Closed-loop stepper controller progress

I've now got a working prototype of my closed-loop stepper controller board (here's a link to a draft data sheet: lobocnc.com/ps3.pdf). It mounts on the back of a motor along with an encoder that provides position feedback. The main advantage of closed-loop control is that you can get much better performance out of the same motor (higher speed - up to 1200 RPM, more usable torque) over using a standard open-loop stepper driver. Also, if you connect it up to a serial port, you can read back the actual motor position as well as the servo following error, which is roughly proportional to the load on the motor. I think this would be most useful on extruders (where you want to keep the motor small) or on printers with lead screw drives that need the extra speed. It'd also be great for CNC machines where the position feedback can be used as a DRO.

I'm trying to decide where to go next. Is there enough interest to make a batch to sell? Would this be a good Kickstarter project, or is it too obscure? In small production volumes, the board would end up having to sell for about $45 -- maybe down to $30 in large volumes. Then there is the cost of the encoder ($23) but minus the cost of a Polulu driver you no longer need ($10). Is that too much for the increased level of performance? In small volumes, assembly will be the biggest cost, but I worry that soldering the surface mount driver chips would be too difficult if it were sold as a kit. Any opinions would be a big help.

Let me get this straight: This is a bolt-on closed loop control upgrade for any Pololu driven machine with steppers that have a shaft out the back? Higher speed and acceleration with position errors continuously corrected? Can the encoder be mounted elsewhere, for steppers that don't have a shaft out the back? I think most people have single-shaft steppers.

Any chance we could print our own housing for the encoder to bring the cost down? Or is the cost not significant?

I do like the idea of closed-loop control upgrade. Especially if it works with my present control electronics. I'd buy into a Kickstarter, for sure.

Quote
Dale Dunn
Let me get this straight: This is a bolt-on closed loop control upgrade for any Pololu driven machine with steppers that have a shaft out the back? Higher speed and acceleration with position errors continuously corrected? Can the encoder be mounted elsewhere, for steppers that don't have a shaft out the back? I think most people have single-shaft steppers.

Any chance we could print our own housing for the encoder to bring the cost down? Or is the cost not significant?

I do like the idea of closed-loop control upgrade. Especially if it works with my present control electronics. I'd buy into a Kickstarter, for sure.

Yes, you've got the gist of it. As for mounting the encoder elsewhere - it needs to be pretty intimately connected to the motor shaft - any play at all between the encoder and the motor shaft will cause problems. Depending on how your motor mounts, though, you might be able to mount the encoder on the front shaft (the encoder has a thru-hole) and then mount the board just off to the side. You could then attach the very end of the motor shaft to your coupling. For a new machine, though, it'd be much easier to start with dual-shaft motors. (A little harder to find, but no more expensive than single-shaft.)

I don't expect the aluminum housing to be that expensive - it'll be laser cut and then you just fold it up by hand on the perforated seams. You could print your own, but I'd worry about heat. The driver chip does have a good amount of heatsink area on the bottom of the board, but the heat still has to get out of the box.

Edited 1 time(s). Last edit at 11/06/2013 02:17PM by LoboCNC.

just a heads up. this does not increase acceleration, it allows incremental corrections for reliability. yes you could say it may allow it to run motors faster, but if any error occures during a print the extruded plastic in the area of the error will be a visible and unwanted flaw. you will likely need a pid algorithum to predict and compensate for errors when they are not large enough to be visible by the human eye.

I think this is a next gen technology and probably should be moved to one of those other forums.

Trying to find a closed loop stepper system , one that is with in my budget no less, has been a daunting task. I found a driver for $80-$140 on alibaba from china but other than that they have all been $4xx-$1xxx.
My point being, I would buy them. Kit or otherwise.
edit, Im not worried about the motors running faster, I want them to run colder

Edited 1 time(s). Last edit at 11/06/2013 10:39PM by cakeslob.

