#linuxcnc Logs

Apr 24 2024

12:19 AM jpa-: rigid: i used plain sensor - but it was some year 2010 model
12:32 AM jpa-: rigid: do you have any result data? for example plots of travel in opposite directions so that you get roundtrip error?
01:04 AM Deejay: moin
01:09 AM solarwind: Deejay I will gouge out your motherfucking eyes and I will fuck you like a pig
01:09 AM solarwind: JK JK JK
01:09 AM solarwind: it's a quote from Whiplash
01:09 AM solarwind: excellent movie, watch if you haven't
01:10 AM Deejay: ah
01:29 AM rigid: jpa-: I have a script that takes 512 measurements along the complete travel distance (256 forward and reverse). I found that traveling forth and back includes backlash and thus isn't useful to measure measuring errors for the sensor.
01:30 AM rigid: without another way of measuring absolute distance like a glass scale, it's hard to come up with good data. Hence I asked. It'd be nice to compare that with something that has actually <5um error.
01:40 AM jpa-: rigid: well, you could do forward-backward-forward and then you can measure roundtrip without backlash
01:41 AM jpa-: then divide by 3 and you have approximate accumulative error over one travel distance
01:52 AM rigid: how would the backlash be compensated then? the travel-backwards distance will be shorter than travel-forward since it's just measuring relative motion
01:52 AM rigid: I could run it 100 times to minimize backlash, but the distance-delta caused by it would still be included if i'm getting that right
01:54 AM rigid: i calculated error by dragging the mouse 5m along a railguide, dividing the result by dpi and then converting it to inch.
01:55 AM rigid: actually I also calculated total distance travelled from X and Y axis since it's very hard to precisely align the sensor parallel to the travel axis.
01:56 AM jpa-: hmm, maybe i don't understand what you are doing
01:56 AM jpa-: i would expect you get the total traveled distance from integrating the movement values from the mouse
01:57 AM rigid: yes
01:57 AM jpa-: because the screw compensation is the total difference, not relative difference
01:57 AM rigid: there are two file formats iirc, one relative and one total
01:57 AM jpa-: so if you travel forward-backward-forward, you should end up at exactly same position, regardless of screw accuracy
01:58 AM jpa-: and then you can check how much error has accumulated in the mouse reading
01:58 AM rigid: you end up at the same position but since you don't measure distance to a fixed point, the measurement will be off and accumulating
01:59 AM rigid: yeah that would work if the machine would already be calibrated and compensating during measurements, i guess
01:59 AM jpa-: the mouse error will accumulate, but that's what the mouse sensor does, and exactly why i think it will not be accurate enough for this
02:00 AM jpa-: but you can do the same over smaller distance if you are only doing local pitch compensation; that will give the error you accumulate over that small measurement distance
02:00 AM jpa-: and no, the machine doesn't need to be calibrated for this
02:01 AM rigid: I really don't get it, if I tell the machine to travel from 0 to 10mm but it ends up at 9mm i'll measrure 9mm+/- sensor error. How will I ever know it's 9mm and not 8mm?
02:02 AM rigid: I need to measure error with known, precise distance (5m rail in my case)
02:02 AM jpa-: if you travel 0 -> 10 -> 0 -> 10, the first 10 and second 10 will be same position and the difference in mouse sensor readings will tell how much the mouse sensor drifted during move
02:03 AM jpa-: after that you know how much the mouse sensor drifts and can consider whether it is useful tool for measuring the accuracy of the movement
02:03 AM rigid: in my example you'll measure 0 -> 9 -> 0 -> 9 and the 9 is unknown. how do I caluclate the error from my measurement x that I'll get?
02:04 AM jpa-: you subtract mouse reading at second 9 from the mouse reading at first 9
02:04 AM jpa-: it doesn't give you the compensation, it is just a test to see if your mouse sensor is accurate enough
02:05 AM rigid: ok, I just get some value then but can't calculate an actual % because I don't know the actual distance travel (effective 9 is unknown, only nominal 10 is known)
02:06 AM jpa-: it will be pretty close
02:06 AM jpa-: even ancient leadscrews are +-1%, so your percentage value error would be that much; and do you really care to know if the mouse sensor error is 0.1% or 0.101%?
02:07 AM rigid: i think traveling a known distance multiple times is just fine to calc sensor error
02:07 AM jpa-: only if you actually have a known distance
02:07 AM jpa-: but what are your results for that?
02:08 AM rigid: I did that with my old 1800dpi mouse and got 0.01% error
02:08 AM jpa-: (the variance)
02:08 AM rigid: (very slow constant movement)
02:09 AM jpa-: somehow that is at least 10x better than what i got, but maybe the sensors have improved
02:09 AM jpa-: can you post actual data?
02:11 AM rigid: yeah
02:11 AM rigid: that's the sensor i'm currently waiting to arrive: https://www.pixart.com/products-detail/129/PAW3395DM-T6QU
02:12 AM rigid: the datasheet says 0.4% resolution error
02:13 AM jpa-: sensor resolution error doesn't really matter much as that you can easily calibrate/average out
02:13 AM rigid: yeah
02:14 AM rigid: and the high resolution gives a lot of elbow room for errors. (my machine has just 30cm at longest axis)
02:15 AM jpa-: i wonder what is the unit for the "path error" graph in the datasheet
02:27 AM rigid: which graph? did you find a datasheet beyond that website?
02:35 AM jpa-: alldatasheet had some though looks like not exactly that model
03:48 AM -!- #linuxcnc mode set to +v by ChanServ