Arc Junctions, corner error max? linear error max? ??

[digger]

Hopefully they already no how to fix this when they start on the new planner soon.

Any news about the date for new planner?

Milosh

[cncdrive]

No, we not working on the new planner yet. It will probably take a half year or so from the time we will start, because it requires lots of things to be changed and then verified etc.
The current planner is already better than Mach3/4, so we are not in a big hurry with it, there are several other new functions which we currently adding and working on instead of fixing a planner which already works better than the concurent ones.
And I'm not sure why those arcs causing a slowdown, but will ask my API developer collegue about it.

[digger]

I believe you that it is better than other one. My impression is that people (myself included) are a bit confused with a slowdown, and we were not experienced that with others programs.

Milosh

[cncdrive]

Yes, you not experienced it with mach3, because it simply cuts off details of the path so it can run fast anywhere anytime, the only bad thing then is that the workpiece will be "garbage" if you need low tolerances.
In other words it is easy to run fast when you can go off the road anywhere with any distance and there are no rules for you.

[cncdrive]

I forgot to mention the other thing is that mach3 often oversteps the acceleration in these cases. We have measured out upto 8 times the setup values, this is another reason how can it run fast. But ofcourse this involves risk that your motors will stall due to overstress in the acceleration. Mach3 simply not obeys your settings, it just doing these things how it thinks it's fine, but ofcourse it is not fine for the machine and for the workpiece.

[Robertspark]

I am not comparing uccnc to other programs

I am just trying to understand the math involved as to why it is slowing down on segments of the same circle (or 4 segments.... Or more.... All with the same center point / origin of the arcs.

Then I'd like to understand the math or improve the math where two arcs intersect and allow for a use of the corner error max parameter on these junctions to allow for controllability / uniformity of the parameters for junctions

Understanding the math is obviously not essential.... Improvements in the motion are the bit that's important

[cncdrive]

Hi Rob,

I was just explaining to "digger" why he did not experience it with other software (mach3/4). It is simply because those softwares not obeying the setup rules risking step loss and they have no path tolerance control at all. So comparing them to the UCCNC planner is kind of funny.

I'm not sure why they slowdown on those arcs and currently we are very busy with the end of the year sales and finishing the latest test release, so if you don't mind I will get back to this later.

[Robertspark]

Balazs, no problem I was just making it clear to anyone else that this is not a comparison thing. Its just looking at uccnc.

Yup I know you are busy on the next development release and those sales are what keeps us all here so they are your priority.
Its no rush on this one as uccnc works its just one of those "hmmm.... why does it do that?" moments + I'm sure the next development release will keep everyone busy for a while

[Robertspark]

Ok, I know that this is on cncdrives radar, however I've been looking at some of the settings to try to understand them better.

The clips below may help with diagnosing what the issue is. To me it looks like the corner error max is being applied to the arcs and is undersizing the junctions ..... over quite a distance of the arc instead of just applying the corner error to the actual junction

So here is another simple gcode.

Basically it is two circles, the first is divided into 4 quadrants, and the second is a complete circle.

Code: Select all

N0000 G0 X500 Y0
N0100 G03 X0.0000 Y500.0000 I-500.0000 J-0.0000 F10160
N0110 X-500.0000 Y0.0000 I0.0000 J-500.0000
N0120 X0.0000 Y-500.0000 I500.0000 J0.0000
N0130 X500.0000 Y0.0000 I-0.0000 J500.0000
N0130 X500.0000 Y0.0000 I-500.0000 J0.0000
N0000 G0 X0 Y0
N240 M30

These are BIG circles and they have just been run in demo mode on UCCNC.
They are in mm, and have a diameter of 1000mm.
The feedrate is 10160mm/min (200"/min)

I played around with the X&Y axis acceleration and also with the linear error max, corner error max and the stop on angles (<89º... <180º etc)

First I set all the configuration settings (linear error max, corner error max etc) to zero
I set the acceleration to 58mm/s/s ........ (why? to prove / test UCCNC obeys centripetal acceleration Acc = velocity ^2 / radius..... so in our case it would be (10160/60)^2 / 500mm = ~ 57.35 mm/s/s
https://youtu.be/oz7PUGawYBY

And ran the code.... the feedrate slows down at each of the quadrant junctions to zero (i.e. its exact stop), but on the full circle the motion follows the arc without any slow downs (NOTE: it follows the arc), and the feedrate achieved the programmed feedrate of 10160mm, but only on the full circle.

Ok, now I changed the acceleration slightly to 57mm/s/s (again to prove centripetal acceleration is correct). And ran the code.... this time the feedrate was not achieved (as would be expected) but the motion was ~exactly the same in that is slowed down to ~zero for every quadrant.


Ok, now I changed the corner error max from ZERO to 10 units (mm) and re ran the code.
The code ran quicker (as one would expect) .... and it did not slow down as much between the junctions of the quadrants (as would be expected) as we eased off on the junction parameter tolerance..... however have a look at the circle..... the motion was unchanged..... it follows the line of the arc.
https://youtu.be/8rcWLEWHKMA

Now I re-ran the code, zoomed into the motion....
Note the effect that the corner error max tolerance has on the junctions between the quadrants how the arcs (of the same circle) would be undercut at the quadrant / arc junctions..... but then when you rewatch the complete circle motion there is no slow down (as would be expected) but there is also no undercut in the arc...... and the feedrate is controlled according to the acceleration and limited the feedrate to slightly below the feedrate set of 10160 as the acceleration was 57mm/s/s ([ SQRT (57 x 500) ] * 60 = 10129mm/min as is expected.

https://youtu.be/nAGOiWO_LK4

[Robertspark]

Ok, same code as the last post, this time I changed the acceleration of the X&Y axis to 5mm/s/s just so I could get an idea of how much the arc junctions are being undercut with the corner error max

Corner error max was still set at 10 units (mm).

https://youtu.be/XKcEKFyVFfs

To me it looks like the cundercut is 20 units (almost) given the circle is 500mm radius and the axis rise to around 480 + a bit.

But again when cutting the circle UCCNC just limits the feedrate to the centripetal acceleration [ SQRT ( 5 * 500 ) * 60 ] = 3000 mm/min {which is correct}

I cannot figure out a way to do this mathematically {yet}

Corner error max is the right parameter to apply just its being applied wrong (in my opinion) and should only be concerned with applying a radius to the junction of two moves