Re: Low cost Servo Controller

CG wrote:

> From: "CG" <cnk@...>
>
> Bertho-
>
> Take a look at the LM629 Motion Controller chip from National Instruments.
> It's an all-in-one motion controller that handles the encoders, current
> sensing, programmable PID loop, and either 8/12 bit DAC output or 8 bit
> sign-magnitude PWM output to the servo amplifier.

That's National Semiconductor, oops, they were bought by Fairchild, Oops,
Fairchild spun them off, I don't even know who they are, now, but maybe back
to National Semiconductor?

The LM628 is analog out, the LM629 is PWM out. Unfortunately, the
frequency of the PWM waveform is pretty much fixed to 4 X the PID
cycle time of the servo loop. It definitely does NOT do current sensing.
It has a VERY sophisiticated PID servo loop, very sophisticated control
of trapezoidal acceleration ramps, extremely fine control of desired velocity
and acceleration due to Vel. and Accel. parameters being 32 bit numbers with
16 bits of fraction. But there is a big hole in the design - synchronization.
Actually, for a linear move, it is pretty sure that 2 or more axes will track very
well, as long as no axis binds up. I don't think EMC has this (yet) but many
CNC controls have a feature that slows all axes down if one axis starts to
lag excessively. So, if one axis binds a little, the other axes slow accordingly,
or just the feed rate is stepped down to allow the slow axis to catch up.
This would be VERY hard to do on the LM628 series. The other area
of worry is that when 2 or more axes are following a circular interpolation
or a series of very small, multi-axis moves (contouring) multiple LM628
series chips could get out of sync. There really is no good provision for
updating more than one axis for continuous moves. The way you do it
for one axis is to watch for the LM628 to go into ramp-down mode
(or just watch the current position) and then jam in a new end point
and velocity. But, for multiple axes, you could get into a situation where
the axis with the longer move gets further and further behind, as you
keep jamming new coordinates into the chips.

I called National Semi, and they had no real help for me, other than suggesting
that I should try this, as they knew of no one using these chips for
coordinated, multiple axis moves, with mid-course updating.

> They show a simple
> circuit with the LM629 using PWM to talk to their own LMD18201 3A 55V
> H-bridge. You need to talk to the LM629 with a micro of course, but the
> language is very straightforward. It seems like a good starting point for
> figuring out the basic functions and architecture for an all-digital
> low-cost system we're talking about. These chips are expensive, as far as
> chips go ($25 ea.), but they do just about everything. I don't have much
> time now, but I've got the data book in front of me if you need more info.
> (National Power ICs Databook).

Actually, I think you'll find they are closer to $60 in small quantity. I think it
has to do with a license fee National has to pay to whoever designed the
DSP processor core.

Anyway, I still don't think the LM628 series will do it all, as it still only
outputs a velocity command, and you still have to arrange for the motor
to accurately follow that velocity. Maybe the PID algorithm can do it
all, at high enough gain to keep the following error close to zero, without
undesirable resonances or hunting. Really, it is pretty hard to get very low
following error when moving AND when stopped, and not also get the
annoying hunting when the machine is supposed to be standing still.

I planned to use these chips for a milling machine control, and researched them
thoroughly before EMC came around. I was not confident at all that they
would really do the job. I know that EMC DOES do the job!

Jon