#linuxcnc-devel | Logs for 2015-07-25

[08:58:46] <jepler> .. 3-card 2kHz test terminated due to power outage
[09:50:22] <mozmck> jepler: can you come up with a software fix for that?
[10:39:09] <jepler> software power generation is a great idea.
[10:39:40] <Tom_itx> battery backup/generator
[10:39:46] <Tom_itx> with autoswitch
[10:50:24] <jepler> oomlout.co.uk/blogs/news/38808897-turning-the-lathe-on-its-head-a-cylindrical-3d-printer
[11:19:00] <kwallace> Does anyone know how well a drag knife works on gaskets? Page 7 here: http://www.tormach.com/uploads/530/TD10093_CNC_Drag_Knife_0312A-pdf.html
[11:21:14] <andypugh> kwallace: It will take many passes. It was 3 passes to get through sticker-paper
[11:21:52] <archivist> that blade has a 3 thou cut, and gaskets are a lot thicker
[11:22:13] <andypugh> Though that might be because my plotter had no pressure, and a CNC machine could probably puch harder
[11:24:30] <kwallace> Gaskets seemed a bit of a stretch. Now I'm thinking maybe a punch attachment. The table holds the sheet and moves it under a 2.5mm punch.
[11:24:33] <archivist> from doing it by hand the lack of a sticky backing like sign material you mean it is hard to clamp and keep still
[11:25:11] <archivist> I think a punch is more sensible too
[11:25:31] <archivist> and time there was a linuxcnc punch
[11:26:37] <pcw_home> couple watt laser?
[11:26:46] <kwallace> http://www.vandf.co.uk/tooling/what-is-cnc-punching/
[11:27:15] <andypugh> I am pretty sure someone did a LinuxCNC punch
[11:27:50] <archivist> where what when, I missed it :(
[11:28:13] <andypugh> For fok who aven’t seen a CNC punch: https://www.youtube.com/watch?v=EU3SKaBFHIk
[11:28:51] <archivist> being in the same area as one is an experience :)
[11:33:08] <pcw_home> Wonder if one of those vibrating engravers would work on paper gaskets
[11:34:24] <kwallace> Andy's video link was great until it mentioned Windows.
[11:34:52] <andypugh> http://www.linuxcnc.org/index.php/english/forum/30-cnc-machines/736-twin-headed-hydro-punch-how-to-cnc?limitstart=0
[11:35:47] <andypugh> http://www.linuxcnc.org/index.php/english/forum/38-general-linuxcnc-questions/6264-punching-machine?limitstart=0
[11:36:01] <archivist> I remember the beam fabricator now
[11:36:41] <andypugh> http://www.linuxcnc.org/index.php/english/forum/30-cnc-machines/27285-punching-press-cnc#41696
[11:38:31] <archivist> I wonder how he got on
[11:40:01] <andypugh> http://www.linuxcnc.org/index.php/english/forum/38-general-linuxcnc-questions/6264-punching-machine?start=20#7102 suggest she got there, but the video has gone
[11:43:16] <archivist> the beam punching(and drilling?) guy came into irc a few times
[11:45:29] <mozmck> http://www.linuxcnc.org/index.php/english/forum/18-computer/29453-dedicated-motor-controller-plus-raspi-vs-desktop#60912
[11:46:27] <mozmck> I think if he wants feel good about the computer responsible for moving a sharp spinning thing, he probably won't use a Pi
[11:50:21] <archivist> is there enough memory in a pi
[11:50:53] <andypugh> Yes, but the latency isn’t great
[11:51:20] <andypugh> 100uS with PREMPT-RT and the Ethernet port is a USB device
[11:51:31] <archivist> ew
[11:51:50] <andypugh> Xenomai might work.
