Clash with Backlash

Discussions about mechanics
Post Reply
HLA
Posts: 73
Joined: Sat May 05, 2012 11:35 pm
Location: Vienna, Austria

Clash with Backlash

Post by HLA »

This thread arose from the discussion on how to get high-quality prints.

One of the common issues with moving mechanics where direction changes are involved is backlash ().

I experience some backlash on my Felix, which becomes evident when printing circles (see picture). Now, there are two sources of backlash and thus two ways to cope with it:
Calibrate mechanical setup: http://www.woodsmokeworkshop.com/wordpr ... 0#more-340
Measure and adopt in software: http://www.woodsmokeworkshop.com/wordpress/?p=359

After trying the mechanical part (without noticeable success), i went on to measure backlash and adjust the settings in SFACT (Tab "Lash") using the Backlash-Maze (see the thread above for the link to the STL file and instructions). I've got following results:
Backlash X: 0,09
Backlash Y: 0,23 (wow)

After typing these values into SFACT, my prints got even worse and the circles are now tilted in another direction. This indicates that the entered backlash values are too high and/or the source of the oval circle is somewhere else.

Will try to print with 0,15 Backlash for the Y-axis while keeping my fingers crossed... stay tuned
Attachments
Backlash.JPG
Backlash.JPG (16.14 KiB) Viewed 1817 times

stratogavster
Posts: 120
Joined: Tue Jul 17, 2012 9:04 pm
Location: United Kingdom

Post by stratogavster »

Hi HLA,

Picked up on your response to my other reply....

You're correct that any meaningful measurement of the pulley diameter may be tricky....but as i'm new to all this firmware / software CNC stuff is it possible to just ask the printer to print 1 line at 10mm at it's known number of steps? in both axis? just to try and disprove the theory of a wrong config vs hardware? surely it can be tuned manually by increasing / decreasing steps needed for the 10mm...and each test 'line' could be measured easily with a vernier?

I'm curious to find out what the firmware thinks the hardware can do.....i'm sure mine is off

HLA
Posts: 73
Joined: Sat May 05, 2012 11:35 pm
Location: Vienna, Austria

Post by HLA »

Hey stratogavster,

In theory, you are right: you would print a line and measure the length. However, such measurement in practice is close to impossible to being precise enough. This is mainly due to the fact that extruded filament has a significant deviation in width. You can notice this with a naked eye. This will cause deviations of around 0,2-0,4mm at least in the measurement. For the tolerance of a good pulley (e.g. better than ISO h6), which can be as low as few um, you would need to print significant length to reduce the error of the filament. This raises the next issue: calipers that measure long lengths are expensive and the print size is limited. I don't even mention the systematic error of taking manual measurements, and the error/offset of the caliper, which is usually not normally distributed to be averaged out by multiple readings.

So if we have a pulley with a good tolerance, then the best way would be to use theoretic values IMHO.

Now to the deviations in your print. You mention that the cube you printed resulted in 24.5x24.2 x-y size. If we assume that the deviation to the digital model (lets take maximum, i.e. 25-24.5=0.5mm) is due to the diameter error of the pulley, then the pulley would have to have a pretty bad manufacturing tolerance (worse than ISO h12), i.e. around 0.1mm - which would be unacceptable for a pulley per se IMHO. This would also imply that a printed object of size 150x150mm would result in an error of 3mm (!). However, I am quite certain that this wouldn't be the case on your printer. Most probably, the error would be of the same magnitude - around 0,5mm. This would be an indication of a backlash.

On your prints, I noticed that your layers are not aligned perfectly. This is also an indication for backlash. In case of the bad tolerance of pulleys, you would observe well aligned layers.

Looking forward to reading your thoughts on this.
Best, HLA

stratogavster
Posts: 120
Joined: Tue Jul 17, 2012 9:04 pm
Location: United Kingdom

Post by stratogavster »

Hi HLA,

Great response! added more sense to what was very much theory to me at the moment...

All makes sense and you're correct regarding the tolerances of that order. I paid little attention at this early stage to exactly how said tolerances would manifest themselves at the outputs of the mechanics, but using your 150x150 example i'm certain you'd also be correct by suggesting the error would be of the same magnitude. I got a bit carried away with the thought of firmware and steppers etc...as I say it's all a bit new to me :)

Have you managed to successfully measure backlash on the Felix?

I'm quite interested in a method, not for software compensation but more for a mechanical redesign to offer more guarantees of lower or zero backlash...hence my dream of repeatable and easy 3D printing

Great links by the way, good find! will read through those :)

Thanks!

