Granted this subject is a bit old I've now had some time to dive back into the TI F2837xD which has dual processors and other features that will make it a good test bed for trying out stuff.
It has a hardware Quadrature Encoder Interface (QEI) so theoretically I should be able to grab encoder counts in any resolution and calculate spindle position relative to Z axis position and create moves to track. I'll report back when I have some more information. John > -----Original Message----- > From: Robert Ellenberg [mailto:rwe...@gmail.com] > Sent: January-20-21 10:12 AM > To: Enhanced Machine Controller (EMC) > Subject: Re: [Emc-users] Spindle speed changes with threading. > > Hi All, > > As others have said, during position-synched moves, the axes follow the > spindle position, so you don't need fine control of spindle speed. However, > you should have both a stable spindle RPM and a high-ish resolution encoder > to get the best results. John, for your example, as each encoder pulse > arrives, the TP will be constantly accelerating / decelerating to try to > track that position signal. Luckily, it can't get too far off because of > the axis acceleration limits, but the amplitude of the jitter will > definitely be higher with a low resolution encoder. > > Here's how spindle synchronization broadly works from within the TP (i.e. > what occurs in motion after START_FEED_SPEED_SYNCH) > > 1. The TP waits (with axes stopped) for a spindle index mark. > 2. Once the spindle has just passed the mark, the machine axes > accelerate until they reach (spindle speed * requested units/rev). > 3. Once the axes reach the expected velocity, then synchronization is > declared, and the TP maintains position synchronization from that point > onwards. At higher spindle RPM, synchronization takes longer, so the > spindle rotates farther before the axes are synchronized. Multiple > threading passes at different spindle RPM will not quite follow the same > path. > 4. After synchronization, the TP computes a position error based on the > measured spindle position. The axis velocity is nominally spindle speed * > uu / rev, but that velocity is corrected up or down as needed to drive the > position error toward zero. > > Note that multi-start threading is not currently possible (because the TP > always synchs at 0 deg, i.e. at the index mark), but we could modify the TP > to start synchronization at some angle after the index mark. > > Finally, the obvious fix for the inconsistency in the acceleration phase is > to just declare synchronization at the index mark right away. With the axes > starting from rest, there would be a huge initial velocity error for the TP > to correct. It will do so eventually, but there are large position over- / > under-shoots until it stabilizes. Luckily, we know what the axis > acceleration looks like (based on axis max acceleration and nominal spindle > speed), so you can anticipate this and start the axes moving before the > spindle reaches the index mark. That way, the axes are moving at the > nominal speed just as the spindle reaches zero. This removes most of the > position error at the start, and the TP corrects for any residual error > very quickly. This is roughly the approach we used in PathPilot, and I > think LinuxCNC 2.8 or 2.9 would benefit from it as well. > > Best, > Rob > > On Wed, Jan 20, 2021 at 12:47 PM John Dammeyer <jo...@autoartisans.com> > wrote: > > > > > > > > From: andy pugh [mailto:bodge...@gmail.com] > > > > > > On Wed, 20 Jan 2021 at 16:09, Kirk Wallace <kwall...@wallacecompany.com> > > wrote: > > > > > > > I left the trail here: > > > > > > > https://github.com/LinuxCNC/linuxcnc/blob/master/src/emc/rs274ngc/interp_convert.cc#L4881-L5028 > > > > > > There seems to have been a lot of time spent investigating theory that > > > could have been settled in 5 minutes with an experiment. > > > > > > -- > > > atp > > > > Dropping an apple from a tree and observing that it falls and smashes on > > the ground doesn't splatter into the words that spell out laws of motion > > made up of bits of peel and apple. > > > > I'm assuming that the authors of this code were clever enough to take into > > account the motor acceleration relative to spindle speed on each pass. But > > that doesn't explain how they do that. > > > > And if there are 60 teeth on the spindle encoder with a single sensor then > > 120 edges are the most you get. That's 3 degrees per edge assuming the > > slots are symmetrical and I don't think there's a rule that they must be. > > Might be 1 and 5 degrees. So assume then an index single rising edge is > > used every 6 degrees. > > > > A half inch shaft has a circumference of 0.3925" and each 6 degree index > > is 0.00654". The implication is depending on spindle speed and motor > > acceleration you might be off by almost 0.006". That's a lot isn't it? > > > > John > > > > > > > > > > _______________________________________________ > > Emc-users mailing list > > Emc-users@lists.sourceforge.net > > https://lists.sourceforge.net/lists/listinfo/emc-users > > > > _______________________________________________ > Emc-users mailing list > Emc-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/emc-users _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