Quote
jamesdanielv
just a heads up. this does not increase acceleration, it allows incremental corrections for reliability. yes you could say it may allow it to run motors faster, but if any error occures during a print the extruded plastic in the area of the error will be a visible and unwanted flaw. you will likely need a pid algorithum to predict and compensate for errors when they are not large enough to be visible by the human eye.

I think this is a next gen technology and probably should be moved to one of those other forums.

In some sense, this is true. If you flat out exceed the torque capability of the motor, a servo system will correct for that (rather than lose steps) but it will leave imperfections in your print. However, with an open-loop stepper, you set the acceleration limit so that you never exceed the torque limit for the absolute worst case loading because it will ruin your print altogether. However, worst case loading may be something you only see rarely (maybe every 10th print?) and probably not on the critical outer perimeters. So in practice, you may be able to set higher accelerations without a noticeable decrease in quality.

There are also a couple of mitigating factors. Firstly, if you have a 1.5A motor, running it as a servo, you can safely set the current limit to, say, 2.0A because the average current used by a servo is much lower. This again lets you boost the acceleration. And run cooler.

Secondly, microstepped steppers aren't as accurate as most people think they are. You may be able to command a position to within 1/16 of a step but the motor position is only guaranteed to be within +/- 1 full step of that position. How far you are off depends on the load on the motor and on how fast you are going. At high speeds, the motor position actually lags the microstep position because it takes time for the current to decay and build in each of the windings. With a closed-loop stepper system, there is an explicit phase advance parameter which tells the commutator to advance the microstep position some number of microsteps ahead of where it would normally be, based on the motor velocity, effectively helping to correct for this error.

There are also some "hidden" factors which limit the torque when open-loop stepping. In particular, step timing jitter and mechanical vibration both can superimpose an oscillatory load on top of the nominal load. If the magnitude of the oscillatory plus the nominal load exceed the torque capability, an open-loop system will lose steps. A servo system will effectively average out the oscillatory loads, so you are really only limited by the nominal load. Ironically, the more rigid your printer drive, the more prone an open-loop system is to losing steps. (That's why belt drives work so well with steppers - they give you some vibration isolation.)

Lastly, closed-loop steppers might be next gen technology for 3D printers, but closed-loop PID control of steppers has been done for at least 30 years (see: HCTL1000). I'm happy to take this to another forum, though, if that's more appropriate.

I would like to see this project move forward.

I don't know what's involved to make this work.
Can you contrast your project to this encoder: CUI Inc, AMT102, $23.63.

Capacitive Encoder: CUI Inc, AMT102.
8192 counts/rev or .04 degrees resolution
[builders.reprap.org]

Digi-Key Part Number 102-1307-ND
[www.digikey.com]

CUI Inc, AMT102 Encoder Installation
[www.youtube.com]

Edited 1 time(s). Last edit at 11/07/2013 02:03AM by A2.

Quote
jamesdanielv
I think this is a next gen technology

I'll follow you here when the stepper is replaced with a DC motor.

The entire point of a stepper is to avoid the need for a closed loop. Once you have this closed loop, using a stepper is moot.

---

Generation 7 Electronics	Teacup Firmware	RepRap DIY

I just think that this deserves its own subject in a forum section of reprap where it is easy to find, and contribute. general is subjective, and likely won't get the response you are looking for.

quote "I'll follow you here when the stepper is replaced with a DC motor."

I'ts hard to say what will happen, some times something works great, then again sometimes the same gets better.

Lobo,

This is interesting and I'd potentially be interested in supporting a kickstarter or indiegogo or whatever.

But here's a thought for you: Why not include the stepper motors with the encoder as a single package (maybe as one of the support levels)? That way people, such as me, would not need to go hunting for appropriately strong dual shaft motors.

Quote
Traumflug
I'll follow you here when the stepper is replaced with a DC motor.

For a given torque, does a DC motor use less energy?
Does a DC motor cost less?
Is a DC motor size significantly different?
Can you provide a link to a DC motor that would be a substitute?

