Firmware Z wobble compensation (Page 2 of 5)

Hi, yes, you should flash the printer with "my" firmware. Just follow the official guide, with the package downloaded from my github instead of lawsy's... You shouldn't notice any difference in the printer behavior unless you explicitly turn on the wobble compensation, as my firmware is a direct fork of lawsy's.

Tonight I will post the gcode of some test prints that should help you calibrate the wobble with little hassle, so stay tuned

I suggest checking the calibrations from the Left front side as my my front right had some misaligned layers that are visible in the gcode.

Using Rince's calibration set, it looks like I will be using:

M97 A0.01 W1.41 P170

I ended up printing out all of the gcodes so that I could see the whole gambit of possibilities.  I'll take pics of each one and throw them up soon as well as some actual prints trying out this setting.

Thanks for the feedback so far guys! I'm glad it's working! 0.1 is quite a wobble you have there... I'm surprised but happy that you found the right parameters.

Unfortunately you cannot. I can send you the stl files if you like, but you would need to add the phase/amplitude increases yourself.

However, the settings that you identify with these calibration prints should also hold at .1mm, even though this layer height is more sensitive to wobbling in my experience. I suggest to get a coarse tuning with the calibration phantoms and then print small cubes (1x1x1cm) at .1mm by varying the parameters around values that you found in order to find the optimal settings.

I've dialed in the sinusoidal z-compensation on my system recently (after a lot of fiddling over the last month or so). I'm a pretty happy camper.

One of impacts that was very measurable was that, after optimizing the compensation, I had to recalibrate the Extrusion Multipliers for my filaments upwards by 10 to 15%. When the threads are all the same height and width, there is a lot more plastic in a .42mm wall than in a banded wall of the same dimension.

Due to the density increase, I'm guessing that the strength of most parts is increased  after compensation, although I have no data.

P.S. As full disclosure, I haven't yet gone the firmware modification route yet, but am running a home-made script using the "Post Slice Filter" feature in Repetier-Host V0.84 to adjust all the z-values in the G-code as it is generated. (I really had to stretch to remember scripting, but managed)

I've been experimenting with determining compensation values *directly* instead of just using trial and error.  I found it very helpful, so here's the procedure I used since I don't have a dial depth gauge:

1) Move the bed to home position and balance a small mirror between the edge of the bed and a stack of blocks or books.  The idea is to have the mirror tilt slightly when the bed moves up and down.  Mine was about 2" across, which is small enough to magnify the bed movement but large enough to still be fairly linear for the range of movement we'll be measuring.  I taped one edge of the mirror to the bed to keep it from slipping.  (see photo, sorry for fuzziness)

2) Mount a laser pointer so it shines off the mirror onto a sheet of paper taped to the ceiling or other flat surface at least a few feet away.

3) Mark the initial laser spot position, and then move the bed down in .1mm increments, measuring each spot.  Measure at least 14 new positions, which covers one full revolution of the threaded rod.  Then go backwards, moving the bed up in .1mm increments until it goes back to home.

4) On the sheet of paper, measure the distance of each mark from the initial spot and write down these numbers.

5) Since we're only interested in periodic errors, we can use any two measurements that are one revolution (14 bed movements) apart as our reference points to scale the measurements.  We can assume that the 14th measurement represents a bed height of exactly 1.4mm since it should have the save wobble as the home position.  Thus, to scale the measurements, multiply each value by 1.4 and divide by the distance measurement for the 14th spot.  The newly scaled values will then equal the Zactual corresponding to each desired Zrod position (.1mm, .2mm.. etc)

6) Chart the difference Zactual - Zrod, which is the error we need to compensate for (see attachment).  In my case, the forward measurements showed a significant amount of non-periodic variability which suggests that I need a stronger spring for my z-axis preload.  The "backwards" measurements, however, taken by moving the bed up, showed an impressively smooth sine-like curve, which we can use compensation to correct.

7) I have yet to test the results, but here's what I've calculated:  The smooth curve in the graph goes from a range of 0 to 0.1 and peaks at about the 9th measurement.  This means we want our compensation curve to have a total range of .1mm and be inverted so that it is lowest when our error is the highest.  Since the compensation curve ranges from -A to +A and has a low point at 270 degrees, then the compensation amplitude for me should be half this, or A = (0.1 - 0)/2 = 0.05mm, and the phase should be 270 - (360 * 9/14) = 38 degrees.

We'll see how it goes when I find some time to reflash and print, but found taking actual measurements really useful not only in determining my wobble compensation values, but also differentiating banding errors due to backlash from banding errors due to wobble.

I just finished my first attempts at using compensation, and I've got to say that I'm stunned how good my prints look now.

As I mentioned earlier, my prints were already very good to begin with, having only minimal wobble, even at .1mm.  By directly measuring the wobble and setting an exact compensation curve, however, I've been able to make it even better and all but eliminated it.  I think this would have been very difficult to  experimentally find a working sinusoidal curve since the wobble was hard to see to begin with, but with direct measurement it was very straightforward.

See the attached before and after photo, where I'm shining a flashlight very low on the surface to maximize what wobble there is.

The only hiccup I ran into was realizing that I had to fix any backlash in the Z-axis before the z-wobble fix could give good results.  Fortunately, using my laser pointer setup while moving the bed up and down made it easy to instantly know when I had sufficiently tightened the spring in my z-axis preload.

I've also been experimenting with a rocking laser pointer mount to make it easy to measure the wobble (without the mirror), and tweaking the firmware to make it easier to set the compensation values without having to do any math.  I'll work on these and post them up this weekend.

Had some issues after I moved my printer, I think my belts and z anti-backlash nut came loose and I had worse banding than I am used to.  So be sure to install a z-backlash solution and tighten those y belts.

I printed out all of Rincewind's gcode parts to see the full gamut of phases.  This zip file has high res images of each block.

Now those are even layer lines.

A0.01 W1.41 P330 for me w/o the ab solution on and looks like a diffrent printer.  This should be part of the all Marlin/Sailfish firmware...it intelligently solves the common issue of using indiscriminately cut cheap threaded rod. This not only helps with the obvious aesthetic issues, but will greatly increase part strength by remedying the root cause of some part failures; the inconsistently fused z layers. This z-axis banding issue is"Solved" now, and I bet SD starts using this before they ship, if not at least the process is easy, straightforward, and gives near perfect results.