search

Frix-x / klippain-shaketune

Klipper streamlined input shaper workflow and calibration tools
GNU General Public License v3.0
568 stars 62 forks source link

AXES_SHAPER_CALIBRATE gives different results than SHAPER_CALIBRATE #139

Closed celtare21 closed 6 days ago

celtare21 commented 1 week ago

K-Shake&Tune module branch

Version

v4.1.0-1-g66f5e32

Describe the bug and expected behavior

The results given by AXES_SHAPER_CALIBRATE from shaketune are quite different than the results given by klipper's SHAPER_CALIBRATE. Don't know if it's also an issue, but the given dampening_ratio values are also quite different than the ones given by ResHelper.

SHAPER_CALIBRATE results: shaper_calibrate_x shaper_calibrate_y

AXES_SHAPER_CALIBRATE results: inputshaper_20240702_213706_X inputshaper_20240702_213706_Y

Also the calculated dampening ratios from ResHelper are: damping_ratio_x: 0.0085 damping_ratio_y: 0.0388

Additional information and klippy.log

No response

Frix-x commented 6 days ago

Hi, thanks for reaching out and creating a topic about this. It is indeed something I need to add to my documentation: that's normal.

The explanation for the different filter suggestions is that the default Klipper scripts use forced values (0.1 and 5) for the damping ratio and square corner velocity (ResHelper does the same, AFAIK), even if they don't correspond to measured values on your machine.

But in Shake&Tune, since I retrieve the configured SCV and calculate the damping ratio from the graphs, I have the ability to inject them into the filter calculation algorithm. Using the real printer values has the effect of producing filter results that can be a bit different from the original scripts, especially if your real values are anything other than 0.1 and 5, as it's most of the time the case with Voron and other printers. The forced values are more suitable for Ender3, etc.. But the new computed filters done that way are also more representative of what really happens in your machine and S&T should end up being more accurate than the other tools.

Regarding the damping ratio, the number I give in S&T is an estimate. It's actually pretty hard to get a measurement for this based solely on a PSD graph, and that only works correctly if there is a main peak well separated from anything around it (and that's for example not the case in your X-axis graph), so the value is probably a bit eronned. But this is also true for ResHelper, and since it uses a different algorithm than I do, it's also giving you different values than in S&T (but it doesn't mean that one is better than the other, they're both estimates). FYI, the algorithm I'm using for the damping ratio is based on a NASA paper and is called the Half-Quadratic Gain method. This method should be more accurate than the Half-Power method usually used by most tools, especially for higher damping ratio values (above 0.05) where we are most of the time.

celtare21 commented 6 days ago

Alright, thank you for the explanation!