Edited 1 time(s). Last edit at 11/07/2013 09:18AM by A2.

Quote
Traumflug
I'll follow you here when the stepper is replaced with a DC motor.

Don't get me wrong, I'm not advocating closed-loop control for most 3D printer applications. Open-loop steppers work remarkably well with belt drives. The loads on the motor are pretty repeatable and you can work pretty close the published speed-torque curve (which is best case). If you need higher speeds, it's just cheaper to use a bigger motor and power supply than to add an encoder and closed-loop controller. For extruders, however, where the load is not that well characterized, and where you want to minimize the motor mass, closed-loop control might be worth it for a higher-end printer. The other good application for closed loop control would be printers with screw drives - lead screws are more rigid and transmit more vibration into the motor, and they also need higher speeds.

Quote
Traumflug
The entire point of a stepper is to avoid the need for a closed loop. Once you have this closed loop, using a stepper is moot.

There are actually a couple of advantages of using a closed-loop stepper over a conventional closed-loop DC servo motor. Steppers are designed for high torque and low-speed, and DC motors for low-torque and high speed. For the same sized motor, a stepper might typically have 4x the torque but run at about 1/4 the speed. This means that steppers can be used direct drive in many applications where a DC motor would need some gear reduction. The second advantage is that because they are made in such high volume, high performance steppers are dirt cheap as compared to a similar performance DC motor.

I think this might be worthwhile. I would like to see a prototype that demonstrates the improvement over an open loop system.

Quote
crispy1
But here's a thought for you: Why not include the stepper motors with the encoder as a single package (maybe as one of the support levels)? That way people, such as me, would not need to go hunting for appropriately strong dual shaft motors.

This is well worth considering. The controller board I made is suitable for a wide range of motors, and my initial thinking is that I don't want to stock a bunch of different motors. However, for RepRaps, I probably could pick one popular motor size and offer that pre-assembled (or at least as a kit). There are a couple of drawbacks, though. I'd be doing this as a hobby business. To build a batch of 100 boards would probably cost me $2500 - $3000 which is doable. To also stock 100 motors (say $10 each), encoders ($20) and enclosures (~$7) would be another ($3700) which may be outside what I'm allowed to spend . Also even as a hobby business, I need to make some profit. I'm willing to go with a 40-50% margin, which is paper-thin for a low-volume manufactured product. If I just sell a board + enclosure for ~$50, then it'd cost someone about $85 once they bought their own encoder and motor. If I buy the encoder and motor for them, I'd need to sell for at least $125. Will that be worth it to most people? What I'm hoping instead is that I can recommended specific vendors and specific models of motors to make life easier.

I would recommend you do a full kickstarter / indiegogo and see how it goes and what people want.

This is high end stuff for people who want the best. The campaign would have to have a demonstration of what happens with and without encoders. A good selling point is that it can prevent print failure.

Some people do massive high resolution prints of over 48 hours (like me) and it's devastating if any steps are skipped. Waste of time and materials, and I don't use cheap materials. Although my printing is not mission critical, for mission critical printing, this would be a must.

Someone pointed out you may still see defects, when compared to step skips and complete print destruction, a little recovered blip can be dealt with.

You can set the campaign up to make it worth your time (desired amount of funding to go ahead) and have options like with and without steppers and encoders.

My bet is for people who want the best they will go with the maximum of what you have to offer and would want it all packaged for them. The tech is about saving time so it makes sense to not have people running around using too much time to finish the units.

I would buy at least three for my hobby without hesitation. Also printers that have them in would be far more valuable and have market edge so other small scale 3D printer manufacturers would want them.

The biggest hurdle is many people don't know or unable to understand the importance of such units. But if they can have a side by side comparison of the results I think they would be far more interested.

I am totally keen. Best of luck putting it together, I will buy some.

