* cradek grumbles bitterly
yes, it's been bad all day, it seems better now
raining there too, eh?
nope, not even that
I'm trying to figure a minimum effort, minimum probability of error approach to collecting screw comp data
borrow a glass scale from somewhere?
no idea where, and that would need interfacing and such
I have a 0.200 x 0.0001 indicator, and a set of gage blocks
I want to record data sets consisting of coordinate (as seen by emc), direction of approach, indicator reading, and gage block stack
a couple hundred of those
mdi'ing all the moves gets old fast
as would shuffling the blocks
how much travel are you talking?
3.3" for Z
(which is what I'm doing first)(
thats only 15 or so block combinations
do you have largeish blocks?
I guess sets have a couple of 1"
yeah, its a semi-binary approach
.1 .2 .4 .5 1.0 2.0 4.0
oops, .3 in there too
doesn't sound too bad
the real problem is not making a mistake
I was thinking in terms of a g-code program with o loops that would move the machine, pause while I read (M0), then move again
a second person would probably help (doublecheck the readings and do the recording)
and at the right time would tell me to replace the blocks
that seems quite possible
I know its possible
but my brain is not cooperating
I wish O...repeat was in 2.2
how many readings do you want?
EMC supports 256
you can do .190 of them at a time
I was thinking something like every 0.030 inches
that would give me 110 over the range, and 6ish readings per turn of the screw (0.200 pitch)
so only 6 readings per stack
12, I'd do up and down without changing the stack
I bet G91 relative mode is your friend
that doesn't change the coordinate display does it?
so if I set G91 I can just do Z-0.030, and Z+0.030
up arrow to repeat on or the other as desired
read the display, write down that value and the indicator reading
maybe you should build one of those 6-axis block stacker robots first
I doubt it could wring gage blocks
it could surely drop them on the concrete floor though
what I really want is a touch probe
although that would require a lot more gage block stacking, since I'd need one stack per point
more accurate though - using the 0.200 range of the indicator also incorporates any indicator error over that range into the readings
what can you stack them on that's stable and (most importantly) perpendicular to travel?
I have a very smoothly ground parallel clamped to the table
you could model the error functions in the screw and only measure at the peaks of cyclical errors
does it indicate flat with your .0001?
that wouldn't tell me anything
I was just thinking that
it would tell you the systematic errors
er, non-periodic errors
the indicator reads vertical travel
measuring the change in the parallel as Y or X changes doesn't tell me anything about Z
Z could be 3 degrees off vertical and it wouldn't show
you could swing the indicator like tramming
it would only tell me if the top of the parallel is parallel to X
cradek: good point
to be honest, I wasn't even worrying about perpendicularity error
that would be a linear term, cosine error
better to check it before you make a lot of bad measurements
even a 1 degree error (0.017" per inch off vertical) would only cause a cosine error of 0.0002 per inch
I can measure the off-vertical at any time, and correct the data
(the data will be going into a spreadsheet before I make the table - I want to plot the errors, see periodic vs random, etc
I have the feeling screw comp is about to be tested
although I certainly hope it has already been tested
hmmmm hmm hmmm hmmmm I think my mom is calling
whats she calling you?
the real test will be when I get to the Y axis
Z is a preloaded ballscrew, I expect little or no lash
Y is acme, it it will be compensating for both error and lash
that will be very interesting
jepler: i been playing with halui, and it draws Signal widgets in different position each time.
That seems impossible unless i cant recognize some random() like call.
took a couple hours, but I got all the data
an O-word program helped a lot
moved the indicator, and told me when to change blocks
cradek: yay, this sounds like they finally did something about udev: https://mail.rtai.org/pipermail/rtai/2007-December/018359.html
anyone know what tomp2 meant?
hmm.. I think he meant halgui
do you understand michael's report?
I think he's talking about a bug with run-from-line
(I thought you fixed those)
when I.m selecting vector 25 and resume from this line, EMC move to end
of 24 vector and then execute 25.