[11:53:02] <archivist> usb in series with ethernet just feels dumb and wrong
[12:02:17] <pcw_home> Yeah in my (limited) experience USB Ethernet will have multi-millisecond delays occasionally
[12:05:01] <pcw_home> 100 usec latency is probably OK for a 1 KHz servo thread, but in my experience,
[12:05:02] <pcw_home> real time Ethernet under Preempt-RT works better when you have more
[12:05:04] <pcw_home> horsepower than the RPI/RPI2 provide
[12:10:00] <kwallace> How can I power one of these? : http://grizzly.com/products/Heavy-Duty-Hand-Punch-w-Stand/H7862
[12:13:45] <Tom_itx> small cylinder
[12:14:28] <Tom_itx> or largeish solonoid
[12:15:03] <mozmck> RPI/2 is ARM too, so it may still have lingering issues? I guess the memory read ordering bug is fixed now?
[12:16:12] <kwallace> Tom_itx, I'm tending toward a stepper and short beefy ball screw to leverage existing hardware.
[12:20:17] <Roguish> kwallace: air cylinder. watch your fingers....
[12:22:21] <andypugh> Air-over-oil cylinder
[12:22:31] <archivist> latched flywheel, watch everything and have interlocks on the guards
[12:22:32] <andypugh> But it is too cheap.
[12:23:51] <archivist> I used to use one of the hand presses that one looks copied from, you can see the frame flex a little
[12:25:44] <archivist> the whitney http://www.knighton-tools.co.uk/acatalog/Punching_Machines.html?sa=X&ved=0CBYQ9QEwAGoVChMI75zHneH2xgIVS1YUCh29eQPx
[12:26:50] <andypugh> $60 with 6 punches hints to me that the punches are not very good.
[12:26:56] <andypugh> http://uk.rs-online.com/web/c/tools/machine-tools-accessories/punch-die-combinations/?applied-dimensions=4294824163&esid=cl_4294967294,cl_4294958554,cl_4294957229&m=1
[12:28:29] <andypugh> Though admittedly the RS bench press is ludicrously costly: http://uk.rs-online.com/web/p/machine-presses/0602432/
[12:28:47] <archivist> we later had an RS lever punch had a flame cut frame
[12:29:40] <andypugh> I used an RS press a lot. They definitely work well enough.
[12:30:02] <andypugh> (University, press and every possible tool. Never used until I found it)
[12:38:16] <jepler> mozmck: no, we haven't merged any patches for the ARM memory ordering bugs and the initial set of patches only cover one of several areas where we imagine that there are memory order bugs lurking
[12:45:17] <pcw_home> interesting thread on forum, someone converting a Novakon to Linuxcnc.
[12:45:18] <pcw_home> it uses a (step/dir) servo for the spindle, wonder how easy that is to make
[12:45:20] <pcw_home> work with rigid tapping
[12:46:17] <pcw_home> (without adding an encoder I mean)
[12:49:36] <jepler> probably at a minimum needs a custom component that takes a counts input, implements the index-enable protocol, and outputs a position-fb
[12:51:48] <pcw_home> I was thinking just scaling the stepgen position would be enough for the counts but it needs the index logic
[12:54:09] <pcw_home> the stepgen firmware has the position latch to implement index/probe so thats a possibility but needs driver code
[12:55:11] <jepler> oh you think there's a physical index on the otherwise encoderless spindle?
[13:00:12] <pcw_home> Maybe not, if not I guess a modN counter could be done (Either in firmware as part of the stepgen or software in the component)
[13:03:08] <pcw_home> a servo driven spindle allows you to slave the spindle to the motion (I think thats what they do with Mach)
[13:04:30] <cradek> if using software stepgen, you could just tie it to a software encoder
[13:04:44] <cradek> seems like with an fpga you could do the same
[13:04:57] <jepler> Subject: [ANNOUNCE] 4.1.3-rt3
[13:06:03] <pcw_home> Currently theres no simple way to wire things like stepgen outputs to encoder inputs in the FPGA
[13:06:37] <pcw_home> (I mean no non one-off hacky way)
[13:06:47] <cradek> heh I meant use wires
[13:07:21] <pcw_home> not enough wires on one parallel port connection
[13:07:30] <cradek> they'd have to agree on step type unfortunately
[13:07:52] <cradek> oh I forgot about connectors
[13:08:07] <pcw_home> yeah maybe easiest is to add index simulation to the stepgen module
[13:09:30] <cradek> no simple fix is there
[13:15:20] <pcw_home> probably the cleanest way would be for linuxcnc to support servo driven spindles, but I have no idea whats needed for that
[13:17:52] <cradek> but stepgen home to index would be nice for other reasons too
[13:18:26] <cradek> wait that's different I think
[13:18:37] <cradek> or is it
[13:27:42] <pcw_home> if you had a physical index input I think its the same
[13:40:14] <andypugh> You can emulate a counts reset with an integer mux2 wired back to itself.
