Minimalist Controller for FF/IMC

I am looking for recommendations for a small controller to implement a feed-forward or IMC model control strategy.

I have a pretty straight-forward temperature controller for a very sluggish batch reactor. Like most things thermal, it is slow to heat up (20-60C in about 90 min), but even slower to cool off (< 1C/hr).

The existing solution is a simple PI controller with a wide PB that will reach temperature in about 90 min. However, the customer has low tolerance for overshoot (delays transfers) and is impatient. We know that the system can reach (and fly-by) the set point in about 40 minutes at full-power. We also seen that you can run full power for 25 minutes, shut it off and let it coast up pretty close to the setpoint in under an hour (requires a good guess for each load recipe).

I would like to implement an algorithm that tracks the integrated heater input and predicts where it will end up. I think I can get away with two AI, one AO, a set-point adjustment, two DI or buttons for man/auto and reset, and a 2 line display for SP, PV, and future process value. I have the math in hand, but can't find HW that is more flexible than PID, but less involved than a full PLC or working up from a bare microcontroller (i.e., PIC, Arduino, etc.)

Any recommendations?

Thanks,

Bob

are you looking a temperature controller or something else?

I have seen these being used, but from memory the cost was more than a PLC.

http://new.abb.com/products/measure...ers/cm30-1-4-din-universal-process-controller

I need to control temp, but I need to integrate/totalize the output/heater signal to predict where the temperature will end up.

Thanks, AustralIan. The ABB CM-30 looks interesting. I had initially written it off as a strictly PID controller, but it has totalize and cascade functions that might work.

The controller isn't that important. Understanding and technique is crucial.

First off I like the idea of using a feed forward. The feed forward should be the inverse of the plant gain. For instance if the system heats 5 deg/% control output and the ambient temperature is 25 degrees C then heating to 200 deg C would require 200-20/5 = 36% control output. Simple but there is a problem in your case. The system heats too slowly. If the 36% control output were applied it would take 5 time constants to reach within 1% of the temperature set point. From would you said the temperature constant must be very long ( ridiculously,and needs to be verified ) if the system only cools 1 degree/hr. So just applying a FF proportional to the SP-Ambient will take too long.

You can't use an integrator or at least not much of one. The IMC model and formula for the integrator time constant says the integrator say that the integrator time constant should be set to the plant time constant which is ridiculously long to be useful.

So here is the trick. Since the system doesn't lose heat very fast lets assume that all the heat added to the system is not lost. So if it takes 36% control output x 90 minutes to heat the system to set point then why not use 100% control output for a shorter time? Think of the energy added vs energy lost. If none is being lost, very little, then 90minx36% = 3240 %min / 100% = 32.4 min. Then turn off the heater and wait as the temperature raises close to the set point. The same amount of energy is added either way. Now a little heat will be lost so when the PV gets close to the SP turn on on the FF at 36% and the PI control. Now the integrator and P control will add little to the FF to make up for energy lost.

It is all about controlling energy in the form of heat folks. It should be obvious that to maintain a constant temperature the heat add must be equal to the heat lost. I think it will be important to record the amount of energy added as temperature rises so adjustments can be made. Also, ambient temperature changes will affect how the FF will act on its own so changes in ambient temperature will need to be taken into account.

I doubt that 1 deg C/hr value. The time constant must be in hours. It would take 5 time constant to cool to ambient.

Also there is no mention of dead time. The method I state above isn't dependent on knowing that but it would be nice to know.

I ran across this company the other day...can't get much cheaper than this.

http://velocio.net/

I have recently started testing the Click PLC by AutomationDirect for a new project we're working on. I've been pleasantly surprised. Very cheap, very easy to program. A lot better than their DirectLogic line. I'd put it on par with a Micrologix or SLC5 by way of power, features, and ease of use, but the cost is a LOT less. Unfortunately it doesn't have specific PID instructions, but the Math instruction is quite flexible, so you could roll your own fairly easily if you have the desired formulas in-hand already.

Good luck!

very expensive
previous brand it call Taylor if i not mistake.

http://www.micmod.com/mod-30ml-multiloop-controllers