Results 1 to 10 of 25
Threaded View
-
12-04-2016, 07:49 PM #13
- Join Date
- Jul 2016
- Location
- Pennsylvania, USA
- Posts
- 255
Hi
So as I understand the thread over on git hub, when the steps go out past 2 ms, bad things happen in terms of catching the 5 ms wide pulse. The MakerFarm's are 4000 steps per mm on the Z axis. At 500 steps / second that would be 1/8 mm per second or roughly 5 mm / minute. My default Z feed is 2 mm / s so it's in the "ok" region. My data before was a default / 10 so 0.2 mm/s. That seemed to work fine on faults. It was junk on spread and one sigma.
Using the above data, to be out around 0.05 mm at 0.2 mm/s your timing would have to be off by 250 ms. If things are sampled at faster than 2 ms, that's a lot of time.
I kinda doubt my one sigma is from timing issues / the machine running to fast. You could trigger the data with a relay and do better than what I'm seeing.
====edit
If you watch the probe long enough as it cycles, you can catch "late" pickup events. The LED "blinks" slightly as it activates. The interesting point is that they do *not* correlate to min or max reading events. The data is still in there with the rest of it.
Contrary to all prior experience, the problem does *not* go away when DEBUG_ALL is invoked. Crazy stuff . So far, it's really hard to see how this is software related.
Off to butcher some poor helpless Marlin functions ....
==== edit
Put a median filter into the probe routine. Running with a fast probe speed (= default on the Pegasus) it improves things by ... surprise! ... square root of the samples (almost). On M48 P40 V4, the standard deviation goes from about 0.030 to about 0.015. With 5 samples that's almost right. There is something else in there other than just 1/F (flicker) noise.
BobLast edited by uncle_bob; 12-06-2016 at 06:40 AM.
Qidi X Plus 3 Paper thin first...
05-27-2024, 01:15 AM in General 3D Printing Discussion