[13:40:35] <andypugh> Then back to a gain -1 sum2.
[13:41:26] <KGB-linuxcnc> 03John Thornton 052.7 c36d828 06linuxcnc 10docs/src/common/user-intro.txt 10docs/src/getting-started/pncconf.txt Docs: add some info on starting linuxcnc. * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=c36d828
[13:42:16] <pcw_home> If it could be hacked together in HAL that would be great
[13:43:00] <skunkworks> would't that output have to be compatable with i/o then?
[13:44:03] <andypugh> Hmm, if you wanted a consistent index position then it gets a bit harder.
[13:44:33] <andypugh> Maybe there is a real need for a sim_index HAL component?
[13:45:39] <andypugh> endex-enable IO pin, and it looks for rawcounts modulus N being zero before resetting output and setting index-enable true.
[13:47:08] <andypugh> It starts to look like a bit of a mess when you add tristate latches and modulo-arithmetic to the HAL mess
[13:47:49] <andypugh> (In fact, if I was writing that index comp it would do modulo arithmetic as a byproduct, because I keep wanting that)
[14:02:09] <jepler> and make sure it works when you have >2^31 counts since poweron and the modulo is not itself a power of 2 (e.g., 360, 400)
[14:02:26] <jepler> (so internally you really need to extend to 2^64 counts and then do modulo arithmetic)
[14:07:30] <andypugh> Yes, I had already concluded that
[14:07:42] <linuxcnc-build_> build #2488 of 1301.rip-precise-rtai-i386 is complete: Failure [4failed compile runtests] Build details are at http://buildbot.linuxcnc.org/buildbot/builders/1301.rip-precise-rtai-i386/builds/2488 blamelist: John Thornton <bjt128@gmail.com>
[14:11:42] <pcw_home> Its would need to look for rawcounts mod N crossing 0 (since you may miss 0 itself)
[14:12:30] <archivist> and overflow
[14:14:19] <pcw_home> jepler: what clock speed is your Core duo test machine? (trying to figure what the added per card overhead is)
[14:20:00] <linuxcnc-build_> build #3295 of 0000.checkin is complete: Failure [4failed] Build details are at http://buildbot.linuxcnc.org/buildbot/builders/0000.checkin/builds/3295 blamelist: John Thornton <bjt128@gmail.com>
[14:20:19] <jepler> pcw_home:
[14:20:19] <jepler> model name : Intel(R) Core(TM)2 Duo CPU E8500 @ 3.16GHz
[14:20:21] <jepler> cpu MHz : 1998.000
[14:20:31] <jepler> so I think it's actually running at the lower speed, haven't sorted out why
[14:21:00] <jepler> hm eventually I had a blip at 2kHz
[14:21:05] <jepler> Unexpected realtime delay on task 0
[14:21:07] <jepler> and all the watchdogs bit
[14:21:17] <jepler> max-time jumped up to 63745712 !