Without any feedback to the controller board sending out the steps how would you keep the axis synchronized together when it recovers the lost steps. This would even be a problem on the extruder where it needs to output a specific amount of plastic in a specific place. If the motor ends up doing a lot of extra stepping to be at the correct step location it could be far behind the axis that has not had to recover any steps. Could it not even result in the extruder still stepping during a travel move if it had to do a lot of recovering prior?

TLDR: Wouldn't having a single stepper running a closed loop driver like this cause a buffering effect at each motor that results in non-synchronous movements?

---

FFF Settings Calculator	Gcode post processors	Geometric Object Deposition Tool Blog
Tantillus.org	Mini Printable Lathe	How NOT to install a Pololu driver

Quote
jzatopa
I think this might be worthwhile. I would like to see a prototype that demonstrates the improvement over an open loop system.

I think that would be key. I can see there are benefits on paper, the question is do they translate to practical benefit, and the more question is the extra cost worth it.

My feeling is this would be a fairly niche product for leading edge type people so is unlikely to get floods of backers. Also crowdfunding is quite sensitive to price, for example this pricey extruder [www.kickstarter.com] didn't get funded. OTOH the QU-BD extruder was aimed at a low price point with wide appeal, and did well, even if the extruder wasn't that good nor particularly novel.

---

What is Open Source?
What is Open Source Hardware?
Open Source in a nutshell: the Four Freedoms
CC BY-NC is not an Open Source license

Quote
LoboCNC
The other good application for closed loop control would be printers with screw drives

For worm gear drive delta printers, an encoder mounted on the pivotal arm (not the motor) would help solve the backlash!
Worm gears are notorious for backlash.

A kit consisting of 4 encoders would be great for worm drive delta printers.
3 encoders for the A, B, C axis stepper motors, and 1 encoder for the filament drive stepper motor.

Does the Marlin software need to be modified?

Edited 1 time(s). Last edit at 11/07/2013 06:28PM by A2.

I'm working on getting a side-by-side test together of the same motor, driver chip, supply voltage, but open-loop v. closed-loop. That seems to be the nub of the matter. Stay tuned...

Quote
Traumflug

Quote
jamesdanielv
I think this is a next gen technology

I'll follow you here when the stepper is replaced with a DC motor.

The entire point of a stepper is to avoid the need for a closed loop. Once you have this closed loop, using a stepper is moot.

DC motors are not all that great - you have the brushes that wear out, arcing from commutation creating electrical noise and the low torque / high speed issue mentioned as well.
A stepper motor is similar to a brushless permanent magnet motor (commonly called brushless DC).
This idea has merit - but as mentioned the big sell is in showing the performance advantage.

BTW - in reading your instruction sheet you mention cutting the motor shaft if too long. Is there not a way to just let any excess shaft stick out? Cutting the shaft of a motor is asking for trouble and is an unnecessary bit of labor if it can be designed around..

Good luck.

Mike

Quote
Mike-Ross
BTW - in reading your instruction sheet you mention cutting the motor shaft if too long. Is there not a way to just let any excess shaft stick out? Cutting the shaft of a motor is asking for trouble and is an unnecessary bit of labor if it can be designed around..

The shaft length shouldn't be a problem for NEMA 17 motors, but I realized after the fact that the back shaft on most NEMA 23's is too long. Any production version would be laid out with a hole for long shafts to poke through. (I guess that's what prototypes are for.)

I've been doing some comparison testing of open-loop stepper control v. my closed-loop stepper controller. My test setup is a linear rail with a screw drive (6 rev/in of travel) and a 1.2A, 50 oz-in NEMA 17 motor. I can run the motor open-loop using a smoothie board with A4982 driver (24v) or with my closed loop driver board with a nearly identical driver chip (A4984). For the closed-loop control, I simply feed the Step & Direction signals from the smoothie board into my driver board.

