AVR-Chat

On Monday 18 April 2005 09:13, Dave Mucha wrote:

> I have a servo motor that accepts step and direction inputs and I want
> to monitor a very dynamic process.

> The process is plasma torch cutting.  Typically a couple hundred volts,
> but a simple voltage divider will yield a useable signal.

> The process is very noisy.  This means that if a heavy filter is used
> on the input, it will make control sluggish.   no filter and the unit
> will wear itself out in short order.

> I am hoping there is a simple solution with a PID type control.

No. PID doesn't like noisy environments either. Implementing the
filtering digitally gives you the advantage of being able to easily
change filter parameters; that's about it.

You can filter the output and still measure at the maximum rate.
PID is proportional-integral-differential; the integral part takes
care of the "long-term" error in the system; the differential bit
looks after rapid changes in response to input. 

> I searched the list, but got tired of only finding stuPID and raPID
> with the PID search.

I Googled and found the first hit:
http://www.stanford.edu/class/ee281/projects/aut2001/depthcontroller/bookmarks.html

Not too many stuPID hits with keywords "avr PID algorithm"

Another apparently useful link from that search:
http://sourceforge.net/mailarchive/forum.php?forum_id=296&max_rows=25&style=nested&viewmonth=200112

> Any ideas where to start ?

Look at where and when the noise is occurring.

See if you can reduce the source of the noise perhaps by selection
of different ground paths or type of wiring; perhaps even a small
change to the machine.  By placing a small sensing processor very
close to the sensor and then sending the opto-isolated data stream,
one can not only almost completely eliminate noise introduced via
signal leads, but it also allows the processor to "float" at the
potential of the sensor. Noise introduced by a variation in "ground"
potential from one point in the device to another due to heavy
pulsed current flows is thus minimised.

Although unlikely in this case, if the interesting signal is
happening at a particular time in the cycle, you may be able to
exclude much noise by sampling in a narrow time window containing
only the signal.  You can still filter that, but the volume of noise
in a narrow time window is consequentially less. By "snooping" outside
the signal sampling window, you can also determine the magnitude of
"background" noise and process your measured signal accordingly.

If the noise is still an issue, break out the spectrum analyser and
see if your interesting signal components can't be filtered out by a
narrow-bandpass.

-- 
/"\ Bernd Felsche - Innovative Reckoning, Perth, Western Australia
\ /  ASCII ribbon campaign | I'm a .signature virus!
 X   against HTML mail     | Copy me into your ~/.signature
/ \  and postings          | to help me spread!