Pololu geared DC motor 30g prototype extruder

Quote
rklauco
My the suggestion of Misan I tried the EU offices - the package came from Austria.
The problem with the magnet of yours is the orientation - check the desired orientation on the documentation from ASM.
Even putting the magnets on the washer will not modify the magnetic field orientation
My idea was to put 3-4 together and glue them using some hot glue or acrylic side-ways.
First I have to test if the strength is sufficient for the AS5600.
Today (maybe) I'll finally have some time.

My understanding is that the AS chips expect diametrically opposed polarization - N on one half, S on the other, or NS across the entire magnet. In my arrangement the washer is just a temporary holder, and we have NSNSNS, which gives just more density, so the resolution is increased as the transitions are sharper (we don't need it, but I already have the magnets).

Wow, I have to browse how did they made extruder with Mabuchi 130 - have 5 fakes and 2 originals on my messy desk...

Well, I would say it may work
I didn't want to wait for a magnet - so I just took 5 magnets with 8mm diameter and 1mm width. I glued them onto the gear wheel using 2-3 drips of hot glue.
Then I burned one AS5600 by accidentally putting 12V onto its output pin.
But later I grabbed another one, spend few hours just to finally read the manual and realize I need 1uF capacitor and as soon as I connected the capacitor, voila!
I had to adjust misan's code quite a bit as he only has code for ESP8266 and AS5600.
Also, I lowered the resolution from 3600 steps per 1 turn of the gear to 3600 steps - much more reasonable.
The movement in the video goes from 20 to 150 and back to 0.
Clearly, I did not work with PID settings (that's why I am overshooting so much), but I cannot tune it now - my kids are sleeping and everything is quiet, so even this small movement in the video made a LOT of noise and woke up my dog
So, I think I am done for today, but it looks like great progress.

Quote
rklauco
Wow, I have to browse how did they made extruder with Mabuchi 130 - have 5 fakes and 2 originals on my messy desk...

Not sure if they indeed used 130s, but the size, vent holes, and arrangement indicate something of that nature. I got some original 130s with encoder from old scanners but worm drives are not very efficient and the entire package is heavier than what we have here. Sourcing the paired worm gear and large helical gear might be an issue.

Quote
rklauco
Well, I would say it may work
I didn't want to wait for a magnet - so I just took 5 magnets with 8mm diameter and 1mm width. I glued them onto the gear wheel using 2-3 drips of hot glue.
Then I burned one AS5600 by accidentally putting 12V onto its output pin.
But later I grabbed another one, spend few hours just to finally read the manual and realize I need 1uF capacitor and as soon as I connected the capacitor, voila!

WOW! First proof that we are not crazy going for something so small

I tried to get the setup more mainstream and adapted the Mk8 gear using #6 brass inserts (didn't have M3 nearby, small pipe should work too). With Mk8 the grip is more uniform as it seems to be machined to touch the sides of the filament too.

Now if China or some local manufacturer reads this thread....
please, please make a version of Mk8 that is 2mm shorter by getting the set screw hole right next to the grooves and with 3mm diameter hole.
Less material=cheaper and lots of happy customers for the first to get to market! Now get those CNC lathes going!

The proof is in the pudding as they say. The test setup is 17g but now anybody can replicate it as there are no more custom parts.





Parts:
1x Mk8 extruder gear
1x 683 bearing (7x3x3) - can be replaced with 623 or 624, just tweak the scad file
1x N20 geared DC 6V micro motor 1:100
1x M3 screw 14mm
2x M3 or #6 brass inserts to use as adapters from gearbox shaft to Mk8

Quote
rklauco
I had to adjust misan's code quite a bit as he only has code for ESP8266 and AS5600.
Also, I lowered the resolution from 3600 steps per 1 turn of the gear to 3600 steps - much more reasonable.
The movement in the video goes from 20 to 150 and back to 0.
Clearly, I did not work with PID settings (that's why I am overshooting so much), but I cannot tune it now - my kids are sleeping and everything is quiet, so even this small movement in the video made a LOT of noise and woke up my dog
So, I think I am done for today, but it looks like great progress.

Do you have a schematic / pic of the board that you used? I ordered a couple AS5600 and AS5601 to try and interface directly to the Azteeg, hopefully they'll be here by next week.

Edited 1 time(s). Last edit at 05/28/2016 12:59AM by ipcalit.

Quote
ipcalit
Do you have a schematic / pic of the board that you used?