In my initial testing, I was able to run back & forth (0-200mm) at very high motor speeds. Both the open-loop drive and the closed-loop drive topped out at motor speeds a little over 1000 RPM. I was using the simple test code of:

G01 X200.0
G01 X0.0

and ran at faster and faster speeds until the motor stalled out.

I was kind of surprised that I was able to get such high speeds out of the open-loop control because in my past experience, it's often problematic trying to go over 500 RPM (unless using a very high voltage driver). I then decided to create a more difficult path. The worst case scenario for a stepper motor is to have an oscillation in the velocity when travelling at a high speed. What I came up with was:

G01 X100.0 Y100.0 (Move in straight line to 100, 100)
G01 X101.1 Y101.0 (Continue moving Y at a constant speed but force X to speed up and slow down slightly)
G01 X101.9 Y102.0
G01 X103.1 Y103.0
G01 X103.9 Y104.0
G01 X105.1 Y105.0 (do this for 5mm in X & Y)
G01 X200.0 Y200.0 (continue in straight line to 200, 200)
G01 X0.0 Y0.0 (Move back to home)

The Y axis was not actually being tested, but including it forced the path planner to create a high speed path for X with a small oscillation superimposed on it. I then executed this path at different speeds. This time, the fastest open-loop motor speed I could achieve was just over 500 RPM. Closed-loop I could still run the path reliably at about 1000 RPM (although it may not have been able to accurately follow the tessellations in the motion).

Now, the path I created was pathologically bad, but it's not hard to image part geometries creating exactly this sort of high-speed oscillation. Every now and again, I'll slice some part and when I print it, there will be drift in one axis indicating slipped steps. But if I rotate the part on the table and slice it again, it'll print fine. I think this is what's happening - the slicer is generating a similarly pathological path.

I'm currently in the process of building a leadscrew driven 3D printer/CNC machine [reprap.org], and the idea of a closed loop system is very appealing to me. Personally I wouldn't mind having a kit since soldering surface mount components by hand isn't too difficult. If the cost savings are enough then it may be worth adding that as an option for the Kickstarter/Indiegogo campaign. Another thing to remember with these campaigns is there are people all over the world that may be interested, so postage becomes another cost to add. I'm in the UK, so shipping from the US may be expensive.

On another note A2's comment about putting the encoder on the pivotal arm rather than the motor itself is a good one. Using a photodetector array like this [docs-europe.electrocomponents.com] and a pattern along the length of the axis would allow for compensation of any backlash as well as missed steps. It might be worth starting another topic on the idea.

Edited 1 time(s). Last edit at 11/10/2013 07:11PM by samp20.

Quote
samp20
On another note A2's comment about putting the encoder on the pivotal arm rather than the motor itself is a good one. Using a photodetector array like this [docs-europe.electrocomponents.com] and a pattern along the length of the axis would allow for compensation of any backlash as well as missed steps. It might be worth starting another topic on the idea.

Closing a servo loop around backlash in a mechanism is actually quite difficult - especially when your axes are constantly changing direction at high speed. For my particular closed-loop stepper system, I also need the encoder right on the motor so that it can be used for commutation as well as for servo control.

And thanks for your other comments - lots of things to consider in deciding to sell something.

Quote
LoboCNC
Closing a servo loop around backlash in a mechanism is actually quite difficult - especially when your axes are constantly changing direction at high speed.

Can you define what "high speed" is?
Then maybe we can operate under a known threshold to control backlash.


CMM (coordinate measuring machine)
In addition I'm also very much interested in a building a CMM.
I used to operate a Mitutoyo B-231 CMM.
Linear Accuracy: .0003''
Resolution: .0001''
Repeatabilty: .0001''

I'm curious what the performance increase might be had with the CUI Inc, AMT102 encoder.

What is the resolution of a NEMA 17 stepper motor?
From what I have read, it appears that 3D filament printed objects are achieving +-.004", is this correct?
What is the present expectation for dimensional tolerances?

