From 12:00 AM CDT Sunday, October 17 - 11:30 AM CDT Sunday, October 17, ni.com will be undergoing system upgrades that may result in temporary service interruption.
We appreciate your patience as we improve our online experience.
We appreciate your patience as we improve our online experience.
I would like to ask for advice from people experienced in control theory.
Recently I work on a very interesting device, it is called inertial guidance vacuum calorimeter (this is an isothermal calorimeter type). I make a completely new LabView program for this device, and I try to increase its stability/precision as much as I can. This calorimeter can measure sample heat powers down to the micro-watt level. The principle of the measurement is the following (please see the attached schematic):
we want to measure the heat flowing from the sample holder to outside, this is done via Peltier-sensor (thermoelectric sensor). To get a valid data, the temperature has to be very precisely constant everywhere in the calorimeter (isothermal method). The inner parts of the calorimeter thermally shielded from the ambient with high vacuum and radiation shields. The double walled vacuum chamber circulates water through a heat exchanger, this is the heat reservoir.
To have the thermal stability, there are 3 controlled loops:
Control loop 1.: this controls using PID the water temperature to be constant (conventional platinum resistor 4W method)
Control loop 2.: this is the first step stabilizing the inner parts' temperature, PID loop using conventional platinum resistor, and drives a Peltier-heat pump between the support and the "base". This can reach only a temperature stability around 0.1 mK.
Control loop 3.: When the ultimate stability is reached with control loop 2, it is turned off, and its driving current output set to constant. Now the loop3 starts, and this is the tricky part: there is a cylindrical heavy copper "inertial mass" with high heat capacity (much higher then the base's) standing on the base. Since it has a high heat capacity, even the very tiny temperature fluctuations in the temperature controlled base yield a measurable voltage in the thermoelectric sensor between the inertial mass and the base. This value is measured with nanovoltage meter (Keithley). So I use this measured value to fine control the heat pumps below the base. In principle, if there is no heat flowing between the inertial mass and the base, we are in thermal equilibrium.
With this concept it is possible the reach temperature stability well below nanoKelvin!
Recently I play with the control parameters, but only using simple PID controls (I have the PID toolkit). The above explained method is a kind of analogy of the inertial guidance control in mechanics. I wonder, if someone could give me advice or direction how I could improve even more the stability of this control. Maybe some more advanced control scheme? Feed-forward?
Thank you very much for advices,
Solved! Go to Solution.
Very interesting instrument!
Before you try to improve the "stability" I think you need to identify the reasons for the instability. For example if the jacket temperature varies a few mK from top to bottom (which does not seem too unlikely), does this produce a measurable effect? What is the temperature uniformity of the "inertial mass" (perhaps better called "thermal mass")? Is your system cylindrically symmetric with the calorimeter cup in the axis? How uniform are the heat flux and temperature distributions due to the Peltier heat pump? (Is this a continuous device or an array of discrete elements distributed around the circumference of the base?)
How fast does the sample heat flux change?
What are the noise limits on your various measurements? What is your control resolution on the temperature controllers?
Thanks, you point toward important things. Some of them I know already: the standard deviation of the water jacket is around half mKelvin. I attached a screen-shot, so you can see the standard deviations of all of the parts (SD=standard deviation) The most important value is the calorimeter signal as you see on the pdf screen-shot, now I have about 50 nanoVolts standard deviation for this. Actually I have already improved the stability of this more then 10 years old device: with one order of degree, but I would push the barrier more on 🙂
I use point-by-point standard deviation VI to continuously monitor the stability, and there are predefined SD limits so the system can automatically change state: water loop equilibrium -> coarse loop equ. -> fine loop equ. -> calorimeter signal equilibrium). To answer the other question, yes, the schematic is a cross section, the device has a cylindrical symmetry.
Well, I guess I should make a kind of list: collect all of the factors what effect stability, and I should find lets say the "bottle neck" of the performance. If I could approximate somehow the ultimate reachable stability what comes from hardware, I would know that if I need to play more with the PID control, or maybe I am already at the barrier...?
(Another note for the fine loop control set-point: it is not zero, since there are some parasitic voltages even at equilibrium over the thermoelectric sensor.)
And yes, the "intertial thermal mass" would be a better name 🙂
Thanks for responding,
ps.: one more thing: recently I run the loops with 1 second rate, the GPIB readings take about 0.6 second. Should I try to be "faster" for my control? The TDS signal is quite "quick" in time...
If I followed everything correctly, the feedback for control loop 3 is measured using the Keithley via GPIB. And those readings take about 0.6 second. The control loop cannot work faster than the measurement system (unless you want a lot more instability!). Most high precision instruments offer a trade-off between resolution and speed. I took a quick look at the 2182 specs and it looks like you may need all the resolution, so speeding up may not be an option.
The bottle neck approach is probably a good one with all you have done so far. It will help you avoid putting too much time into something which will not make much improvement.
Another thought: How noisy/drifty are the power supplies/amplifiers driving the Peltier devices? I worked on a system several years ago where the target was ~10 uK stability and noise and drift in the power circuits were major limitations. Most engineers designing such devices never think about parts per million or fractions thereof. Especially with slow feedback from the nanovoltmeter, drift or noise in the power circuts could be significant. Also look at how constant the outer loops are in open loop constant output mode.
Thanks for the advices! It is good to get some feedback, "more eyes can see more".
Well, the fine loop drives a Keithley 2400 sourcemeter, the sourcemeter is in the 1 mA range, the PID range for fine loop is: +-1mAmpere.
This is from the Keithley manual:
Source accuracy is calculated similarly, except source specifications are used. As
an example of how to calculate the actual source output limits, assume that you
are sourcing 0.7mA on the 1mA source range. You can compute the reading limit
range from source current one-year accuracy specifications as follows:
Accuracy = ±(0.034% of output + 200nA offset)
= ±[(0.034% x 0.7mA) + 200nA)]
= ±(238nA + 200nA)
In this case, the actual current output range is 0.7mA ±438nA or from 0.69956mA to 0.70044mA.
I accept your answere as a solution, since I was looking for some advices here,
I am glad to be of some help. Projects like that are definitely ones where a change of perspective occasionally can be useful.