Well, I tried to do the schematics - don't kill me, this is my first Fritzing attempt
However, I run into a problem.
When I start up the whole thing, it measures the angle - and it jumps a LOT.
E.g., no motor movement (completely disconnected from all sorts of power), I have this reading:

Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-117.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-73.00 PID_output=80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-102.00 PID_output=80.00 Target=0.00
Position=-73.00 PID_output=-80.00 Target=0.00
Position=-116.00 PID_output=80.00 Target=0.00
Position=-29.00 PID_output=80.00 Target=0.00
Position=-73.00 PID_output=-80.00 Target=0.00
Position=-256.00 PID_output=80.00 Target=0.00
Position=43.00 PID_output=-6.39 Target=0.00
Position=72.00 PID_output=-80.00 Target=0.00
Position=58.00 PID_output=-80.00 Target=0.00
Position=43.00 PID_output=-80.00 Target=0.00
Position=14.00 PID_output=80.00 Target=0.00
Position=0.00 PID_output=79.76 Target=0.00
Position=87.00 PID_output=-80.00 Target=0.00
Position=29.00 PID_output=-80.00 Target=0.00
Position=29.00 PID_output=80.00 Target=0.00
Position=29.00 PID_output=80.00 Target=0.00
Position=14.00 PID_output=51.97 Target=0.00
Position=28.00 PID_output=80.00 Target=0.00
Position=29.00 PID_output=80.00 Target=0.00
Position=14.00 PID_output=80.00 Target=0.00
Position=14.00 PID_output=80.00 Target=0.00
Position=29.00 PID_output=80.00 Target=0.00
Position=14.00 PID_output=80.00 Target=0.00
Position=0.00 PID_output=-80.00 Target=0.00
Position=0.00 PID_output=80.00 Target=0.00
Position=14.00 PID_output=80.00 Target=0.00
Position=0.00 PID_output=80.00 Target=0.00
Position=29.00 PID_output=-80.00 Target=0.00
Position=72.00 PID_output=80.00 Target=0.00
Position=-256.00 PID_output=80.00 Target=0.00
Position=14.00 PID_output=80.00 Target=0.00
Position=43.00 PID_output=80.00 Target=0.00
Position=15.00 PID_output=-80.00 Target=0.00
Position=0.00 PID_output=80.00 Target=0.00
Position=0.00 PID_output=80.00 Target=0.00
Position=73.00 PID_output=-80.00 Target=0.00
Position=0.00 PID_output=80.00 Target=0.00
Position=14.00 PID_output=-80.00 Target=0.00
Position=14.00 PID_output=-80.00 Target=0.00
Position=-15.00 PID_output=80.00 Target=0.00

I don't understand what's going on. There is no other source of magnetic field around as far as I am aware...
@Misan: Do you have similar experience, or is it just me?

Are you reading the I2C or the analog output? My experience is that measure is quite accurate.

I use the recommended magnet, if not I have no experience of what might happen.

Please note there are a couple of capacitors in the recommended schematic.

Edited 1 time(s). Last edit at 05/28/2016 09:24AM by misan.

I don't have the recommended one, that's for sure.
I am using the capacitors after the fiasco with the 3.3V rail
But I think I found the reason.

When there is no filament in the extruder, the shaft is a tiny tiny bit up. That's by design - I achieve a bit better grip by creating the pressure by this slight dislocation.
So, when filament is loaded, the measurement is rock steady. When it is out, I receive a field fluctuation like the one above.

While the current situation is fixed by simply loading the filament, it makes me wonder - how big the impact is on the measurement if the filament diameter varies even slightly.
To compensate I introduced simple division by 8 - so now the precision is 512 pulses per 1 rotation of the gear. As the diameter is 10.5mm, the resulting length of the extruded filament is ~33mm. That gives resolution of 0.065mm per pulse or 15 "steps" per mm. That's acceptable for me.
I will experiment with 1024 pulse resolution (division by 4) to test the measurement and system stability.

Oh, and I forgot, of course I am working with i2c.

Edited 1 time(s). Last edit at 05/28/2016 10:44AM by rklauco.

Quote
rklauco
When there is no filament in the extruder, the shaft is a tiny tiny bit up. That's by design - I achieve a bit better grip by creating the pressure by this slight dislocation.
So, when filament is loaded, the measurement is rock steady. When it is out, I receive a field fluctuation like the one above.

I my design the driving gear is stable and just the bearing moves slightly to accommodate the filament. This should not create variations in the magnetic flux.

