ControlLogix PIDE Chilled water loop, Long lag time

There is a chilled water cooling loop in our process cooling a roller.
It has been about impossible to tune. this is in a tertiary loop. From the time valve opening it takes about 4 minutes for the slug of cold water to hit the transmitter. Transmitter is at the roll. Valve is about 200 feet away. From the start of a cycle is about 14
minutes from upset to the next upset. Best I can achieve is a 12 Deg. F.
swing in temp. I have tried a very good tuning software
but the settings it provided were so low it over temped the loop.
This is in a ControlLogix using a PIDE block. Flow rate is about 50 gpm
in a 2 inch line. Any ideas??

Christoph said:

I have tried a very good tuning software
but the settings it provided were so low it over temped the loop.
This is in a ControlLogix using a PIDE block.

Click to expand...

A simple PIDE loop probably can't be tuned to meet what you want to do. You need to use a Smith predictor or some advanced control. This about it this way. If you were to take a shower, and it took four minutes for the temperature to change when you made an adjustment, how long would it take you to get the water to the right temperature? If you didn't know about where to put the valve, you would probably move the valve very little at a time, and then wait for 4 minutes to see if you made too much of a change, or need to make more. The PID does not wait for the 4 minutes. It keeps seeing the bad temperature and continues to adjust. If it made the perfect adjustment, it would mess while it waited for the result of the measurement. So the best tuning you'll get is an extremely slow (low P gain, high reset time, no derivative).
Now after you've used the shower a few times, you'll figure out about where to set the valve, and move it to that position in the beginning. Then you'll wait 4 minutes and make another adjustment, or if a couple of minutes later you find that someone else starts the shower in the basement, you'll make another adjustment them. You do this because you make a model of how the shower works, then make the 'big' changes based on your model, then make small changes based on the error between your model and what you measure. This is what the Smith predictor does. It consists of a model of the plant, and a PID controller.

I agree with proof that a process control application where dead time dominates response of final control element will require slow PID tuning to avoid instability or persistent oscillations. And with a reliable model, the Smith predictor is a classical approach to dealing with true dead time when the instrumentation cannot be relocated to eliminate transport delay.

That said, the part of the OP that seems a bit off is the 4 minutes of dead time. By my calculations with the supplied flow/pipe information, the transport delay should be less than minute. If there is truly no response in the process measurement for 4 minutes after a valve move, then I might look elsewhere in troubleshooting. For example, is there valve "stiction" (control signal is changing, but valve is not physically moving because of mechanical resistance) which shows up as dead time, or is the sensor installed in such a way to significantly hinder response (e.g., not in contact with the bottom of a thermowell). These are the kinds of things to rule out before adding complexity to the controller.

With a long lag time I suggest you consider floating control. You could put a time delay between loop executions.

you should take into account the effect of dead time and you should add feed forward controller to the loop

Any chance you can make mechanical changes to the sysrm and add a bypass loop and pump? Then move the valve to be on the outlet of the loop that. Add a thermocouple to the coil inlet and setup a cascade loop.

Move the valve closer to the transmitter. That will reduce the dead time.
Use a Smith Predictor. This requires a model of the system and a delay queue.
Delaying the time between loop executions will not help.
Using a feed forward or bias to estimate the control output is a good suggestion. Then the feed forward estimate supplies most to the control output and the weak gains you computed with the tuning software will just trim the estimate.

I would have some words with the clown that put the valve 4 minutes head of the transmitter and ask he can predict the future.

The Smith Predictor method works well. One of my most popular videos is
https://www.youtube.com/watch?v=uhLMyOlwCoM
The video shows using sliding mode control but a simple PI controller may be good enough.