Fuzzy logic and model based control is NOT one of my strong points and I didn't dabble in auto-tuning of the PID loops either, but I manually tuned enough of them. In that respect, fuzzy logic might be easier.
Follow along with the video below to see how to install our site as a web app on your home screen.
Note: This feature may not be available in some browsers.
Just my opinion but I think a digital type control system such as one using Fuzzy Logic is more appropriate for a microprocessor that PI or PID, which is basically using a microprocessor to simulate an analog control loop. You said you did this 20 years ago, likely as an analog loop. The problem with PI(D), as you noted, is that determinng the PID circuit gain and time constants required to prevent oscillations and generate a stable system is difficult, especially if you don't know the system response times and gains, as is the case here. It's easier to determine the needed control parameters in a Fuzzy Logic loop, more obvious how to tweak the loop if it's not working properly, and things like preventing integrator windup are not a concern. The Fuzzy program basically consists of a series of simple If-Then-Else loops, something that a microprocessor readily performs.
I'm suggesting a commercial 5V power supply and a commercial 24 V PS just to prove a point. Tap into the circuit at the proper spots and probably your problems will go away.
Your then left with a few bypass caps and one suggested BW limiting cap. Most of the complaints were power supply related. I'd start there.
I have no idea what the requirements are, but this is the idea: https://www.automationdirect.com/ad...l power supply&utm_campaign=dc+power+supplies if you don't have a lab supply.
Linear supplies are usually out of favor these says.
Bypass caps are a part of good design, but most of the time they don't break one. An EMI filter may be regulatory and basically prevents the PIC oscillator from radiating into the power line and stuff nearby from getting in.
That transistor didn't look right, but I concentrated on the other stuff.
It wouldn't be so bad if you could say it needs a new power supply, would it?
Other stuff, like surge suppressors add cost, but contribute to reliability and insurance issues. It's one think if my french fries aren't done and quite another if the gas valve sticks.
Here are some Fuzzy Logic references:Since I'm new with this I would prefer Fuzzy Logic for an easier approach.
The RTD circuit basically amplifies a reference. The gain is variable and probably non-linear. Then it's amplified again and attenuated. The non-linearity will be an issue.
This **broken link removed** is the simplest thermister circuit, but note it is ratiometric.
You want to quickly determine if the power supply is at fault, so if you need a 5V and a 24 V power supply controlled by a single outlet strip to do so, so what? The point is, it's quick. Do you have to worry about power supply sequencing...don't know.
Just get it out of the way.
Actual solutions come later. A quick look around, this http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=102-1959-6-ND seems promising (500 mA), but the input voltage required has tighter specs. I don't know your power requirements.
The problem is the very high input/output voltage differential and the tricks the designer used to solve it. A pre-regulator that doesn't work, diode drops etc. You also effectively have a "line regulated" power supply for the temperature sensor. Both are bad, but you have to prove it. An LDO regulator is probably not appropriate.
Thermister circuit: I haven't looked at it detail. Went for the bigger fish, first.
Yep, I think you identified a temperature measurement issue and one part of it is electrical.
what do you mean slow?The linear regulator after a switching regulator won't likely work. It's too slow.
If you look at the Ripple Rejection graph on page 7, it shows about -18dB rejection at 1MHz and increases below that frequency (for example -40dB at 100kHz), with a 10µF filter cap on the Adj pin. While not great, it is a significant rejection.I picked the LM317T which, I belive, is similar to the 7805 except adjusttible. If you look at the data sheet and the transient response curves https://www.electro-tech-online.com/custompdfs/2011/07/LM117-2.pdf, you essentially see there is almost no regulation for uS signals. Switching regulators can operate from say 20 kHz to 1 MHz and probably outside that range. Using a linear regulator after a switching supply is like trying to ask an audio amplifier to amplify RF frequencies. It won't happen. The high frequencies just pass right through because the regulator won't respond to the change.
The circuit is more suitable for a thermister interface. Just not sure how your supposed to calibrate it.
Usually you have gain and span controls. Span looks like it's the pot that's in parallel with the sensor and gain is the pot later in the circuit. And yes, they seem interrelated.
I believe it does mean if you measure to the input of the last divider at a fixed temp and set that voltage and then set the divider voltage, it's a two point calibration. The effect of the parallel resistor will help linearize the sensor. You CANNOT, in this circuit, use the table data directly because it's the curve, it's accuracy, the fixed and variable resistor temperature coefficients and the parallel combination of the curve and the fixed resistor.