two things. Our (belt driven) Essemtec runs using servos on 30mm belts in X (moves the head) and dual 50mm belts in Y (moves the gantry). It clearly states a 40micron resolution on the linear encoders and a placement accuracy of 51microns for "chips" or 39microns for QFP at 3sigma, the sigma value being something the Chinese machines simply don't quantify or qualify in any way. The commercial machines will have thermal compensation algorithms built in.

Placement accuracy is affected by your vision model, coplanarity, stability and a whole host of other factors, your robots precision of movement is only half the story. Neoden might be able to be that accurate with a brand new machine freshly calibrated, but they'll never know when that specification falls can no longer be met by the belts if they are measuring on the motor rather than the length of travel. I rather doubt the belts deliver even on day one, that would require a level of QA user experience on here suggests its missing.
Essemtec dropped belts and doubled the encoder resolution on subsequent models. Larger manufacturers had already done that at least a generation earlier, if they used belts at all, screws and linear motors got more affordable, speeds got higher and both are less hassle to support and maintain.

[Smallsmt]

I don't understand how it can be if the head moves about 60mm per revolution, and the stepper has 200 steps per revolution
 
Neoden 4 literature.... X-Y repeatability +/-0.02MM   really?   Looks like they are using 16 microsteps to get that number.  I'm not convinced that mcirosteps gives increased accuracy.  Am I wrong?

These accuracies are impossible due to the belt's temperature response.
Systems that do this with belts are equipped with active measuring systems.
The guides of the machine will probably not do this either.
Once again, resolution has been confused with positioning accuracy.

[MR]

Ditch a magnetic scale on it with 5um resolution and close the loop. Let a PID handle the rest, story done - accuracy achieved.

[thm_w]

Ditch a magnetic scale on it with 5um resolution and close the loop. Let a PID handle the rest, story done - accuracy achieved.

Used to be the update rates were bad, 5-20Hz. So it would slow down the machine unless you are using it for initial calibration only.
Any recent scales that have better update rates and don't cost a fortune?

[MR]

20hz? No way. You mean 20m/sec. I would not trust those numbers unless I test them myself though. But even 1-2m is more than good enough for an average PNP machine since acceleration and deceleration are the most important parts not the max travel speed.

aliexpress.com/item/1005005087476488.html

Aliexpress is not the best source to buy such items.
I'm using glas scales myself on my CNC machine.

Chinese parts are okay as long as you understand what you get from them - and it's not what you would get from upper priced manufacturers.

[Kean]

@thm_w are you confusing the serial data update rate of a DRO with the quadrature output of the scale?

[Prime]

Ditch a magnetic scale on it with 5um resolution and close the loop. Let a PID handle the rest, story done - accuracy achieved.

That helps but it still doesn't fix the backlash. Long belts stretch.

So the head overshoots. The tension also fluctuates with temperature. You can put a tensioner on but it's a solution in search of a problem.

Mycronic ditched belt drives nearly 10 years ago for a reason. The service techs who used to maintain our machine used to tell us how good the linear drives are in comparison.

[MR]

A closed loop system using a linear scale will also help with backlash because the PID will try to correct until the target is reached.
Closed loop isn't that difficult.
Mid range systems should just be fine with it.
Entry level systems (belt drive) <-> high end systems (linear motors)

Of course if you strictly want to ride the open loop horse it will have its disadvantage.

[langwadt]

I took apart a scrapped mitutoyo motorized XYZ table,  I think for a microscope using in a semicondutor factory. It was ballscrew drive but there was no position feedback from the motors at all, only analog tacho, the positioning was done with optical scales

[Coordonnée_chromatique]

I don't understand how it can be if the head moves about 60mm per revolution, and the stepper has 200 steps per revolution
 
Neoden 4 literature.... X-Y repeatability +/-0.02MM   really?   Looks like they are using 16 microsteps to get that number.  I'm not convinced that mcirosteps gives increased accuracy.  Am I wrong?

hello,
witch accuracy are you talikng about ?
Absolute, relative, relative to what, kind of movements, size of movements, mechanical moments, 
0.02MM is the radial clearance of a high quality 608 ball bearing... alone

[coromonadalix]

all pnp's i saw   had stepped motors with encoding wheels   down to 0402  parts sizes ??   if it was not good  they would have used something else

and before smt placement the tech guy  did  fiducial placement cross checks vs parts placements, and did one pcb   before a batch  just to be sure ,,,

[ajb]

all pnp's i saw   had stepped motors with encoding wheels

Steppers or Servos?  Any serious machine ought to be using servos, although the motors may look the same the operation and performance is significantly different.  Well, technically a servo system can be built around any kind of motor, but AFAIK most of the motors you see in these classes of servo systems are 3-phase permanent magnet setups, like a common "brushless DC" motor, not steppers.  Beyond just closed-loop positioning, servos typically offer much more sophisticated control, including better overload (crash!) detection and much better speed and torque characteristics.  But anyway, as has been mentioned, the key part is to close the loop with at least a high resolution rotary encoder on the motor, or preferably a linear scale on the axis.  Motors with factory-installed rotary encoders with up to 24 bit resolution are readily available, so resolution is not a huge problem.  That's of course not the same thing as repeatability or accuracy, but with a placement machine you also have a vision system.  With vision, you can load an engraved glass plate into the machine and run a calibration cycle to measure and null out irregularities in the machine kinematics.  In the CNC machine world, spindle probe and calibration plate or a ball bar test can be used for similar purposes.  Glass scales on the axes are pretty much the gold standard, although you can do MUCH better with crazy interferometry-based position feedback, down to the point where heating from people in the room becomes an error source.  But rotary encoders can be pretty good, when paired with adequate mechanics: preferably a high-quality ball screw with properly pre-loaded bearings, as well as quality linear guides, all on a big heavy machine base. 

