G2 slow feed down

[ger21]

Why can't the Y axis start accelerating earlier, rounding the corner. As long as the rounded corner does not exceed the error tolerance, this would be the desired result. The higher the error tolerance, the sooner the Y can accelerate, keeping the speed higher.

[cncdrive]

But there is no real corner between an arc and line connected like that.
And the software can't just cut it straight down even if the error tolerance would allow it it would still not look nice.
Check the code and you will see.
And yes, the higher the error tolerance and higher the acceleration the less the software slows down on the transition, there should be a point when it is no more noticable, it is measurable.

And as said Mach3 resolves it with overacceleration of the settings, but a 8x overacceleration is just not the right solution, it is just cheating which involves the risk that the stepper motor will stall.

[ger21]

But there is no real corner between an arc and line connected like that.

It should work like the example pic in the manual for corner error.
Calculate a fillet between the line and the arc, and follow that fillet.

No it won't look good, but in some cases better performance is preferable to accuracy.
And since it would be controlled by the corner error setting, the user could find a middle ground that was acceptable.

[cncdrive]

Gerry, yes, it is already doing that, but if the acceleration of the axis is not high enough then it will still not be able to accelerate up in time even if the path is modified/optimised like that.
And so is why it is not a complete decceleration, only a minor slowdown which slowdown is smaller as the acceleration is higher.

[Derek]

My collegue got back and we studied this issue and got to a conclusion that it has to work how it works now.

That makes sense but I think overall there is more to it than that. I see the slowdown happen in all kinds of movement. X and Y moving and then getting to an arc comes to mind. When I get time I'll do some testing with some larger arcs intersecting 45 degree lines and see what that looks like. X and Y should be at the save velocity on the straight and if the arc radius is large enough there should be little or no pause at the intersection.

And let's take a trip back to 2006 where I posted this:

When I do 3D work my cam program writes everything as line segments. Even a perfect arc. I have noticed that even with lookahead set at 300 or more there is a noticeable speedup and slowdown as UCCNC processes each line segment. If I convert the same line segments to arcs in the cam program it is one fluid motion. I use a .001" tolerance in the CAM program so the line segments are pretty short.
I'm running in CV mode.
I normally convert to arcs as much as possible but I would like to get the same smooth motion with line segments if possible. Not sure if it is a settings issue or a CPU issue or if "that is just how it is" issue.

Thanks
Derek

http://forum.cncdrive.com/viewtopic.php?f=2&t=31#p67

Why can't the Y axis start accelerating earlier, rounding the corner. As long as the rounded corner does not exceed the error tolerance, this would be the desired result. The higher the error tolerance, the sooner the Y can accelerate, keeping the speed higher.

This would be my preferred solution. A lot of the stuff I carve is not that critical dimension wise and having the option to have smooth movement in a controlled fashion would help. As we have all said none of the settings available to us makes a difference in this instance.

[ger21]

Gerry, yes, it is already doing that, but if the acceleration of the axis is not high enough then it will still not be able to accelerate up in time even if the path is modified/optimised like that.

If you add a "blended" tangent arc between the line and arc, then the required Y accel would start at 0. The size of the blended move would be dictated by the accel setting. But there's no reason why it has to slow down, unless the accel is so slow that the blended move would be too large.

[cncdrive]

Unfortunately I don't think that we can do it better. I see no way to make it better. There is a reason (which I described) why it works like that in the UCCNC and in Linuxcnc too.
We could do tricks like in Mach3, but that would make it worse in overal like it is in mach3.

Gerry, the motion is already blended, otherwise there would be a full decceleration to zero feedrate, but there is only a small bump in the feedrate, because the Y axis has to startup from zero speed.
That bump in the feedrate is larger as the acceleration is lower and it becomes unnoticable if the acceleration is high enough.

I understand that you guys want a perfect solution, but unfortunately the world and it's physics is not perfect.

[Robertspark]

Balazs, thanks for you and your software engineer taking the time to check out the issue.

No we don't want Mach3's solutions or erratic behavior on acceleration.

[ger21]

Yeah, I'm not complaining, just making a suggestion to try to make it better for those with machines with low acceleration.
I don't have the understanding of how it actually works, so if it's as good as it can be, then so be it.

[cncdrive]

We have compared to the newest one. But as said there is no magic, it just can't work better without cheating which cheating makes things worse in overal.