why does adding backlash comp require a lower speed and acc?
skunkworks: emc's backlash compensation works as follows: when you reverse direction, move the motor an additional B units
you want to do that 'as fast as possible'
'as fast as possible' is defined as accelerating at the inifile maxaccel, up to maxvel
-- but that is above and beyond the acceleration already being used in the reversal, and above the velocity requested by the move
so while we usually say that emc "doesn't exceed machine constraints", it actually does when taking up backlash
ah - thanks :)
which should be OK in many circumstances, since theoretically you aren't pushing the table or cutter when doing backlash comp - you're only reversing the motor and screw
jmkasunich, do you think andy holcomb's problem could be a problem with the stepgen position loop going one count too far or something?
(seems like a longshot, and everyone and their brother would have seen it if so)
it's a bit ugly looking really
"normal" stepper system with backlash?
yes, all values from commanded
inifile maxvel is 1.0, maxaccel is 28.0
wow, that's quite a pseudomachine you've got there ;)
it's my real machine
that's very high accel, I think
but then again, I don't really know
I don't think that's atypical for itty bitty stepper machines
looks like the old 3-axis "max" has .5667 and 20
I wonder how many steps stepgen might output between "capture-position" and "update-freq"
funny considering that I have smaller motors running at a lower voltage than chris
SWPadnos: if andy's report is accurate, though, it's position-cmd that is changing
and I assume he's in "machine on", otherwise no steps should be commanded
backlash in part of position-cmd though, isn't it?
yes I suppose so
so stepgen is going full tilt one way, taking up backlash
fwiw I haven't seen this on zenbot
on a slow-ish CPU, motion may take a little while to execute, so you get one extra step, and now backlash goes full tilt the other way
its scale is 32000
unfortunately, that's the question Andy didn't answer ;)
I wonder if it's possible to use the BACKLASH setting as something like DEADBAND
ie, if you're within the backlash, don't bother moving the motor
my axis.0.motor-pos-cmd is stable at -1.720783e-05 for as long as I care to watch
what CPU speed?
or maybe more importantly, what's tmax on the motion functions
I think you're wrong talking about stepgen -- he says the commanded position is changing, and that's before the position gets to stepgen
yes, but backlash calcs are dependent on the feedback from stepgen
hm I didn't realize that's what you were getting at -- I'm not sure whether that's true or not
I assumed that "reversal" was commanded reversal, not feedback reversal
he says the commanded position is changing three or four digits down
stepgens synthesized position feedback is what motion uses to decide when to apply backlash comp
meanwhile he also said that the axis (its angular) moves something like 20 degrees
(if it needs to reverse to get back "on target"
SWPadnos: thats not true
comp is based on commanded direction and position only
and I don't think you want to treat backlash like deadband. On my backlash-happy mill I can still get good 5 mil features (like a trace narrowing to fit between two pads, then widening again) with 9mil backlash. if it was "like deadband", that wouldn't work anymore
jmkasunich: hi and good evening
jepler, ok, thanks - I wasn't sure it was a good idea
jmkasunich, hmmm - ok, right. there's no PID in motion
did he ever mention how he created this config?
don't think so
he's been dabbling (at least) for quite some time though, I think
I just read his latest - it seems he is saying that turning off backlash comp solved his problem?
seems that way
I wish people would speak precisely
what is the problem exactly?
just what I was thinking.
From: andyholcomb <email@example.com>
Subject: Re: [Emc-users] Problems with axises in auto and mdi
some axes wiggle back and forth sometimes, but not in manual mode
"then I do what I do to make it go crazy"
In manual mode it works great, MDI and Auto mode it will
randomly go nuts on on one or two varying axises. Nuts, sometimes
going back and forth about 10 or 20 degrees on a axis with no movement
on the digits on the screen. Or, moving slowly in one direction with no
movements on the location display of the screen.
