Hi Steve,

Some ideas come to mind, but you may be on your own to implement them due to the specialized nature of your thermal system.

(1) Use your standard PID. The performance gain you have seen (from 15 minutes to 7 minutes) may be as a result of the PID parameters being near optimal for the heat pumping capability presently available. In other words, keep tuning. If you have reached the best attainable response with the PID and need better perfomance, then move on to one of the following techniques.

(2) Use an open loop heat/cool mode (with appropriae safeties) until the actual temperature is within a to be determined range of the desired temperature, then turn on the PID and let it do the final trim. The maximum available heat/cool pumping capability WILL be used before transitioning to the PID. Using this method, you'll want to identify a value (or range of values based on load) with which to initialize the integrator when transitioning to the PID. The derivative term can be initialized by observing the rate of change of temperature during the time just before switching to the PID. I used this method on a pneumatic system and went from a settling time of twenty minutes down to three minutes.

(3) Use a feedforward technique termed "Posicast". The term is taken from the action used by a fly fisherman to place the fly at a precise spot. In control sytem language, an initial value of setpoint is chosen to cause an over/undershoot in the controlled value. The maximum/minimum value of the over/undershoot will eventually be the desired settling point. At an 'appropriate' time the setpoint is changed a second time to cause the controlled value to fall precisely on the maximum/minimum value of the over/undershoot. A bit of system modeling and analysis is needed to implement this without a fair amount of testing. It's also a method to be used in conjunction with a PID since it does not account for variation in system parameters that occur with age, ambient conditions, etc. Here's a demonstrative link -

http://code.eng.buffalo.edu/tdf/article01.html

(4) Optimal control and non-linear optimization methods that assign a cost function for (1) deviation from desired states & (2) expenditure of control effort, and allow for limits of states and control effort (or energy) are also available, but require a good deal of modeling effort on the designer's part.

Cheers,

Bruce