PID Autotune Removed from Studio 5000 v33

[AustralIan]

DrBitboy, could you add quantization error?

Like, if you had a 16bit input which was currently rising by 1 bit every scan of the PID, with a gain of 0.4 the proportional term would not move as much in the velocity form as in the... other form.

In the 'positional form':
CVn = Bias + K * (PVn + Quantisation_n)
But in velocity form it is:
CVn = CV0 + K * SUM(PV_n - PV_n-1 + Quantisation_n)

Thanks cardosocea, I never thought about this until today.

 

[drbitboy]

Yes, quantization is a problem, I first ran into it emulating a Honeywell TDC 2000 more decades ago than I care to remember.

So the velocity form requires floating point math, which hopefully pushes the quantization error down to a level where it does not matter.

 

[drbitboy]

, you did well drbitboy. There is NOTHING like being able to derive the formulas yourself. That shows true understanding.

I did it in my head when I first read that document. It's much tougher to write it out in a cogent form for someone else to read.


I still call it bookkeeping.

 

[Peter Nachtwey]

The effects would be the same using either mode but the derivative term could become useless. The sum of Quantization_n averages out but that isn't the real problem. The real problem is that the error between the target and actual velocity multiplied by the derivative gain could cause the output to saturate. In this case the control is severely compromised.



Think about this. The derivative term is actually a gain on the error between the target and actual velocity. Back in the dark ages a Tempsonic rod had a resolution of about 0.001 inches so if the scan was 1 millisecond, the speed would change in 1 inch per second increments. This didn't work so well when ramping up or ramping down. The target velocity would change smoothly but not the actual velocity. This would cause the derivative term to change up and down when ramping. I could do a simulation to make this clear.


Even when traveling at 9.5 inches per second, the feedback resolution was such that in one millisecond the calculated speed was 9 inches per second and the next was 10 inches per second. Since the target speed was 9.5 inches per second, the derivative term always saw a 0.5 inch/second error.