^^ this thread, I won't mislead you with an attempt to summarise
since he talks about degrees, and says "on a axis" I read that to mean on his angular A axis
but later he says 'in the 4th or 5th digit'
that's not 10-20 degrees
but it probably means "my X or Y or Z linear axis motor shaft turns by 20-30 degrees"
and he says the dir signal is changing but step "is fine"
exactly - he doesn't speak precisely, and his reports are self contradictory
so I ignored him
I'll ask him to post his config files again
that's frustrating; I like to help but I despise fishing for information
4 mil backlash on 20TPI is abuot 30 degrees. naive cam which changes an axis by a tiny bit in every move? that could give "crazy" behavior like this.
if he has jittering on the last digit or two, EMC/stepgen might be going back and forth one step, but his drives could be missing the direction change so they'd walk one direction (at least for a while)
x.001 / x-.001 / repeat
jepler: so maybe it's when running a certain gcode
he seems to say it sits there and jitters at the MDI promot
he didn't say that the axes lose position
just that they wobble, and they stop when he goes to manual mode
does jogging apply backlash comp?
(I guess it should)
he didn't say they don't lose position
that's true, he didn't say either way
SWPadnos: yes it does
if the motor moves 20 degrees, and the feedback doesn't change (1st msg) or changes only in a very small digit (2nd) then yes, he lost position
I think he was talking about two different things tehre
the first was the DRO, which didn't change (it wouldn't, since it's only 3 or 4 digits)
the second was halscope/meter showing motor-pos-cmd and stepgen feedback
ok, so those two are consistent with each other
which did change in the 4/5/6 digit or something
not consistent with 20 degrees - his motors or drives must be misinterpreting the signals from stepgen, or else stepgen has a bug that nobody else has seen
motor-pos-cmd or joint-pos-cmd?
the DRO doesn't change when backlash is applied, does it?
depends on whether the DRO is in cmd or fb mode, and how well stepgen's fb tracks its command
this gcode "makes my X axis go crazy", because it's continually reversing and taking up backlash. http://pastebin.ca/1192152
it should not change (and won't, in cmd view, but in fb view it might)
jepler: I understand, but that seems to contradict what he is describing
the X DRO does change in the 4th digit while running this code
They jitter sometimes as soon as I go into mdi or auto mode, if I move x
in manual mode and then then go to mdi mode, the x axis may start to
jitter, it seems like it only jitters if the axis has been moved. I was
messing with it earlier I moved all of my axises in manual mode and them
went to mdi mode, the x and the z axis were jittering, I entered g00
x0y0 and the x and the y move as they should; but the z was still
jittering during the move.
hmmm. are there actually joint-pos-cmd outputs from motion?
I guess I should check the manual
crap -- I've misplaced all my broken drill bits
(they're what I use for touching off "Z")
I just asked andy to come to IRC - dunno if it will work
need to get back to cutting, but I'll check here once in a while
have I ever mentioned how much I hate "built up edge"
I guess I'll just have to break another
uh oh, he's using plu-toh
yay! config files
andy holcomb is using pluto?
no he's not
he just pasted a few extra files for some reason
* jmkasunich refrains from saying something unkind
these are not stepgen configs
er, this is not a stepgen config
no timings are set
his base period is small enough to get me a realtime delay message
it looks like a copy of our standard config
OK, I have the config running and went into mdi mode -- what am I supposed to do to see a problem?
or something like that
Y backlash .024, ouch
and 0.02 on X
I got a 'joint 0 following error' on a G0 move
hard to get 40ksteps/sec with 14000 base period
I bet stepgen accel headroom has to be much greater with backlash, since up to 2x accel is used while reversing
you'd need 12500 or less
SWPadnos: that too
so it sems likely that these aren't the correct files
it's running better with maxvel changed to 0.7 and STEPGEN_MAXACCEL changed to 45 in all instances
homed, G0'd X away from origin then back .. commanded position is stable
4 float OUT 8.673617e-18 axis.0.motor-pos-cmd ==> Xpos-cmd
what chat client is installed on ubuntu by default?