Quote
rklauco
To compensate I introduced simple division by 8 - so now the precision is 512 pulses per 1 rotation of the gear. As the diameter is 10.5mm, the resulting length of the extruded filament is ~33mm. That gives resolution of 0.065mm per pulse or 15 "steps" per mm. That's acceptable for me.
I will experiment with 1024 pulse resolution (division by 4) to test the measurement and system stability.

Mk8 is 6.8-7mm dia at the grooves and about 20-21mm circumference, so the resolution at 512 pulses should be 0.04 (or 0.02 at 1024). I looked at some Gcode files I had and it seems that it should be enough.

Quote
rklauco
Oh, and I forgot, of course I am working with i2c.

Try the analog input too. I'm curious if you get identical measurements or close enough.

Quote
ipcalit
I my design the driving gear is stable and just the bearing moves slightly to accommodate the filament. This should not create variations in the magnetic flux.

How did you achieve the flexibility in the bearing?
I'm interested in the technical solution as mine is rock solid...

Quote
rklauco

Quote
ipcalit
I my design the driving gear is stable and just the bearing moves slightly to accommodate the filament. This should not create variations in the magnetic flux.

How did you achieve the flexibility in the bearing?
I'm interested in the technical solution as mine is rock solid...

The trick is to use the material properties and printing technique in your advantage.
1) I'm using PETG instead of PLA or ABS, so there's some give at 0.8-1.2mm wall thickness. Depending on the thickness of the part holding the M3 screw you can get a stiffer or more flexible bearing mount.
2) Bring the bearing about 0.3mm closer to the gear in comparison with the calculated position considering the filament as just touching both the gear and the bearing. This creates enough pressure to make grooves into the filament as it advances. The material going around the gearbox has a larger surface and won't give in as much as the "U" part that holds the screw and bearing.
3) Look carefully at your layers orientation vs mines. The stress in your model from feeding the filament goes against your Z leading to layer splitting, whereas in mine the stress is on X/Y which is much stronger. I had to print using supports, which is a pain with PETG, but worth it. Also, notice I made a cutout on the opposite side of the bearing as I have no load there. This reduces weight and allows me to clean the gear from filament bits and pieces. Same idea with the E3D Titan.

Edited 1 time(s). Last edit at 05/28/2016 05:29PM by ipcalit.

Quote
rklauco
I don't have the recommended one, that's for sure.
I am using the capacitors after the fiasco with the 3.3V rail
But I think I found the reason.

When there is no filament in the extruder, the shaft is a tiny tiny bit up. That's by design - I achieve a bit better grip by creating the pressure by this slight dislocation.
So, when filament is loaded, the measurement is rock steady. When it is out, I receive a field fluctuation like the one above.

While the current situation is fixed by simply loading the filament, it makes me wonder - how big the impact is on the measurement if the filament diameter varies even slightly.

Got my AS5600 samples and 6mm magnets -- BIG thanks to AMS for quick delivery and great packaging. Looking on page 33 of the specs I think you're right regarding small deviations of the magnet from the top of the AS5600:

"The maximum allowed displacement of the rotational axis of the reference magnet from the center of the package is 0.25 mm when using a magnet with a diameter of 6mm."

AS5600_DS000365_2-00.pdf

This means we need some precise reference for where the AS5600 sits in relation with the gearbox. Will try so sketch something the following days.

Did you use the schematic from Figure 38 for interfacing to your controller via I2C? I see you used a board... purpose made or repurposed from other projects?

I have used SOIC8 adapters from eBay

Then press fit into square hole in a plastic part like this [www.thingiverse.com]

Quote
misan
I have used SOIC8 adapters from eBay
Then press fit into square hole in a plastic part like this [www.thingiverse.com]

b30bd63ea4f0953412342891a1c65d57_preview_featured.jpg

In your design it seems that you're centering the entire board over the magnet. Depending on the accuracy of your soldering skills and the printed parts did you experience any fluctuation based on 0.25mm deviation from axis like rklauco mentioned? I'm thinking about centering just the AS5600 over the magnet to get tighter tolerances.

Also, which driver did you use? Any experience with TI DRV8838 or DRV8835 (dual) from Pololu or did you just use the classic L293N/L298N?

[www.pololu.com]

No, I did not notice any relevant variation. However, I have not performed a dual-technology test (using a second optical encoder attached to the same shaft). So I might be fooled by my own perception.

Recently I saw the project Mechaduino that does a big effort to compensate the errors of their magnetic encoder. Mecaduino

I have tested drivers 6612 [www.pololu.com] and L298N but I am waiting samples of this other one for beefier motors [www.pololu.com]

If you do not have enough current for your motor then performance will be subpar.