[thm_w]

@thm_w are you confusing the serial data update rate of a DRO with the quadrature output of the scale?

The only reasonable hobby prices back in the day were capacitive sensing scales which are read at <50Hz.
So you are right the quadrature rates were always fast (glass or magnetic), and now prices have come down.

[MR]

all pnp's i saw   had stepped motors with encoding wheels

Steppers or Servos?

Steppers have more torque at lower speeds when compared to BLDC servos of the same form factor.
Many low end pick and place machines use steppers.
For entry level Pick and Place machines acceleration / deceleration matters most before traveling speed comes into play making Steppers perfectly suitable.

I'm using BLDCs (X / Y) and Steppers (Z / Theta) on my PNP, steppers with appropriate form factor ratings would perform faster in the current situation on X / Y but in general I aim to replace my 200W BLDCs with 400W on each axis (X - 2Y), I'm using ballscrews however.

In old days they also used some ancient drivers for steppers which generate hell of a noise... the situation changed a bit with some dsp driven controllers.

[langwadt]

all pnp's i saw   had stepped motors with encoding wheels

Steppers or Servos?

Steppers have more torque at lower speeds when compared to BLDC servos of the same form factor.
Many low end pick and place machines use steppers.
For entry level Pick and Place machines acceleration / deceleration matters most before traveling speed comes into play making Steppers perfectly suitable.

I'm using BLDCs (X / Y) and Steppers (Z / Theta) on my PNP, steppers with appropriate form factor ratings would perform faster in the current situation on X / Y but in general I aim to replace my 200W BLDCs with 400W on each axis (X - 2Y), I'm using ballscrews however.

In old days they also used some ancient drivers for steppers which generate hell of a noise... the situation changed a bit with some dsp driven controllers.

adn stepper can just as well be run as bldc servos, they are just two-phase instead of three-phase

[TankSparks]

I think stepper motors with encoders is so you can run them faster, otherwise you run them 30% slower than they are capable of if there is no feedback or risk lost steps.   Doesn't give you more resolution. 

My CNC has 10 threads per inch screw,  vs Neoden with 20mm pulley moving 60mm per revolution.  I don't think microsteps work for positioning, but works for smoother acceleration.   half steps would be ok though for positioning.   So for the Neoden 60mm / 400 steps = .15mm per step. 

[NorthGuy]

I think stepper motors with encoders is so you can run them faster, otherwise you run them 30% slower than they are capable of if there is no feedback or risk lost steps.

This is a misconception. The feedback doesn't prevent loosing steps. It can correct later, after it detects that the steps have been lost. This is Ok for PnP, but it is too late for CNC. If you let one motor skip steps while others are still moving, the trajectory of the tool will be wrong and it'll be cutting in the wrong place. Moreover, lost steps will kill the inertia, so, you'll likely to lose more steps, and it may be impossible to get out of the situation without stopping everything.

Instead of using undersized motors with encoders, buy regular motors and size them correctly. It'll be cheaper and more reliable.

Optical feedback where you look at the actual thing you're moving is useful because it takes away errors (that is screw/belt unevenness, backlash, temperature, bend etc.), but encoders on the motors are not.

[H.O]

I have cheap closed loop steppers on my CNC-machine. They are driving (rolled) ballscrews with 4mm pitch. Motors are standard 2-phase motors with 200 steps/rev meaning a "native" resolution of 0.02mm/step.
The control resolution is set to 0.002mm and with an indicator on the table I can defintiely see it move on each "click". There's not always exactly 5 "clicks" per 0.01mm, sometimes it's 4 only to be 6 on the next. How much of that comes from the motor/drive and how much of it comes from the mechanics I don't know.

That does in no way mean that I believe its absolute accuracy within the envelope of the machine is 0.002mm but its resolution is definetly better than 1/2 step.

On a pick and place machine, where the load is fairly constant (compared to machining) I would expect the achievable accuracy to be better compared to a machining application (all else being equal).

On a belt driven machine using steppers makes sense since they have high torque at low speed. A servo would either need to be oversized or geared down in order to produce the same kind of torque as the cheap stepper. On a ballscrew driven machine with say 10 or 20mm travel per rev (instead of 60 as was mentioned) a direct drive servo starts to make more sense (to me).

Finally, the first post mentions 0.02mm repeatability which, again, is not the same thing as resolution OR accuracy.

[thm_w]

This is a misconception. The feedback doesn't prevent loosing steps. It can correct later, after it detects that the steps have been lost. This is Ok for PnP, but it is too late for CNC. If you let one motor skip steps while others are still moving, the trajectory of the tool will be wrong and it'll be cutting in the wrong place. Moreover, lost steps will kill the inertia, so, you'll likely to lose more steps, and it may be impossible to get out of the situation without stopping everything.

Instead of using undersized motors with encoders, buy regular motors and size them correctly. It'll be cheaper and more reliable.

Optical feedback where you look at the actual thing you're moving is useful because it takes away errors (that is screw/belt unevenness, backlash, temperature, bend etc.), but encoders on the motors are not.

Depends on the complexity of the feedback mechanism. The printer I have has a error output from each stepper axis, so you can use that to trigger a pause on the machine until the motor is able to catch up to its correct position, then error output will be de-asserted. Although I doubt marlin would support something like that, so its left unused.
You can also specify the desired error window in degrees.

I don't disagree with what you are saying about using the right size of motor though.