That looks like a good demo, Sam. I just added it to my dropbox folder of nc test files. As both you and Phill have identified, the issue here seems to be how short it's cutting the corner, even at relatively low speeds. Here's a screenshot of the same issue on my machine: https://dl.dropboxusercontent.com/u/10948059/linuxcnc-trajectory-tests/corner_cut_arcblends.png
This happens because there are no tolerances applied to the blend here, and arc blending is "greedy", in that it can consume essentially all of a segment to get the fastest speed. A tight corner leads to huge tolerance for a very small increase in blend radius. I just pushed a simple fix for this, which is to limit the blend tolerance to the parabolic equivalent. Even with no tolerance applied, the worst-case blend tolerance is still bounded by the size of the segments. The maximum tolerance for parabolic blending is effectively 1/4 of the segment length, at the extreme case when the corner is 0 deg. (i.e. reverse direction). This wasn't explicitly programmed since it happened naturally, but we don't have that luxury for arc blending. My fix just adds this as a maximum tolerance. Here's what it looks like now with parabolic blends: https://dl.dropboxusercontent.com/u/10948059/linuxcnc-trajectory-tests/tight%20corner%20arc%20blend.png Here's the same motion with arc blends now: https://dl.dropboxusercontent.com/u/10948059/linuxcnc-trajectory-tests/tight%20corner%20arc%20blend.png -Rob On Sun, Jan 5, 2014 at 1:46 PM, sam sokolik <sa...@empirescreen.com> wrote: > well I just ran the program - what seems to hapen is the angle between > the last 2 moves is very small. So it blends them quite early.. > If you say G64P.001 - it then runs the path pretty much the same speed - > makes it all the way to the corner (well within .001) > > http://imagebin.org/285301 You can see the blend between the 2 lines > right in the middle - it is odd though that it seems to be right at x0.. > maybe coincidence.. (that shows both runs - G64 and G64P.001) > > this really isn't a good test to see the performance improvement for the > new tp.. This would.. (and I think it should fit on your machine) The > new tp peaks (with my config) at 270ipm vs 60ipm on current TP. > > g20 g64 g90 > > g0 z.1 > g0 x1 y0 > g1 z-.1 f999 > > o100 repeat [370] > g91 g1 @-.0025 ^4.5 > o100 endrepeat > > g90 g0 z.1 > m2 > > sam > > > On 01/05/2014 11:32 AM, Robert Ellenberg wrote: > > Hi Phill, > > > > Those screenshots look the same, are they duplicates? In any case, the > fact > > that it skipped the parabolic blend entirely at that corner is not > supposed > > to happen. I'm going to run that test again on my machines to confirm. > What > > commit are you building on? I had to do a roll-back of some changes a few > > days ago, so that could be contributing to this issue. > > > > -Rob > > > > > > On Sun, Jan 5, 2014 at 2:18 AM, phill carter <phill.car...@oztime.com.au > >wrote: > > > >> I ran the arc-arc test after changing Y3 in lines 3 & 7 to Y2 and Y-3 in > >> line 6 to Y-2 to better suit the envelope of my mill. > >> Using 2.6 standard it ran ok, using 2.6 with arc-blend when running line > >> 6, it jumped to line 7 when it reached Y0 rather than Y-2. > >> Screenshots at https://www.dropbox.com/sh/wb05n11coih47ld/gJ6AY1Drch > >> Cheers, Phill > >> > >> On Sat, 4 Jan 2014 01:43:23 -0500 > >> Robert Ellenberg <rwe...@gmail.com> wrote: > >> > >>> I just noticed a weird bug in a recent commit on the circular arc > branch > >>> that causes acceleration violations when pausing / stopping. I've > rolled > >>> back the circular arc branch to a known good commit to be safe. If > anyone > >>> has pulled from that branch in the last three days, please reset to the > >>> current circular-blend-arc-alpha branch, which should point to this > >> commit: > >>> 689e5cb185f87deba62391cd184f21f360ce2246 > >>> > >>> I'll push an update once I'm sure the bug is fixed. > >>> > >>> > >>> On Mon, Dec 30, 2013 at 1:20 PM, Robert Ellenberg <rwe...@gmail.com> > >> wrote: > >>>> On Dec 30, 2013 12:24 PM, "sam sokolik" <sa...