Re: [Emc-users] jerk control
On 8/24/21 1:08 AM, andy pugh wrote: > I have a feeling that kinematics is not a problem in most cases, the > kins functions run fast enough to be used for finite-difference > differentiation / numerical integration. There is an elegant way to compute exact derivatives without computing analytical derivatives by hand. Here is a nice introduction: http://ceres-solver.org/automatic_derivatives.html#chapter-automatic-derivatives Best Regards, Alexander ___ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
Re: [Emc-users] jerk control
Well I have two high speed CNC machines.. But they are all analog control. Probably no good for tinyg On Thu, 26 Aug 2021, 04:12 dave engvall, wrote: > IIRC emc was conceived as a vehicle to test intercommunication between > processes and as such higher level features such as lookahead, > smoothing, etc were not part of the master plan. In fact emc would not > have had stepping if Matt Shaver had not requested it. > > Now on to the real subject of this email. Is there a user that has the > resources to test linuxcnc vs tiny g for smoothness vis a vis jerk. I > suspect that tiny g does not have integrated tools to do this so simply > milling under extreme conditions may be the only way to access the > utility of its jerk component. > Ideas, comments. > Ray Henry used to talk about tuning by milling before we had good tools > in emc. It could also be that he was dealing with analog controls. ;-) > > Dave > > On 8/23/21 5:51 PM, andrew beck wrote: > > Andy do you know what the tormach uses for more than 3 axis path > blending? > > > > > > > > On Tue, Aug 24, 2021, 11:11 AM andy pugh wrote: > > > >> On Mon, 23 Aug 2021 at 21:27, andrew beck > >> wrote: > >>> Just had a look at tiny g looks great. > >> I did try to implement a zero look-ahead finite jerk planner for laser > >> rastering. It was interesting, and I learned a bit. > >> > >> It is easier the less general you make it. > >> > >> Ideally LinuxCNC would have a 9-axis finite-jerk planner that handled > >> arbitrary kinematics with feed-override control. > >> > >> Tiny-G is a 3-axis (I think) planner with trivial kinematics and no > >> feed override (AFAIK). > >> > >> At the moment I would be happy just to see LinuxCNC handle more than > >> 3-axis blending. It's in Tormach. > >> > >> I have a feeling that kinematics is not a problem in most cases, the > >> kins functions run fast enough to be used for finite-difference > >> differentiation / numerical integration. > >> I am not sure about the more computationally intensive ones, such as > >> genserkins. (I think that is fast forwards, slow inverse) > >> > >> -- > >> atp > >> "A motorcycle is a bicycle with a pandemonium attachment and is > >> designed for the especial use of mechanical geniuses, daredevils and > >> lunatics." > >> — George Fitch, Atlanta Constitution Newspaper, 1912 > >> > >> > >> ___ > >> 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 > ___ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
Re: [Emc-users] jerk control
So for an abort I think would be still following the proscribed machine path in a coordinated fashion but stop as fast as possible within constraints. Sort of like dragging feed override to 0 as quickly as possible without violating constraints. For an e-stop it will be an uncoordinated stop. Each joint will halt as quickly as possible independently while still obeying joint constraints. I think that jerk limit will apply for all joint and cartesian space. On Mon, Aug 23, 2021 at 4:54 AM andy pugh wrote: > On Mon, 23 Aug 2021 at 04:40, Chris Albertson > wrote: > > > Actually for a machine tool, why not run the > > simulation off-line and use as much time and computer power as it takes. > > Feed-override? > > Do you allow infinite jerk on abort? You might think that is an easy > question, except that continuous jog is implemented as a move to the > limit that is aborted on key release. > > And, do we need to jerk-limit in joint space or cartesian space, or both? > > -- > atp > "A motorcycle is a bicycle with a pandemonium attachment and is > designed for the especial use of mechanical geniuses, daredevils and > lunatics." > — George Fitch, Atlanta Constitution Newspaper, 1912 > > > ___ > 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
Re: [Emc-users] jerk control
IIRC emc was conceived as a vehicle to test intercommunication between processes and as such higher level features such as lookahead, smoothing, etc were not part of the master plan. In fact emc would not have had stepping if Matt Shaver had not requested it. Now on to the real subject of this email. Is there a user that has the resources to test linuxcnc vs tiny g for smoothness vis a vis jerk. I suspect that tiny g does not have integrated tools to do this so simply milling under extreme conditions may be the only way to access the utility of its jerk component. Ideas, comments. Ray Henry used to talk about tuning by milling before we had good tools in emc. It could also be that he was dealing with analog controls. ;-) Dave On 8/23/21 5:51 PM, andrew beck wrote: Andy do you know what the tormach uses for more than 3 axis path blending? On Tue, Aug 24, 2021, 11:11 AM andy pugh wrote: On Mon, 23 Aug 2021 at 21:27, andrew beck wrote: Just had a look at tiny g looks great. I did try to implement a zero look-ahead finite jerk planner for laser rastering. It was interesting, and I learned a bit. It is easier the less general you make it. Ideally LinuxCNC would have a 9-axis finite-jerk planner that handled arbitrary kinematics with feed-override control. Tiny-G is a 3-axis (I think) planner with trivial kinematics and no feed override (AFAIK). At the moment I would be happy just to see LinuxCNC handle more than 3-axis blending. It's in Tormach. I have a feeling that kinematics is not a problem in most cases, the kins functions run fast enough to be used for finite-difference differentiation / numerical integration. I am not sure about the more computationally intensive ones, such as genserkins. (I think that is fast forwards, slow inverse) -- atp "A motorcycle is a bicycle with a pandemonium attachment and is designed for the especial use of mechanical geniuses, daredevils and lunatics." — George Fitch, Atlanta Constitution Newspaper, 1912 ___ 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