[14:21:56] <pcw_home> oops, thats a rather long vacation
[14:28:46] <pcw_home> I have the same CPU
[14:28:48] <pcw_home> model name : Intel(R) Core(TM)2 Duo CPU E8500 @ 3.16GHz
[14:28:49] <pcw_home> cpu MHz : 3158.689
[14:32:43] <pcw_home> (HP DC7800) it runs one card at 3 KHz pretty reliably but rarely has ~300 usec delays
[14:32:45] <pcw_home> so I get errors at 3 KHz (about once a month) Of course I am using it as
[14:32:46] <pcw_home> a normal desktop at the same time
[14:32:48] <pcw_home> The H97/G3258 is completely reliable at 4 KHz (at least 6 months without incident)
[14:37:43] <pcw_home> the rare ~300 usec delays have been in the hm2_eth read time but also sometimes show up in the motion controller time
[14:41:46] <pcw_home> The H97/G3258 shows none of these rare large-ish delays
[14:42:38] <jepler> 99th percentile of 120000 2kHz servo thread iterations: (thread1.time @ 2GHz)
[14:42:41] <jepler> one card: 377568
[14:42:43] <jepler> two cards: 435033 (+75465)
[14:42:46] <jepler> three cards: 536446 (+83413)
[14:42:47] <jepler> of course that's just over 1 minute, not over a long period of time
[14:45:20] <pcw_home> so at 2 Ghz thats ~ 40 usec more per card
[14:47:40] <pcw_home> (probably not much data xfer for those added cards though )
[15:21:40] <jepler> yes, cards 2 and 3 are configured all gpio
[17:51:30] <jepler> aha I knew I had written a general "extend a small-width counter to 64 bits" function. it's hiding in pluto_common.h. http://paste.debian.net/286030/
[17:51:44] <jepler> no doubt there is some more elegant way to write it
[17:53:40] <jepler> and I wonder if the type of newlow should be unsigned
[18:10:58] <KGB-linuxcnc> 03John Thornton 052.7 4e20811 06linuxcnc 10docs/src/hal/parallel-port.txt 10docs/src/hal/rtcomps.txt Docs: figure links are no longer needed with asciidoc * 14http://git.linuxcnc.org/?p=linuxcnc.git;a=commitdiff;h=4e20811
[18:19:03] <andypugh> jepler: That’s interesting. I have written a few bits of code that extend to 64 bits, and they are nowhere near as long.
[18:24:41] <andypugh> All I have done is: http://paste.debian.net/286043/
[18:25:26] <andypugh> Which I have proven to work for every scenario that I have tested it with. But leaves me worried that there are scenarios that I have not tested with.
[18:25:42] <Roguish> Hey all: I get this error on the linuxcnc.org website: Internal Server Error
[18:26:29] <andypugh> So do I
[18:26:37] <andypugh> It was working earlier
[18:26:48] <Roguish> i agree.
[18:27:18] <Roguish> where is the server these days?
[18:29:06] <andypugh> I am not sure. We had a 403 problem a week or so ago, which was fixed.
[18:31:07] <andypugh> I have forgotten who it has that has admin access to the server.
[18:34:32] <Roguish> is it in a commercial farm, like Amazon? or on a personal server? does it need some funding? possibly a round of contributions?
[18:34:36] <linuxcnc-build_> build #3286 of 1200.rip-lucid-i386 is complete: Failure [4failed compile runtests] Build details are at http://buildbot.linuxcnc.org/buildbot/builders/1200.rip-lucid-i386/builds/3286 blamelist: John Thornton <bjt128@gmail.com>
[18:35:19] <andypugh> Roguish: I believe it is somewhere between the two.
[18:35:54] <Tom_itx> i wonder if it's ever backed up
[18:36:14] <andypugh> My impression is that it lives on a commercial server but we don’t pay
[18:36:40] <Roguish> even as squeaky as I am, I would be more than happy to contribute some $.
[18:42:11] <andypugh> I thought that is was Alex in Romania, but apparently not: http://whois.icann.org/en/lookup?name=www.linuxcnc.org
[18:44:37] <andypugh> Anyway, the server is now back up
[18:49:19] <linuxcnc-build_> build #3296 of 0000.checkin is complete: Failure [4failed] Build details are at http://buildbot.linuxcnc.org/buildbot/builders/0000.checkin/builds/3296 blamelist: John Thornton <bjt128@gmail.com>
[20:42:30] <cradek> it's at dreamhost
[20:42:51] <cradek> although I appreciate the sentiment, I don't think any of us involved with it need or want contributions
[22:57:37] <mozmck> If you want to disable soft limits, do you set the MIN_LIMIT and MAX_LIMIT to ridiculously large numbers - or both to zero?
[23:16:58] <Tom_itx> http://linuxcnc.org/docs/2.7/html/config/ini-config.html#_axis__lt_num_gt_section
[23:17:01] <Tom_itx> doesn't really say
[23:17:39] <Tom_itx> seems you would set them past the hard limits
[23:19:27] <Tom_itx> or as you said, a ridiculously large number