Re: [Emc-users] Maybe a minimally printed harmonic drive?

[Gene Heskett]

On Sunday 02 May 2021 16:34:43 Jon Elson wrote:

> On 05/02/2021 01:00 PM, Chris Albertson wrote:
> > The below is exactly true.  The problem is that PID is widely used
> > and PID is only "reactive" it can only deal with what has already
> > happened and the time lag is the source of oscillations.In the
> > past, PID was the only control option because we had limited
> > computing power.
>
> True for classic PID.  The PID calculation in LinuxCNC also
> has feedforwards (FF0 - FF2) which
> are quite useful, as they give an approximation of the
> required drive to the motor without the output, move, sample
> delay.
>
Yes, and that I have found usefull.  But using them intelligently isn't 
at all well explained by the docs we have. Time constants and their 
effects are are not at all well explained. The bottom line is that if 
and when you hit the magic settings, you still have no idea why this 
works and that doesn't.
> Jon
>
>
> ___
> Emc-users mailing list
> Emc-users@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/emc-users


Cheers, Gene Heskett
-- 
"There are four boxes to be used in defense of liberty:
 soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
If we desire respect for the law, we must first make the law respectable.
 - Louis D. Brandeis
Genes Web page 


___
Emc-users mailing list
Emc-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/emc-users

Re: [Emc-users] Maybe a minimally printed harmonic drive?

[Eric Keller]

A well-tuned PID with feedforward is capable of controlling a machine tool
without any issues.  Sure there are better ways to implement a control, but
the results are not going to be significantly different on something as
simple as a machine tool.  After the academic controls community played
around with robust control for 15 years, it's really difficult to get
funding to do any more work in that field. I'm not sure what it would take
to revive it.
Eric Keller
Boalsburg, Pennsylvania


On Sun, May 2, 2021 at 4:38 PM Jon Elson  wrote:

> On 05/02/2021 01:00 PM, Chris Albertson wrote:
> > The below is exactly true.  The problem is that PID is widely used and
> PID
> > is only "reactive" it can only deal with what has already happened and
> the
> > time lag is the source of oscillations.In the past, PID was the only
> > control option because we had limited computing power.
> True for classic PID.  The PID calculation in LinuxCNC also
> has feedforwards (FF0 - FF2) which
> are quite useful, as they give an approximation of the
> required drive to the motor without the output, move, sample
> delay.
>
> Jon
>
>
> ___
> 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] Maybe a minimally printed harmonic drive?

[Jon Elson]

On 05/02/2021 01:00 PM, Chris Albertson wrote:

The below is exactly true.  The problem is that PID is widely used and PID
is only "reactive" it can only deal with what has already happened and the
time lag is the source of oscillations.In the past, PID was the only
control option because we had limited computing power.
True for classic PID.  The PID calculation in LinuxCNC also 
has feedforwards (FF0 - FF2) which
are quite useful, as they give an approximation of the 
required drive to the motor without the output, move, sample 
delay.


Jon


___
Emc-users mailing list
Emc-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/emc-users

Re: [Emc-users] Maybe a minimally printed harmonic drive? --> MPC and other control theory

[Nicklas SB Karlsson]

Den 2021-05-02 kl. 20:00, skrev Chris Albertson:

The below is exactly true.  The problem is that PID is widely used and PID
is only "reactive" it can only deal with what has already happened and the
time lag is the source of oscillations.In the past, PID was the only
control option because we had limited computing power.   People are
starting to use MPC for motion control now.  MPC is predictive and can
account for "spring back", backlash and the time it will take to
decelerate.   But it does require modern compute power.   If you google,
you find many papers on MPC applied to milling machines starting in about
2010.


So what's the difference between MPC Model Predictive Control and feed 
forward? Solve equation analytical instead of using analytical solution?


Read some about it but did not figure out what they are doing but should 
not be able to predict something not in the model.



Eventually, this will filter down to the open-source DIY hobby community.
If you have ever seen a machine running with this kind of controller it is
visible even just looking at it by eye, the machine looks "alive" and moves
more like a human or a dog rather than a robot.
A consultant at work my workplace brought two dogs, kept them in a 
nearby room during meeting and they pee on the floor. Machine dog eat 
oil and shit screw is much better.

The problem is the complexity.  It requires some advanced math and the
computer has to be able to run thousands or likely millions of simulations
of the entire machine per second.   This is maybe why we don't see
research on this done in the 1990s or 2000s.   It is a good use for a
6-core Intel i9 CPU


Read model based control theory at University. Non-linear system may be 
linearized around working point while linear system may be stated 
directly. In both cases it will be a system on matrix form. There are 
standard methods to calculate feed-forward once model have been stated. 
For servo motors rather simple calculations give fast dynamic response 
though it might be possible to get more accurate control by modeling 
more details and improve feed-forward.