I use xchat, but I can't remember if I had to install it
hmm, andy replied to my message, but I didn't get it myself
I've had that happen, then the earlier message shows up sometime in the next day or so
EMC: 03alex_joni 07TRUNK * 10emc2/tcl/ (tkemc.tcl TkEmc):
EMC: Patch by Dewey Garett:
EMC: * add separate sliders for linear/angular joints:
EMC: uses DEFAULT_LINEAR_VELOCITY/DEFAULT_ANGULAR_VELOCITY if existant (or DEFAULT_VELOCITY if not)
EMC: * new key bindings (semicolon, apostrophe) for incrementing angular jog speed
EMC: * fix typo in spindle override slider that made it's popup behave different
EMC: 03tissf 07TRUNK * 10emc2/docs/src/ (5 files): french translation update
EMC: 03tissf 07TRUNK * 10emc2/docs/src/common/ (user_intro_fr.lyx integrator_intro_fr.lyx): french translation update
EMC: 03tissf 07TRUNK * 10emc2/docs/src/config/ (ini_homing_fr.lyx stepconf_fr.lyx): french translation update
EMC: 03tissf 07TRUNK * 10emc2/docs/src/install/installing_emc2_fr.lyx: french translation update
thanks for your continuing work on the docs!
jepler: np a pleasure for me
I have a recurrent problem with ini_homming_fr.lyx
All the characters are underlined in blue, I suspect that to be the error source to the compilation
But I do not see why!
could it be an HTML-style link problem?
(text getting displayed as a link)
all the texte is black but underlined in blue
a very clear message of the compiler!
LyX: Missing quote [around line 1838 of file /tmp/lyx_tmpdir153998srxmW/153996jfMWz]
and the first thing that it does... it eliminates this file!
well, it is a temp file ;)
the error's not temp!
if only we knew which file that one was generated from
the suspicious one is config/ini_homing_fr.lyx for me
HOME", c'est le point final de la séquence de prise d'origine.
La valeur par défaut est zéro.
I do not see any error except can be the format of the ""
who seems different of the other files
there's only one quote on that line, and no matching end quote (")
yes, I added the " lacking, that changes nothing apparently
tissf: I don't know why
no! missing "
If jepler does not know... that will be difficult
OK -- I think I figured it out
in that file there are many lines in the .lyx file which say "\lang french". I removed all those lines, and now the blue lines are gone
jepler oh ok can you commit it ?
I found a hint that it might be due to language when I searched for: lyx blue underline
yes I will check it in .. give me a moment
EMC: 03jepler 07TRUNK * 10emc2/docs/src/config/ini_homing_fr.lyx: fix all letters being underlined in blue when edited in lyx
can someone independently verify what I think is a bug in canned cycles?
"In incremental distance mode: when the XY-plane is selected, X, Y, and R numbers are treated as increments to the current position and Z as an increment from the Z-axis position before the move involving Z takes place"
cradek_ is now known as cradek
jepler: thank you
a sequence to trigger it is g90g0x0z1 / g90g17g81z-1r.1x.1l10
a sequence to trigger it is g90g0x0z1 / g91g17g81z-1r.1x.1l10
cradek: G90 is absolute mode?
I meant the second sequence above - the first line was a typo
I had to add an F-word too
so you expect the cycle to go down to z0 but it goes to z.1 instead?
are you also surprised that it moves to z1.1 before the first cycle?
or does it say it's supposed to do that?
"At the very beginning of the execution of any of the canned cycles, with the XY-plane selected, if the current Z position is below the R position, the Z-axis is traversed to the R position."
if "the move involving Z" is the traverse to R then it's wrong. If it's the feed down to the depth specified by Z then it's right. I am not sure the language is precise enough to say which one was intended
oh I see what you mean. I didn't consider "move involving Z" that way
how's that for refusing to "steak out"(sic) a position
yew guise arrr wired [sic]
well I didn't really want to be the last one to touch canned cycles anyway.
huh, I'd never thought about how for repeated canned cycles you have to move "one beyond" the first hole to start
yeah it's a bit odd
g90g0x0z1 / g91g17g81z-1r-.1x.1l10f20
that drills down to z-.1
is that different from what you expect?
cradek: I'm not sure
hmmm. does it leave the cutter at Z-0.1?