@empirescreen.com> > >> wrote: > >>>>> Could you explain your 'small gain'? from what I am seeing - you > >> have > >>>>> to slow down to do a parabolic blend (If I understand it - it is a > >>>>> read-ahead buster) > >>>>> > >>>>> There are paths (like steve.ngc) which would be able to keep the > >> speed > >>>>> up if there was arc-arc, arc-line blends and you can optimize the > >> path > >>>>> (depth). And yes - I know it is a lot of work. > >>>>> > >>>>> (you are doing a great job - it is very impressive how quick you got > >> to > >>>>> this point) > >>>>> > >>>> What i mean by small gain is more statistical. It seems like most > >>>> engraving-type g-code is made up of line segments, so the biggest > >> overall > >>>> gain is from making arc blends between lines, simply because there are > >> so > >>>> many line intersections. It is true that hitting an arc-arc or > line-arc > >>>> blend is a lookahead buster, though, so it doesn't take much to cause > >>>> annoying slowdowns like in Steve.ngc. > >>>> > >>>> I'm planning on a little more hardware testing first, but if code > like > >>>> Steve's and Julian's is common, then it looks like blends with > existing > >>>> arcs are a must-have. Even the Stellabee code showed a little bit of > >>>> slowdown due to the occasional arc , though overall it's a small > >> effect. > >>>> Does anyone else have example code that shows this kind of slowdown? > >>>> > >>>> -Rob > >>>> > >> > ------------------------------------------------------------------------------ > >>> Rapidly troubleshoot problems before they affect your business. Most IT > >>> organizations don't have a clear picture of how application performance > >>> affects their revenue. With AppDynamics, you get 100% visibility into > >> your > >>> Java,.NET, & PHP application. Start your 15-day FREE TRIAL of > >> AppDynamics Pro! > >> > http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk > >>> _______________________________________________ > >>> Emc-developers mailing list > >>> Emc-developers@lists.sourceforge.net > >>> https://lists.sourceforge.net/lists/listinfo/emc-developers > >> > >> > ------------------------------------------------------------------------------ > >> Rapidly troubleshoot problems before they affect your business. Most IT > >> organizations don't have a clear picture of how application performance > >> affects their revenue. With AppDynamics, you get 100% visibility into > your > >> Java,.NET, & PHP application. Start your 15-day FREE TRIAL of > AppDynamics > >> Pro! > >> > http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk > >> _______________________________________________ > >> Emc-developers mailing list > >> Emc-developers@lists.sourceforge.net > >> https://lists.sourceforge.net/lists/listinfo/emc-developers > >> > > > ------------------------------------------------------------------------------ > > Rapidly troubleshoot problems before they affect your business. Most IT > > organizations don't have a clear picture of how application performance > > affects their revenue. With AppDynamics, you get 100% visibility into > your > > Java,.NET, & PHP application. Start your 15-day FREE TRIAL of > AppDynamics Pro! > > > http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk > > _______________________________________________ > > Emc-developers mailing list > > Emc-developers@lists.sourceforge.net > > https://lists.sourceforge.net/lists/listinfo/emc-developers > > > > > > ------------------------------------------------------------------------------ > Rapidly troubleshoot problems before they affect your business. Most IT > organizations don't have a clear picture of how application performance > affects their revenue. With AppDynamics, you get 100% visibility into your > Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics > Pro! > http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk > _______________________________________________ > Emc-developers mailing list > Emc-developers@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/emc-developers > ------------------------------------------------------------------------------ Rapidly troubleshoot problems before they affect your business. Most IT organizations don't have a clear picture of how application performance affects their revenue. With AppDynamics, you get 100% visibility into your Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro! http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk _______________________________________________ Emc-developers mailing list Emc-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-developers