Edited 1 time(s). Last edit at 06/07/2016 03:04AM by misan.

I tested L293, but it wasn't able to withstand the ~30kHz I am using to drive the motor.
The L298 is much better although a bit too big for me.
However, it's a bit too big, so I worked with TB6612FNG and that was fantastic. I did have some bad experience with 1 batch from China that had all the capacitors badly rated for 10V and they blew up as soon as I connected 12V.
However, the TB6612FNG can only handle 12V motors. As I have some 24V (and I do plan to use them due to great torque) I ordered and yesterday received the A4953 (up to 40V) and will test them today/tomorrow.

I'm keen to jump in as well, have ordered the parts and am waiting on the slow boat from China.

Not sure if you guys have come across this driver, but I've ordered one to try out.

[www.ebay.com.au]

Quote
ipcalit
Any experience with TI DRV8838 or DRV8835 (dual) from Pololu or did you just use the classic L293N/L298N?

The DRV8838 and 8835 are both only for 11V and 1.2A. That's not really enough for majority of stronger DC motors - usual 2D printer motors work with 24-36V and although they are able to work with 12V or less, the torque is lost quickly.

Quote
nebbian
Not sure if you guys have come across this driver, but I've ordered one to try out.

[www.ebay.com.au]

Reading the specs, it seems like it's only good for 12V, but I have some trouble with understanding the datasheet - it looks like "Output HIGH" will only get 7.7V, not 12V...

Quote
rklauco

Quote
ipcalit
Any experience with TI DRV8838 or DRV8835 (dual) from Pololu or did you just use the classic L293N/L298N?

The DRV8838 and 8835 are both only for 11V and 1.2A. That's not really enough for majority of stronger DC motors - usual 2D printer motors work with 24-36V and although they are able to work with 12V or less, the torque is lost quickly.

They should be enough to drive the small extruder and are both cheap enough in board form directly from Pololu. They are definitely not powerful enough for anything like 370, 540, 550s at 24-36V in the X/Y stages, but at this time I'm relying on my CoreXY steppers to do the heavy lifting.

Quote
misan
No, I did not notice any relevant variation. However, I have not performed a dual-technology test (using a second optical encoder attached to the same shaft). So I might be fooled by my own perception.

Recently I saw the project Mechaduino that does a big effort to compensate the errors of their magnetic encoder. Mecaduino

I saw Mecaduino and the error compensation is valuable, but the overall approach is not really a good match for the 3d printing world. We need the controller to perform close-loop over all three axes (general case, just two for cartesians) and the extruder at the same time, not just for each motor. There's no gain in performance or quality if X axis is temporarily blocked and on the path to recovery, whereas Y and the extruder keep on going about their business as usual.

Quote
misan
I have tested drivers 6612 [www.pololu.com] and L298N but I am waiting samples of this other one for beefier motors [www.pololu.com]

If you do not have enough current for your motor then performance will be subpar.

I have the L298N but as rklauco said, fast PWM is not possible. The VNH5019 seems a bit overkill with its 12A at 24V - you could drive a CNC with it, or lower Z after printing and just kick the print off the bed

Edited 1 time(s). Last edit at 06/07/2016 12:50PM by ipcalit.

Well, I am disappointed. I tested the A4953 and while it works, it gets extremely hot even during position hold.
And while doing back and forth 120mm movements it got so hot I could not put finger on it.
I had to remove it and put back my TB6612, which works flawlessly.
I had to admit, though, that my temporary board is WAY off the reference design - so probably the missing heat-distribution is accountable for most of it.
What worries me, though, is that the chip should be able to work with 40V and 2A (80W), while I was feeding it 12V and 0.3A (3.6W), so less then 5% of it's designed power capabilities
Someone has different experience with A4953?

Quote
rklauco
Well, I am disappointed. I tested the A4953 and while it works, it gets extremely hot even during position hold.
And while doing back and forth 120mm movements it got so hot I could not put finger on it.
I had to remove it and put back my TB6612, which works flawlessly.
I had to admit, though, that my temporary board is WAY off the reference design - so probably the missing heat-distribution is accountable for most of it.
What worries me, though, is that the chip should be able to work with 40V and 2A (80W), while I was feeding it 12V and 0.3A (3.6W), so less then 5% of it's designed power capabilities
Someone has different experience with A4953?

