Open Loop Tuning using Ziegler Nichols

[bigbang]

I'm trying to tune a difficult temperature loop with a long response time (around 30 minutes) and thought I'd go back to proven methods rather than using the "suck it and see" approach.

Only problem is I'm confusing myself trying to determine the process gain which equals the %Change in Process Value/% Change in Manipulated Value.

How do you determine the % Change in Process Value. What is the change a percentage of?

For example if the temperature changes from 95Deg C to 105 Deg C with an output change of 35 to 42 % what is the process gain.

 

[xnuke]

You need to know the full range of the instrument and actuator to find the percent change. For example, let's say you're measuring your temperature with an instrument range of 0 - 200 C and you're manipulating a control valve that can operate 0 - 100% open. If you put in a 7% change in valve position and you see a 10 C change in temperature, you have a change of 5% (10 C/ 200 C * 100%) in the temperature for your 7% change in valve position, so you have a process gain of 5/7.

Out of curiosity, why is the system difficult to tune?

xnuke
"Live and act within the limit of your knowledge and keep expanding it to the limit of your life." Ayn Rand, Atlas Shrugged.
Please see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[bigbang]

Thanks for that.

I'm finding it difficult mainly due to the slowness with a period of around 40 minutes , my impatience and lack of tuning experience with these slow loops.

I thought I had it nailed the other day with classic 1/4 amplitude dampening after a 6 Deg C setpoint change(94 to 100 Deg C). However soon after there was a fairly big change in a process flow which sent the output ballistic which was unacceptable to the operators.

My next thought is to introduce a feedfoward signal from the flow that caused the upset.

 

[controlnovice]

Feedforward may help with the process changes.

For difficult temperature control, try playing with the derivative as well.



______________________________________________________________________________
This is normally the space where people post something insightful.

 

[ScottyUK]

With such a long dead time you can afford to use a smoothing function on the input with (say) a sample time of 1 second and averaging the current and previous (say) nine samples. This will reduce the effects of any noise on the sensor while allowing you to use more derivative action without creating a jumpy response from the final control element (valve?).

When tuning loops with long dead time or very slow response, start as you have but make the steps in setpoint larger as you start to approach optimum loop tuning. Leave the derivative out initially - get the loop response as good as you can with just PI then introduce derivative action.


----------------------------------

Sometimes I only open my mouth to swap feet...

 

[ADMIRALKEN]

Have a look at this site:

http://www.controlguru.com/

There's a lot of good loop tuning information there.

 

[HEC]

Bigbang

you have initially tuned for a process condition, flow/temperature ie process load. In your second post "fairly big change in a process flow which sent the output ballistic" this is a change in the process load i.e. flow. This will require very diferent tuning parameters compared to teh steady state process. Consider cascaded control loops from the flow or inlet temperature to provide a degree od feed forward control.

Mark Hutton