Non-PID Speed Control

[drbitboy]

Isn't that what @Ken Moore wrote in post #7 i.e. slip parameter?

Whoopsie! That should say post #5.

 

[Peter Nachtwey]

dead band better than filters. try it.

NO! I have. It causes oscillation between the two limits. Our controllers have a dead band eliminator to eliminate the dead band that some valves have.

 

[L33er]

NO! I have. It causes oscillation between the two limits. Our controllers have a dead band eliminator to eliminate the dead band that some valves have.

So deadband cause oscillation, use external limits then?

 

[drbitboy]

So deadband cause oscillation, use external limits then?

There may be oscillation between limits (pillar to post) with deadband, but without a deadband the PID will respond to noise (measurement, shot noise) and actually increase the noise in the PV data.

 

[Peter Nachtwey]

There may be oscillation between limits (pillar to post) with deadband, but without a deadband the PID will respond to noise (measurement, shot noise) and actually increase the noise in the PV data.

You want to ignore the noise while in the dead band. That is not a good idea because it ignores the underlying PV signal. If the noise is Gaussian ( random ), then a filter will smooth it out without adding a bias. The filter can be very strong (low cutoff frequency). This forum only discusses first order filters but there are other filters that will reduce the noise significantly.

 

[drbitboy]

You want to ignore the noise while in the dead band. That is not a good idea because it ignores the underlying PV signal.

This is where the wheels come off your point: the deadband limits are set to ignore noise but not signal; an inherent assumption is that PV maintained within those limits is adequate, so PV signal in that range can be ignored i.e. the statement above is correct but not relevant.

Also, some noise will pass through a filter, though it is attenuated of course; that attenuated noise will be increased by a PID without a deadband.

 

[Phrog30]

You know if there is a thread regarding PID Pete & Bit will be going at it.

 

[Peter Nachtwey]

This is where the wheels come off your point: the deadband limits are set to ignore signal but not noise;

How do you tell the difference?

an inherent assumption is that PV maintained within those limits is adequate,

Bad assumption

so PV signal in that range can be ignored i.e. the statement above is correct but not relevant.

This ignores how the PV is changing within the dead band.

Also, some noise will pass through a filter, though it is attenuated of course; that attenuated noise will be increased by a PID without a deadband.

There are filters and then there are FILTERS. I can use my mathemagic to filter just about anything. I can reduce the noise so it is insignificant compared to the width of your dead band. As long as the noise doesn't introduce a bias the filtered PV will be much closer to the actual than some PV in the dead band somewhere.

 

[drbitboy]

Originally Posted by drbitboy
This is where the wheels come off your point: the deadband limits are set to ignore signal but not noise;

How do you tell the difference?

By whether the [signal + noise]. i.e. the measurements, exceeds the deadband limits.

Both filtering and deadbands are attempting to do the same thing, i.e. identifying signal in measurements. Filters are analog and do it with all measurements; deadbands are discrete and simply say whether the signal is at the setpoint, or, stated another way, whether the process is "in control" (cf. here for one definition of that term).




I apologize for the ambiguity of the statements in my earlier post; I think I have corrected, in blue, some of them below, which corrections should address the objections.

Originally Posted by drbitboy
an inherent assumption is that PV measurements, i.e. PV signal plus noise, maintained within those limits is adequate,

Bad assumption

So the (corrected) assumption statement boils down to one of the normality of the noise (e.g. Anderson-Darling test).

Originally Posted by drbitboy
so if PV measurements i.e. signal plus noise, stays in that range then the variation can be ignored

This ignores how the PV is changing within the dead band.

It does not ignore how the PV signal is changing, but it does ignore how the PV measurement (i.e. signal plus noise) is changing when that signal is at setpoint. I.e. deadbands around a setpoint, if sized properly, ignore noise and let the signal through; it feels fuzzy, but it works.

 

[L33er]

Hey, I am a simple-minded guy here. If dead band cause oscillation, why do PID instruction have dead band in the first place?

 

[Frost999]

Hey, I am a simple-minded guy here. If dead band cause oscillation, why do PID instruction have dead band in the first place?

For many processes, oscillation inside of the DB is acceptable. That's not WHY a deadband exists, but it is often (some times, whatever) exploited as a crutch for "difficult" PID tuning.