special pid gain schedule to increase PI control stability

[Blokk]

Dear All,

Maybe someone could give me advice about a good PI control technique to increase my system's performance...

I have a heater wire surrounded with thermally semi-conductor material, and a heat-pump outside transferring the produced heat from inside to the environment. I use a voltage-generator to drive the heater wire, and I measure the 4W resistance of a Ni wire as a temperature value. I use PI control from LabView 2010 PID toolkit. The setpoint, and process variable are in Ohms, and the output is in Voltages (between 0-5 Volts).

The system is a calorimeter with the following working principle: using PI control, I reach the setpoint value. I measure the electrical power used to heat the copper wire. I make a measurement without a sample first, and after that I make another with a heat-producer sample (lowest heat production is about mW). By calculating the required heating power difference between the calorimeter+sample and empty calorimeter, we get the heat power what our sample produces (for example a radioactive isotope). So it is a very simple methode.

The setpoint in celsius is about 26 degrees. After a cold start, the calorimeter starts to heat the wire and the surroundings. After approx. 4 hours, the system reaches thermal equilibrium, when every part has a constant temperature, and there is a constant heat flow from inside to outside.

Sorry for the quite long description, but this was needed regarding to my question:

So I want to measure sample heat power via measuring the heating power of the copper wire. So when I get an acceptable deviation around the setpoint (lets say setpoint is 800Ohms of the Ni wire, and the deviation is +-0.001 Ohms), I calculate the heating electrical power by multiplying the heating voltages and the current running through a high precision resistor. So the precision what I can reach with this calorimeter is mainly depends on the PI control fluctuations in the output heating voltages of the PID.vi . I also use a "control input filter" before the PID vi. After the cold start of the calorimeter I use a relatively large P gain. When the setpoint and the thermal equilibrium is reached after a few hours, I get a standard deviation (SD) in the PID.vi output value what results in a +-15 mW SD in the calculated heating power. If I decrease the P gain to half value, I get approx. a half value in the calculated heating power SD, like +-7 mW . And so on, I am still making the experiments, but so far I have reached about +- 2 mW SD with acceptable constant Ni wire resistance.

Do you think is this a good design idea? For example, I could make a kind of P gain scheduling: I check the SD of the last 1000 samples of the process variable, and if it goes under a certain value, I decrease the P gain value of my PID.vi . Of course, there must be somewhere a reachable lower limit, what is decided by the noise and sensitivity of my measuring equipments.

Or got someone a better idea how to increase the sensitivity of such a system?

Thanks very much for any ideas,

Best regards!