Stratogavster

Lioric
Posts: 116
Joined: Wed Aug 15, 2012 1:13 am

Post by Lioric »

Do not use that backlash calibration object, it is dependent on so many other factors to be already calibrated that it makes no sense before you already know that you have all the other in place first

For me it produced absurd values that resulted in even worse circles. What worked better is the hollow cube that I posted in the other thread

Kjetilei
Posts: 334
Joined: Fri Mar 16, 2012 8:26 pm
Location: Stavanger, Norway

Post by Kjetilei »

Hi Loric,

Are you referring to the object linked in the woodsmokeworkshop blog? Thanks for the feedback.

HLA: Would you mind if you/I/anybody added as much as possible about backlash problems/tricks/solutions in this thread? To keep the discussion in one place.

Lioric
Posts: 116
Joined: Wed Aug 15, 2012 1:13 am

Post by Lioric »

Yes, that calibration object from woodsmokeworkshop blog, at least for me produced very deviated results that made circles a lot worse. After inspecting the printed object is clear that it depends on other number of factors to be properly calibrated first

For me the object that I posted in another thread was better and produced results more in relation to fixing the backlash issue. I will post the link here later

HLA
Posts: 73
Joined: Sat May 05, 2012 11:35 pm
Location: Vienna, Austria

Post by HLA »

@Stratogavster: I am glad it could bring some clarity. I agree that mechanical adjustments to reduce backlash are way better than software compensation. However, there are mechanical limits and backlash can never be eliminated completely - thats the nature of it. Nevertheless, Guillaume's prints demonstrate that there is potential to significantly reduce backlash without massive intrusion into SFACT's settings. If you have any suggestions to modify the mechanics - I am open for experiments as my previous posts show :)

@Lioric:
After measuring the y-backlash to be 0,23mm and ruining another print, the mentioned maze-test proved to be useless - thanks for confirming this, Lioric. As you pointed it well out, there are too many factors that affect the maze. Even a different slicer might create different paths which create different backlash patterns. So I tried the good-old iterative trial and error method:
Increase Lash value -> print calibration object-> measure deviation -> if deviation present {repeat} else {end}.

This resulted in some strange things: circles got better, rectangles got worse :shock: So I decided to play with mechanics: Idlers on the y-axis got additional small washers to reduce belt friction with large washers, added bearing-lubricant to the idler, tightened belt more, increased power to y-motor. This seemed to improve the backlash only slightly. So I guess the thread will stay alive for a while...

@Kjetelei: Please post as much information and thoughts as you would like. Lets win the battle with backlash!

stratogavster
Posts: 120
Joined: Tue Jul 17, 2012 9:04 pm
Location: United Kingdom

Post by stratogavster »

After more considerable thought and study I have a question to ponder.....

What is the difference between 'backlash' and 'repeatable positional accuracy''?

My understanding of backlash in this application is simply unwanted return slack on the linear slides....after playing I can sense very little, my theory being the system uses tensioned belts, not lead screws. What I can sense lots of is what i'm calling 'play' - The linear bearings allowing axial movement, of course unwanted!!

After watching a part print I was paying a lot of attention to vibration and any sight of feel of backlash, but what I was more concerned with was the printers ability to hit the same real X and Y positions layer after layer.

Any chance anybody could demystify my interpretation of the 3 mentioned phenomenon -

1.) Backlash
2.) Play
3.) Positional Accuracy

4.) ...and their relationship


Thanks :)

HLA
Posts: 73
Joined: Sat May 05, 2012 11:35 pm
Location: Vienna, Austria

Post by HLA »

1) Backlash: 'lost' motion when changing the direction of movement. Occurs in systems with non-fixed motion-related connections such as gears, belts, pulleys, etc.
Check http://en.wikipedia.org/wiki/Backlash_(engineering)

2) Play: Try to wobble the components (e.g. extruder) with your hand when the stepper holds the position of the axis. If you sense that the extruder can wobble, then you have play. Backlash is also a kind of play, but very specifically for a play when you change the direction. Play in general can occur in any direction and usually refers to the bearings and axes. This is something you cannot avoid since no play means a very rigid connection resulting in high friction. Thus, when calibrating a mechanical system, you need to make a compromise: friction/play.

3) positional accuracy: the accumulation of all things that result in less then perfect positioning (=error) + resolution of steppers, electronics, etc. (=limitation).

Sorry for less than ideal technical description, but hope this explains it.

Post Reply