Trial and Error Control Approach 1

[ilan7]

The regular control loop aims to keep the process variable nearby a desired set point by changing the controlled variable (the output) as a function of the error (difference between process variable and set point). With the trial and error approach, the control system aims to "improve" the process variable (e.g. raise the yield or efficiency) by continuously changing the controlled variable and "checking the result": if the process variable improves, the system will keep on changing the output in the same direction. If it does not, the change direction will go the other way. Of course, many other issues are involved, such as the change cycle, the amount of change, etc.
Anyone heard of an application using this trial and error approach to an automatic controller?
Thanks to all

 

[hacksaw]

it has been the basis for closed loop control for about a century or so. Yes, I think you can say it has been discussed from time to time....

 

[CARF]

This is what an automatic PID controller does. It's not trail and error: This control type is based on solid process knowledge and mathematics (feedback control).

Cheers,
MVD

 

[hacksaw]

in computing the error and adjusting the control move in such a way that it is more or less proportional to the error and to further adjust the control move if the off-set continues is a variation of the trial and error adjustments made when the control valves were adjusted manually.

 

[CARF]

No, the PID controller should be tuned to the dynamics of the specific process.
These dynamics can be calculated with dynamic process models or measured, i.e. with a step response on the process. Once you know the process dynamics (behaviour) one can select the most appropriate Proportional, Integral and Deravitive action in the controller. Trail and error (for me at least) suggests "monkey around a bit and see what happens". This is not how correctly tuned automatic controllers work.

Please see also:

Http://www.controlstation.com/

 

[hacksaw]

point taken on tuning and loop configuration, but it is hardly as deterministic as you suggest. even the very best controller will experience performance degradation when subjected to random load changes. the ideal tuning parameters set only define a starting point in my own experience.

 

[CARF]

A fuzzy control system might follow a more trial and error control approach as described by ilan7. Here you can define your own set of fuzzy rules the controller should 'listen to'.

It's mostly a set of IF THEN rules fed by linguistic variables as colder, warmer, hot. The controller output is set by addition the series of (sometimes contradictory) requirements.

See also:

http://www.fuzzytech.com

And demo on:

http://www.iit.nrc.ca/IR_public/fuzzy/fuzzyShower.html

Hope this helps.