Couldn't find your A4953 but Pololu has the G2 18v17 and VNH5019 drivers for ~$30. If you read their real-world report of the TB6612, DRV8833/5/8 all these drivers get hot with low voltage at 100% duty cycle. Perhaps the trick is to use high voltage but low current motors to get the drivers stay relatively cool. Also, did you try adding some simple Alu heatsink to the chip or clamp/screw it face down over some part of the metal frame of the printer?

Well, I simulated 2 8-hour long prints with X and Y axis on a single TB6612 and it was as cool as it started. The motor got worm to touch, but the TB did not.
I did not put any sort of heat sink on the A4953, I can try it tomorrow and report back the results.
I would still like to make the A4953 work - I have a board in mind with ATtiny85, A4953, 2 capacitors and form factor that would make it a drop-in replacement on RAMPS boards... But maybe I am wasting my time and it would be easier to make an add-on board for Arduino Mega, similar to RAMPS shield...
However, I like the experiments

OK. I put little aluminium heat sink on top of the chip - the same one I am using on my stepper drivers. That seems to help - now the temperature stabilizes around 40 degrees, I can touch the heat sink no problem.
I have to change my wiring a bit - I have the 100uF replaced with 470uF - that seems to improve the reliability of motor start, but I cannot find 100nF around - have to do a little shopping.
But I am starting to like it a bit more

Quote
rklauco
Well, I simulated 2 8-hour long prints with X and Y axis on a single TB6612 and it was as cool as it started. The motor got worm to touch, but the TB did not.
I did not put any sort of heat sink on the A4953, I can try it tomorrow and report back the results.
I would still like to make the A4953 work - I have a board in mind with ATtiny85, A4953, 2 capacitors and form factor that would make it a drop-in replacement on RAMPS boards... But maybe I am wasting my time and it would be easier to make an add-on board for Arduino Mega, similar to RAMPS shield...
However, I like the experiments

Given that you'll be switching to DC everywhere, I see no reason for plain RAMPS compatibility for a couple extra MOSFETS there, just do your own DCRAMPS as you have more flexibility with the layout and chip selection. That's what Panucat did with their Smoothieboard clone and the packaging is really nice and compact compared with lots of add-on boards in the name of upgrade-ability. With some work you could even have a Raspi zero board add-on that has an ATmega or CortexM0 to keep all motors in sync. With octoprint and cheap parts from inkjets that would sell like hotcakes.

That was my initial idea - just instead of Zero (due to virtually no availability) I plan to use Orange Pi One.

Quote
rklauco
That was my initial idea - just instead of Zero (due to virtually no availability) I plan to use Orange Pi One.

Not familiar with this clone, but I got a couple Raspi2 and Odroid-C1s that I can play with. Awaiting on some SOI8 adapters for the AS5600 and the TB6612 and DRV8835s drivers to test.

So. I received on Friday the geared motors with extended shaft and encoders.
First of them has some strange encoder - it does work, but the output seems to be very strange - floats somewhere at 2.5V and spikes to 5V and 0V like when there is a capacitor in the way. Therefore it is not working with my Arduino code as the spikes are too short and are not detected by the chip. I'd have to add additional electronics or something. I can post a screenshot of the oscilloscope output.
The second one is much easier to work with - it has a regular hall probe (2 of them).
However, the resolution is quite poor (did not have the chance to measure it, I just quickly tested it).
The problem here is the transmission - while the motor I did the original tests with was 1:100 gear ration one, this one is ~1:30. And as the motor gear is bigger than the one with 1:100, I cannot simply switch the gearboxes
I'll do further investigation on how I can work with it and what is the resolution per turn - I hope I'll have enough time today in the evening.
However, this seems much better solution then AS5600 from my perspective as I was not able to make it work properly - I still received "jerks" and the detection of overturn was not working for me properly, too.

I am thinking about some different approach.
The idea of measuring the actual movement of the filament stuck in my head.
What I am thinking of is a solution like this:

• do not measure the motor movement
• instead measure the movement of the filament
• introduce a "spring-like" mechanism to push the filament string against a steel rod
• the rod would be covered with heat-shrinking sleeve where the filament meets it to ensure better friction
• the diameter of the rod would be between 7-8mm
• the rod will have something like this mounted at the end of it
• we can measure the movement using the encoder with a resolution at least 100 lines (with possibility to quadruple this to 400) while the diameter of the rod will be approximately 8mm - resulting in a resolution of 0.063mm per impulse
• here is the possibility to decrease the diameter of the measuring rod - 3mm would result in 0.024mm per impulse. The less the diameter the less the friction on the moving parts

Does anyone else think this is reasonable? Or is it just a "blind" way?