^^ but I don't think it ever worked this way
oh -- I don't care about halgui.
and maybe we should just take run from line off the axis menu
cradek: I didn't understand that report
I think he is saying that 2.1 used to work by moving to the end of the previous line when you do run from line (I think this is wrong), but 2.2 gives a following error when you use run from line
(and I haven't tried it)
odd - I left my machine on overnight (running EMC, but idle), and this morning I have the "unexpected realtime delay" popup
if it's still running, you could see how many you have with that hal param
sleepy computer - it took several seconds just to bring up a shell
only one I saw that seemed remotely related was motion.servo-overruns, and it was zero
yeah that's the one
did your message say 'on task N'?
yes, N = 1
hm, that's the "other" error then
jepler added it recently. it comes from rtai instead of timing tests
it was a handy message before I added rtai_smi to the list of modules to be loaded, because it would remind me to load that
I think this is the first time I've left the shoptask powered overnight
I don't think I'm going to sweat it too much
I have a few hundred numbers to type into a spreadsheet
that sounds so exciting
yeah, I can't wait
thats why I'm playing a game right now instead
actually I am kind of eager to graph the results and see what it looks like
just do it, I guess
are your screw pitch and indicator length different so you might see which one causes a particular periodicity?
yes and no
no, the indicator and screw are both 0.200 pitch
but yes, cause the program sometimes added 0.100 of gage blocks to the stack and sometimes added 0.200
(depending on where the readings were in the indicator range)
I told it not to make readings within 0.010 of either end of the indicator, so it couldn't always add 0.200 to the stack and still be in-range
I'm anxious to see the results now too
type type type
my portable does throw overuns as the screen goes blank.. (smi also)
I never looked into it.. as the screen starts to fade out as the latency test is running - you see the overuns countin up
data entry done, now for some formulas
skunkworks: I think it's common for the 3d screensaver to cause problems
I usually disable it
Hi alex. (don't know if I have disabled it yet)
I hate to say this, but open offices spreadsheet carts sucketh
something as fundamental as "use this column for X and that one for Y" seems to be impossible
gawd, the interface designers should be shot
cosmetics over everything seems to be their motto
click on something like an axis, and you get all the useless cosemetic crap - you can change its color, etc
but can you change actual _axis_ stuff like scaling? nooooooo
you have to jump through fricking hoops to do that
well hidden hoops
jmkasunich: I don't think it's OO's problem
they amde it as close to M$ office as possible
(people are actually used to jump through those hoops you mention)
in excel, you right click on an axis and you get the axis menu
what was your last excel version :?
I think work is running 2003
I use excel a LOT at work
well.. I guess you're right
I just tried it now.. and it does suck
you could use gnuplot, which sucks differently
do the math in a spreadsheet and export the data for plotting
yes, that would suck in a whole new way
I'm gradually finding out where the hoops are hidden, but why did they feel the need to reinvent things
what kind of graph are you after?
(I think what you're after is use the first column as label..)
X Y plot - one column is X, two more columns are Y
excel calls it a scatter plot, OO calls it XY plot
I have the graph type I want
ah, ok.. XY then
and am slowly beating it into submission
but its making me crazy
in excel, you can right click the graph, and one of the items in that menu is "source data"
I would have guessed a regular graph with the first column as the X axis, and the other 2 as the values
in "source data" you can add new lines, delete lines, and for each line you can specify the range with the X coords, the range with the Y coords, and the title that appears in the legend
that entire functionality is either missing or deeply buried
I noticed that you can double click the center of the graph
OO requires you to select all the regions when you start to create the graph, and if you want one line to use a different set of cells for X as another line you are screwed
then you have additional menu items
titles, axis, grid, etc?
yeah.. unusefull stuff
title legend axis grid
err.. not usefull
but you can modify data range if you right click the graph initially
(doesn't have series like excel does..)
or at least I don'
t see them
yeah, that is what I'm missing
its pathetic really, both excel and oo have about 20 graph types, and the only one I ever use is scatter (or XY), because its the only one that is real
real = you specify two sets of numbers that define points
and it plots the points
bar graphs and all that crap are for powerpoint
seriously, maybe you should learn gnuplot if you want to do this all the time
it has lots of 'real' stuff
but that kills the connection to the data
when I'm doing spreadsheet work, I do things like plot the results of a bunch of formulas, then change a cell that contains a formula parameter and look at the graph to see if that improves the curve fit
ah, sounds like college :-)
sounds like engineering
chemistry lab results
nah, you fudge that data
yield was ... 98%, wow!
wah, I need to clean a volume control but it's so tedious to get to it
I find myself doing things like trying to come up with a model that takes operating current, bus voltage, PWM frequency, and modulation index, and produces a result that matches junction temperatures as measured with a thermal camera
at least it updates the graph when you change values :D
sounds interesting, your job is not quite like mine
unfortunately I've also found myself collecting data to prove that the $20000 thermal camera doesn't meet the vendors specs
+/- 2 deg C, right.... maybe... in the very center of the screen... on a good day...
[18:09:06] <jmkasunich> http://jmkasunich.com/pics/quill-screw-error-1.png
blue/red are the two directions?