SWPadnos: no, Z0.9
cradek: so I am thinking it's a doc bug, and the docs should say something like "Z is an increment from the retract plane" which I think describes the current behavior
and is maybe even sensible
current position -> apply "R" (if specified) as increment to get retract plane -> apply "Z" as increment to get depth
I agree that's the behavior and the docs should be clearer, but I'm not sure the behavior is desired
I guess the question is, what numbers is it natural for the person coding the program to know
EMC: 03cradek 07TRUNK * 10emc2/src/emc/rs274ngc/interp_cycles.cc: I see no reason to forbid g95/fpr canned cycles.
EMC: 03cradek 07TRUNK * 10emc2/src/emc/usr_intf/axis/scripts/axis.py: on lathes, always draw dwells so they can be seen
EMC: 03cradek 07TRUNK * 10emc2/share/axis/tcl/axis.tcl: large DRO font option
EMC: 03cradek 07TRUNK * 10emc2/src/emc/usr_intf/axis/scripts/axis.py: large DRO font option
cradek: I can't wait to try that
don't tell jepler
he asked if I gave the two choices of font that will make all users happy - I said 'yes' with a straight face.
EMC: 03flo-h 07TRUNK * 10emc2/src/po/de_axis.po: updated some translations
EMC: 03tissf 07TRUNK * 10emc2/docs/src/ (8 files): french translation cleaning - fix lyx errors - new filenames for docclean
EMC: 03tissf 07TRUNK * 10emc2/docs/src/common/ (5 files): french translation cleaning - fix lyx errors - new filenames for docclean
EMC: 03tissf 07TRUNK * 10emc2/docs/src/config/ (6 files): french translation cleaning - fix lyx errors - new filenames for docclean
EMC: 03tissf 07TRUNK * 10emc2/docs/src/gcode/ (4 files): french translation cleaning - fix lyx errors - new filenames for docclean
EMC: 03tissf 07TRUNK * 10emc2/docs/src/gui/axis_fr.lyx: french translation cleaning - fix lyx errors - new filenames for docclean
EMC: 03tissf 07TRUNK * 10emc2/docs/src/hal/ (8 files): french translation cleaning - fix lyx errors - new filenames for docclean
EMC: 03tissf 07TRUNK * 10emc2/docs/src/install/ (compiling_emc2_fr.lyx installing_emc2_fr.lyx): french translation cleaning - fix lyx errors - new filenames for docclean
EMC: 03tissf 07TRUNK * 10emc2/docs/src/ladder/ladder_intro_fr.lyx: french translation cleaning - fix lyx errors - new filenames for docclean
EMC: 03tissf 07TRUNK * 10emc2/docs/src/quickstart/stepper_quickstart_fr.lyx: french translation cleaning - fix lyx errors - new filenames for docclean
cradek: looks like your font change will leak display lists when repeatedly changed
cradek: this probably causes x server memory to grow and not be reclaimed until axis exits, or maybe makes axis fail with an opengl error after awhile
sure - rain on his parade. ;)
EMC: 03jepler 07TRUNK * 10emc2/src/emc/usr_intf/axis/scripts/axis.py: do not leak display lists when switching font
with many cooks, all broth is shallow
er .. something like that
good night all
many cooks make light broth?
what is it called - step/dir you get individual steps instead of a servo where you get a smooth velocity movement
wait - did that make sense?
is quantization the word you want?
cradek: yes - thanks
[23:11:24] <LawrenceG> http://imagebin.ca/view/u11asXn.html http://imagebin.ca/view/3iFDcau.html
cradek: free shipping code for enco (supposidly) WB8SP
ordered already, thanks :-)
skunkworks: wait a couple days into the month, then search on any old machining web bbs
this was on cnczone ;)