Implementing feedforward control in heated box

Question

I'm making a temperature-controlled box. Issue I'm having is that it overshoots and undershoots a lot because it takes a long time between turning the heater on/off and seeing a difference in temperature in the box. I have a SHT30 sensor with 0.1°C accuracy measuring the ambient temperature.

I've reached the conclusion that I need feedforward control, so I've calibrated a NTC thermistor and put it on the heater. The temperature of the heater is my feedforward signal.

...now what?

I checked online and with ChatGPT how one would go about this, but I see nothing useful.

Basically, at 50°C, the box is just below its max temperature, and I want the feedforward to contribute very little to decreasing the PID output. At 30°C, the box is close to room temperature, and I want the feedforward to contribute more to decreasing the PID output. At 30°C setpoint, there are oscillations of +/-2.5°C. Closer to 50°C, these oscillations are less than 1°C, which is fine. My goal is to have all oscillations for any setpoint temperature to be less than 1°C.

The peaks of the overshoots and undershoots are maybe 5 minutes apart. It's only a 20W resistive heater (with a fan) for a 10L insulated box, so changes to both heater and air temperature are very slow.

(The reason why I'm not considering changing my PI tuning is because it has the correct idea of putting output to max right away. It takes ages to heat up the box as is, so I really don't like the idea of having very small PI values.)

Answer 1

The best way to do temp control (I've done temp control for 10 years and been able to do controllers/systems that do microdegrees) is to use a PID. There also needs to be a decent mechanical/thermal setup.

For adequate thermal control there needs to be enough control authority in the heater to change the temperature and hold it. One way you can find this out is run the heater at full power (if it won't damage anything) and see what the end temp is compared with ambient. If the heater can't warm the box up more than a few degrees then you will need more insulation, it would be good to have a heater that could warm 30C or more above ambient. Insulation is always a good idea to keep room temp swings (which can be 2 to 10C) out of the box, the more the better.

The sensor (NTC in this case) also needs to be located close to the heater but not so close that it causes instability. A fan may also be needed to be able to avoid convection from the heater and keep the entire box temp homogeneous (there will be globs of hot air in a still air environment that will come off the heater and cause issues for temp control). A good way to do this would be to use a heatsink on the heater with a fan on it and place the NTC anywhere from a 1/2" to an inch away from the fan toward the edges.

The accuracy of the setpoint will be determined by the NTC, you will need an sensor system that is not noisy and not thermally noisy, most NTC sensors and electronics should be able to maintain 1C.

Answer 2

If you need to control temperature in the box, then sense temperature in the box, not on the heater. Then PID can solve your problem, but you have to find adequate P,I,D parameters (in your case D solves "delays" or "thermal mass" problems). If you can sense only temperature of heater and outside box (and not enviroment inside box), then you can measure thermal parameters of your box (box-ambient thermal resistance eventualy heater-box resistance and heater and box thermal mass) and calculate temperature in box by model.