PID help request from Peter

Peter,

I have been following all of your posts since 2009. We (PLCS.NET forum), are all amazed at your acquired empirical knowledge to identify a LOOP characteristic from simple model description, and then offer the best strategical approach to resolve.

We use the Delta RMC products for EXPERT velocity and pressure control of our processes, and tuning is easy with RMC Tools software. Hydraulic positioning and pressure control are easy with your Delta products.

But, currently, we are treading in unfamiliar territory. We are heating a tank of oil media, but there is about a one-hour thermal lag between CV command and PV measured response.

We have a square central tank that the process takes place. The oil media in the main central tank is in direct contact with the manufactured goods to be heated. Thus the final PV (Process Variable).

The central tank is made of Stainless Steel (which has VERY LOW thermal conductivity). The central process tank is surrounded by another Stainless Steel tank of a different heating oil media.

The heating supply is electric resistance strip heaters surrounding the exterior of the outer tank.

What we observed though strip-chart data acquisition that there is about a one-hour delay between heat applied to the strip heaters to a detectable increase in internal tank process PV.

We also observed, once the external media is heated, taking away the CV does not net a balanced decay. Heat applied is sluggish, heat decay is 10 times slower.

We assume aggressive P and I gains will cause us oscillation due to the CV to PV delays.

Many of your posts speak about poles, and other terms that are above my experience.

Can you describe a possible approach to a control loop solution?

Also note, we are not limited to a single PID only loop, but are open to dynamic gain adjustment, if prudent. (Or other Min/Max CV Strategies)

Thank You

Plastic

Don't know if I dare post when you have just asked Peter, but I do have experience of a similar system. First put some insulation over the electric heating otherwise half of your energy will be wasted. Then we split the system in two, with a temperature sensor in the oil in the outer tank and an over temperature safety cutout. The oil in the outer tank didn't need complex control so we have a simple On/Off with hysteresis. Heat transfer is all about surface area, thermal conductivity and temperature difference. We added a stirring system on the inner tank to move the warmed material away from the tank walls, we have some sticky material. We couldn't improve surface area or thermal conductivity but we realised that we could run the outer tank hotter, so we have a temperature boost function until the inner material gets closer to set point. Once the inner material is close to set point we disable the boost and just control the outer tank. There is no chemical reaction so the inner one can't get hotter than the outer one.

Not as complex as the PID system you were hoping for but it gets the stuff to with a couple of degrees of ideal.

It sounds like a Smith Predictor might be the way to go.

Plastic, there are a couple of ways your system can be controlled. destination unknown's suggestion about using a Smith Predictor is a good choice. Since you are using on-off heater strips, sliding mode control with the Smith Predictor is an option. I have a video here. It is one of the popular videos on my Peter Ponders PID YouTube channel
https://www.youtube.com/watch?v=uhLMyOlwCoM


Can you get the temperature up to the set point?
You are right about not making the controller gains too aggressive. One hour long dead time is a huge problem.
Does the ambient temperature change much?



Can you record a response to changes in the control output and post them here? I realize this may take many hours due to the slow response. The time constant(s) for your system seem long so one reading every minute is probably good enough.



Many years ago Ron Beaufort sent me data in this form
"time CO PV"
0.00 10.00 114.00
1.00 10.00 114.00
2.00 10.00 114.00
3.00 10.00 114.00
4.00 10.00 114.00
5.00 10.00 114.00
6.00 10.00 114.00
7.00 10.00 114.00
8.00 10.00 114.00
9.00 10.00 114.00
Ron's data was from his "Hotrod" system which is a soldering iron or wood burning iron. The data was collected every second



I can read this data and compute the system model using the python program I posted a link to a few months ago. Estimating a good model is the key to success because any method that will work will require accounting for the delay time. This will require the use if a FIFO or similar to simulate the delay in the PLC.


It seems to me there must be faster ways to heat up oil.

BryanG said:

Don't know if I dare post when you have just asked Peter, but I do have experience of a similar system. First put some insulation over the electric heating otherwise half of your energy will be wasted. Then we split the system in two, with a temperature sensor in the oil in the outer tank and an over temperature safety cutout. The oil in the outer tank didn't need complex control so we have a simple On/Off with hysteresis. Heat transfer is all about surface area, thermal conductivity and temperature difference. We added a stirring system on the inner tank to move the warmed material away from the tank walls, we have some sticky material. We couldn't improve surface area or thermal conductivity but we realised that we could run the outer tank hotter, so we have a temperature boost function until the inner material gets closer to set point. Once the inner material is close to set point we disable the boost and just control the outer tank. There is no chemical reaction so the inner one can't get hotter than the outer one.

Not as complex as the PID system you were hoping for but it gets the stuff to with a couple of degrees of ideal.

Click to expand...

on off controller sounds ideal, I wonder what sort of accuracy is required?

"..It seems to me there must be faster ways to heat up oil."

I agree. Just DoLight It:doh:


Kidding aside, in general , for a slow response , a Smith predictor or PID controller and for faster response , use a PI controller, right?

BachPhi said:

"..It seems to me there must be faster ways to heat up oil."

I agree. Just DoLight It:doh:

Click to expand...

I would consider an external tank just to heat the oil and then have tubing that would flow into the tank where the piece is kept. The hot oil would be recirculated to the tank where the pieces is and back to where it is heated up.

Kidding aside, in general , for a slow response , a Smith predictor or PID controller and for faster response , use a PI controller, right?

Click to expand...

No, the integrator of the PI would wind up because it wouldn't see the PV changing. The Smith Predictor uses a model predict that the tank will heat up. This keeps the integrator from winding up.


Another option requires knowing the open loop or plant gain. A temperature control plant gain usually has units of degrees/% control output. Each % of control output will increase the steady state temperature by a fixed amount over ambient. It should be possible to predict the control output fairly accurately. Now it is possible to superimpose technique similar to Tom's floating control but the update rate would need to be as long as the dead time. This means corrections would be very slow. If the ambient temperature changes too quickly this method will not keep up unless we know the ambient temperature. Then we can keep updating the feed forward or bias value immediately.



We need more data such has how fast is this process or how fast does it need to be.


First we need the response data.

My bad, I meant it PD not PI for faster response.

Thank You!

While we are waiting for data, it is Quiz Time!

BachPhi said:

My bad, I meant it PD not PI for faster response.

Thank You!

Click to expand...

Yes, a PD controller generally has a faster response than a PI controller.

Why?????

Which is faster? A PD controller where the derivative gain is multiplied by the error between the SP and PV or.... a P-D controller where the derivative term is only in the feedback loop so the derivative gain only acts on changes in the PV?

The -D means the derivative gain only acts on the changes in the PV.
This is a diagram of a P-VA or or P-DD2 controller. Kdf is the proportional gain. Kaf is a second derivative gain but it is only in the feedback loop. The same goes for Kvf which is the derivative gain. Multiplying by s takes the derivative of the input.