I can't figure out how to change the damned legend
that graph gives me the impression you didn't "get" the exact tpi you thought
is it really off the total .002-3?
pretty close to 0.002 per 3.000" slope
that would be 0.008 per foot, which is within spec for rolled ballscrews, ground ones are usually better
no clue what I have, its surplus
interesting that .2in is quite close to 5mm
but not this close
well.. you can change the scale, or not bother and leave it like this :)
there is also some 0.2" periodic error
yes but it's not always there (not at -2.0)
so I wonder if it's the indicator instead
not at -2.6 either
its also quite spiky, I was hoping that something like bearing runout would be a 0.2" period sinewave
I wonder if the spikyness is real
I'm sure there is some randomness to the data
but there are many places where red and blue spikes match up
jmkasunich: next you need to do it again, using other combinations of blocks
and see if they match :)
and a different indicator
other combinations, in order to have the indicator in a different working pos.
I am considering doing about 10 samples over the 0.2" range of the indicator, where I don't move the machine at all, and change the blocks instead
that would spot any major indicator errors
(replacing them with gage block errors)
is your indicator cheap or expensive?
hmm, probably the same one I have
I wouldn't expect .0007 noise from a starrett .0001 indicator
some random noise is going to come from the balls rolling in the ballscrew itself, and in the ball bearing supporting the ballnut
neither was new
finding these results would make me happy - the slope is very real (and easy to fix) but the rest doesn't need correction IMO
doesn't need = not useful to spend time trying
the linear error could be screw pitch error, or it could be cosine error if the axis isn't perfectly parallel with the gage block stack and/or the indicator plunger
hm, the heater in the shop does run sometimes even with the thermostat turned all the way down (~ 42)
I wonder if that's a feature
not for your electric bill
I bet it doesn't run much, wish I knew how much it is
how is it controlled? thermostat switches 24V which pulls in a contactor?
any way you could hook up a mechanical electric clock so it runs when the heater does?
normal home thermostat
that's a great idea
but, everything is 220, might be hard to find a clock here
the heater is 240V though
can you get into the heater?
connect the clock from one pole to neutral?
I ran neutral to it, if the contactor switches both sides it would still be ok
I was assuming it would switch both, but that may not be valid
yeah, seems like it should
I'd have to look
it's fairly accessible
the other problem might be finding a mechanical electric clock around
yeah, there is that.
I don't think I have one here - most are digital line powered, or quartz battery powered
I have one right here but I'd hate for it to get broken (glass front etc)
do you have an hour meter tucked away in your junk bins anywhere
I'll be able to find something
no but I'm sure surplus center has them
well, if you're gonna buy something, a clock is easier and cheaper
go to a thrift store or something
doubt there is a real mechanical electric clock to be purchased anywhere in the city
or have quartz battery ones taken over
gotta buy those batteries you know
and who can read those non-digital clocks anyway?
(now I'm trolling a bit)
just make a 24V AC to 1.5V dc power supply, and use a quartz clock (one with hands, that won't care about power interruptions)
is there a blower on the heater?
put a pinwheel in the airstream and connect to a 1.5V generator
thanks mr. goldberg
always happy to help
* cradek is up to his knees in LISP
on a serious note - anything "connected" by way of either air flow or air temp is inherently isolated
you could hook it to a parport safely, and let hal accumulate the on time
temp switch that is turned on by hot air?
not sure. I think the delays after element on and off are strictly time
a clock and two gator clips is definitely the simple way to do this test
unless you don't have a clock
surely I can find one here somewhere.
if not you need to change your domain
well, after subtracting out the 0.00067" per inch linear ramp, there is about 0.0002 of a broad U in the data
along with the +/- 0.0004 high frequency stiff
[19:35:36] <jmkasunich> http://jmkasunich.com/pics/quill-screw-periodic-error.png
no simple once-per-rev pattern
but there is a pattern
I'm not gonna sweat it
change scale and forget about it
I just have to figure out whether I should be making scale larger or smaller
looks like everything is mostly within 3 tenths - that seems pretty great to me
yeah, except it didn't actually work out that nice when I tested it
I changed the scale, and then measured at 0.280, 1.280, 2.280, and 3.280
got 0.0004, 0.0007, 0.0004 and 0.0000 (I set the indicator zero at the last point)
those are rough numbers, I didn't actually write them down
I probably should take 5-10 points each way again, just to see what I have now
I know I'm not gonna correct periodic errors, so I don't need to take anywhere near as many
maybe I should take data every 0.2
that would cancel out screw periodic error and indicator error, leaving only gage block error and cosine error
and cosine error should be linear