Experimental Physics and
Industrial Control System

"J. Lewis Muir" <[email protected]> · Wed, 22 Feb 2012 16:53:21 -0600

On 2/22/12 4:29 PM, Paul Nord wrote:
> I've got a network device that's misbehaving.  It "sometimes" reports an absolutely ridiculous value.  I'd like to set an alarm on this PV but I don't want the alarm sounding once or twice an hour when there's not really anything wrong.  (Never cry wolf.)
> 
> In the typical failure mode it reports a "crazy" value once and then goes back to the normal value.  (The device actually returns the voltage value for a channel when you ask for a current.  And it formats a reply and happily returns "Current: 3600 A", which would be impressive for a supply running at 3600 V.  It works out to 13 MegaWatts.  The real current is just ~100 uA.)
> 
> What's the best way to deal with this?
> 	A CALC pv that takes this value as an input but returns zero if the reading was outside of some range?
> 	A CALC pv that counts the number of readings which are outside of the proper range?  (e.g. A>1.0?(VAL+1):0 )
> 
> Are there other good mechanisms?

Hi, Paul.

For a glitchy device reading where I don't want to see the
glitch, I typically keep track of the most recent non-glitch
value read from the device and a count of the number of
consecutive glitches.  If the value read is not a glitch, I use
it and reset my glitch count to 0.  If the value read is a
glitch, I increment the glitch count.  If the glitch count is
less than or equal to the max allowed glitch count, I use the
most recent non-glitch value.  If the glitch count has exceeded
the max allowed glitch count, I set the value to some in-range
default value (e.g. 0) and set the alarm status to INVALID.
This can be done in a C/C++ driver or using just records in
EPICS Base.

Lewis

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System