360 deg / 200 steps = 1.8 deg
Sine 1.8 deg over 1 inch = .0314"

360 deg / 400 steps = .9 deg
Sine .9 deg over 1 inch = .0157"

What kind of resolution can one expect from the CUI Inc, AMT102 ?
8192 counts/rev or .04 degrees resolution (are these specifications correct?).
Sine of .04 deg over 1 inch = .0007".

A2

There are two problems with trying to servo out backlash. The first is that every time you switch directions,the motor has to quickly traverse across the backlash before resuming the normal path motion. This will give you little glitches in your path, but that may not be too bad if they don't occur too often. However, say you have 3 deg. of backlash in your gearhead and you are making a bunch of quick back & forth motions of only a few degrees. In this case, half the time the motor will be moving across the backlash and your axis won't be moving at all.

The second problem comes in when an axis is unloaded. Say you are moving just along X with no change in Y. If any vibration causes the Y axis to move across the slop in the drive train, the servo will try pushing it back. But without any load on the Y axis, it's really easy to have the Y axis motor jitter back & forth across the backlash, creating a lot of vibration. You need a pretty sophisticated control algorithm and a pretty good dynamic model to handle this type of situation.

Regarding the AMT102 encoders, they can be set for 8192 counts/rev (or 0.043 deg/count) but the actual accuracy is only spec'ed at 15 arc min, or 0.25 degrees. For accuracy, there's no real reason to set the resolution higher than 400 pulse/rev (or 1600 counts/rev.). They probably include higher resolutions for compatibility with systems already using a different encoder resolution. Also, the higher resolution can be helpful in servo control systems for achieving greater damping.

Quote
LoboCNC
...But without any load on the Y axis, it's really easy to have the Y axis motor jitter back & forth across the backlash, creating a lot of vibration.
You need a pretty sophisticated control algorithm and a pretty good dynamic model to handle this type of situation.

Springs on all the joints!
I will design out the backlash.

Quote
LoboCNC
...the actual accuracy is only spec'ed at 15 arc min, or 0.25 degrees.

Resolution: Sine .25 deg over 1 inch =.0044"

Stratasys Dimension Rapid Prototyping (RP) machine
The product will be accurate to ±0.05 mm. (±.002 inches )
[reprap.org]

What steps are required to integrate your encoder to a RAMPS controller?
For example, will I have to make code modifications to Marlin, purchase additional hardware?
Sorry if these questions have been addressed in your paper, I'll read it more carefully.

Is it possible for you to keep the external dimensions of the encoder enclosure within the stepper motors foot print i.e. 42.3 mm?

Ok, so my expectation for the encoder is for robustness, which allows me to validate my equipment, which is great!
I would like 4 of your encoders

A2

Edited 1 time(s). Last edit at 11/15/2013 01:31AM by A2.

Idea: How about using the encoder for auto bed leveling?
A simple cantilever type of configuration might be all that is needed.

A2

Example of an magnetic encoder attached to an extruder:
[www.grassrootsengineering.com]

Edited 1 time(s). Last edit at 11/30/2013 10:37PM by A2.

Quote
A2

Example of an magnetic encoder attached to an extruder:
[www.grassrootsengineering.com]

Thats a DC motor with encoder. If you read the blog you see he has replaced that with a geared stepper motor (without encoder it seems)

Quote

The old plastic extrusion tool (left) next to the New extrusion tool (right). The new extruder takes in a feedstock of 1.75mm ABS plastic filament. The filament is compressed between a stepper motor with a planetary gearbox which drives fine toothed gear and a spring loaded idler wheel. The filament is then forced down through a 0.35mm nozzle heated to 220°C. This new extruder is about a quarter the size of the old extruder and also effectively doubles the resolution of the the machine. (Note: New extruder is shown without nozzle/heater portion attached)

---

What is Open Source?
What is Open Source Hardware?
Open Source in a nutshell: the Four Freedoms
CC BY-NC is not an Open Source license