Question: Sample Time

[lbussy]

Background:

• I am using this for a slow-acting temperature control system. It will take minutes for the system to react.
• The temperature sensors sample once every 10 seconds (these are DS18B20 sensors so not incredibly granular).
• I am using that to feed a circularbuffer<5> and my temperature reading is the average of that buffer.

tl;dr my &Input is a 60-second average.

Further: My current efforts have QuickPID loop() every X seconds based on Arduino's Ticker.

I suspect that your code is intended to sample temperatures over time, taking action when sampleTimeUs expires. If I'm correct, my circular buffer and Ticker are redundant to the QuickPID functions, and I could eliminate those and just run QuickPid::Compute() in my main::loop(). Right so far?

I "need" the running average to smooth out the UI, so I'll keep that no matter what.

So all that to set up my question: Should I use the average for %Input and QuickPID::SetSampleTimeUs() to something like 10 seconds and call QuickPID::Compute() every second or so? I'm leaning that way (or even longer periods) because of the slow-acting time of the system.

OR

Should I use the current "live" temperature reading, a more frequent loop, and a longer sampleTimeUs time?

[Dlloydev]

I haven't done any testing on a slow acting control system yet, but its good first of all to get the time parameters identified and set correctly. A good thing to estimate firstly is the time constant of your system. Let's say for example that it's 5-minutes (300 seconds). It's suggested elsewhere that the sample time (PID compute period) should be at least 5-10x less than the time constant, so keeping with the example, this could be set to 30 seconds. With sensor readings coming in at 1 reading per 10 seconds, you would get 3 new readings (that are averaged in your circular buffer) for each sample period. I think the sensor rate should be more than 2x the PID compute period so this seems OK.

I took a quick look at the DS18B20 datasheet and it says the resolution can be set to 12-bit, which means you can get about 0.08°F resolution. For 12-bit mode, the max conversion time is 0.75 seconds.

I think using the running average temperature is OK, but you might find better response by using the raw temperature data. I wouldn't set the PID rate so fast that you get less than 2.5 temperature readings per PID compute period. As per the example, you could use 30 seconds: SetSampleTimeUs(30000000)

[lbussy]

Thanks for your thoughts on this. It will be a bit before I have a physical representation of the environment, so I'm sure you understand I'm trying to get close enough to use as a starting point right now.

I just identified a need to use two loops for this application, the main control loop (area being controlled), and a loop controlling the max temperature output of the heating system in order to avoid damage. This should be fun!