Nicklas SB Karlsson



___
Emc-users mailing list
Emc-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/emc-users

Re: [Emc-users] Maybe a minimally printed harmonic drive?

[Chris Albertson]

The below is exactly true.  The problem is that PID is widely used and PID
is only "reactive" it can only deal with what has already happened and the
time lag is the source of oscillations.In the past, PID was the only
control option because we had limited computing power.   People are
starting to use MPC for motion control now.  MPC is predictive and can
account for "spring back", backlash and the time it will take to
decelerate.   But it does require modern compute power.   If you google,
you find many papers on MPC applied to milling machines starting in about
2010.

Eventually, this will filter down to the open-source DIY hobby community.
If you have ever seen a machine running with this kind of controller it is
visible even just looking at it by eye, the machine looks "alive" and moves
more like a human or a dog rather than a robot.   It is hard to describe
but it comes from "knowing" what is about to happen and taking the future
half-second or so into account.The control will take action to prevent
oscillation or spring because it is predictive, not reactive.  This is
in use now for some years, it's not speculative.

The problem is the complexity.  It requires some advanced math and the
computer has to be able to run thousands or likely millions of simulations
of the entire machine per second.   This is maybe why we don't see
research on this done in the 1990s or 2000s.   It is a good use for a
6-core Intel i9 CPU

On Sun, May 2, 2021 at 7:22 AM Gene Heskett  wrote:

>
> That is the unmentioned in polite company problem associated with glass
> scales as even resolvers aren't completely instant. The rssolution of it
> might include a higher frequency than 1 kilohertz for the servo loop.
> Stepgen based feedback doesn't have that lag. They report the position
> they've sent to the motors, no lag other than the quantization of the 1
> kilohertz servo loop.


-- 

Chris Albertson
Redondo Beach, California

___
Emc-users mailing list
Emc-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/emc-users

Re: [Emc-users] Maybe a minimally printed harmonic drive?

[Gene Heskett]

On Sunday 02 May 2021 09:29:09 dave engvall wrote:

> Hi Gene,
>
> Your   comments on the servo-stepper are interesting. How well would
> they drive a BP sized mill?

I can't think of a GOOD reason, other than the mass of the knee would 
probably need some belt reduction in the knee. 3NM is in the 425 OZ/IN 
range and drives my GO704 at 70 or more IPM.  XY would be std ball 
screws of course.
>
> I'm thinking about picking up one to play with.
>
> Any reason you know of that they could not be used with and external
> encoder, eg glass scale for position?

None, it would be two different feedback paths that don't need mixing, 
let the motor and controller be run by the TP, and let the scales tell 
the TP where its at. PID's optional, all it would be is an additional 
rubber band in the control loop. I have no PID's in the sheldon config 
at all. Motion tells the motors what to do and they just do it. Feedback 
comes from the stepgen, and the motors do as they are told.  And if 
either can't do it, they stop linuxcnc in its tracks, litterally without 
a mark on the work that isn't springback.  Whats not to like?

The only problem I can see would be the time lag in the scales, they 
would have to be real time. Any lag would translate to overshoot and 
possible oscillation.  That would limit the useable accels of the 
machine as the TP or moiion would have to be able to anticipate the = 
point before it got there.

That is the unmentioned in polite company problem associated with glass 
scales as even resolvers aren't completely instant. The rssolution of it 
might include a higher frequency than 1 kilohertz for the servo loop.  
Stepgen based feedback doesn't have that lag. They report the position 
they've sent to the motors, no lag other than the quantization of the 1 
kilohertz servo loop.  And if the motors can't do it, they stop linuxcnc 
quite close to instantly.  You of course have to hook all that up in the 
hal file.

> Dave
>
> On 05/01/2021 05:51 PM, Gene Heskett wrote:
> > On Saturday 01 May 2021 12:34:01 Chris Albertson wrote:
> >> Gene,
> >>
> >> I did a search on the part number to gave and, they look really
> >> good but cost $400 per axis vs about $60 for what I have.   Yes,
> >> they are much better.Did you find a better deal?
> >
> > Egad, where are you looking? 6 months ago I bought 4 of them in two
> > orders thru ebay, 3 3NM's and a 2NM, and paid about $130 USD per
> > axis.
> >
> >  >wEZdaQa5B>
> >
> > 85$ copy, but $45 shipping.  So $130/axis, discounts for 3 or more.
> >
> > Cheers, Gene Heskett


Cheers, Gene Heskett
-- 
"There are four boxes to be used in defense of liberty:
 soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
If we desire respect for the law, we must first make the law respectable.
 - Louis D. Brandeis
Genes Web page 


___
Emc-users mailing list
Emc-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/emc-users