Continuous weighing with a PID control.

[nonuke]

Why don't you use a flowmeter to control the VFD using PID or Fuzzy from Siemens, and just use the load cells connected to a Siemens Siwarex module just to help you verify that the system is working ok. We have lots of dosing systems working this way.

 

[Banker]

300Kg for bin + granulat at a max rate of 1500kg/h. How much does your bin weighs?

Bin = 100 kg granulate 60 kg max.
I fill the bin when there is 10 kg left and then stop filling
when 60 kg.
If i go under 10 kg i can see the screw feeder speed up. I thing it is because it is not fully covered with granulat.

Banker

 

[robertmee]

Input: wanted (theoretical calculated) 50%, feedback from loadcell (calculated) 47%
Output: PID output value, for example, is 3%. Speed given to drive 53%.

Feedback from loadcell gets better because the drive increases his speed. Calculated feedback from loadcell is now 51%. PID reacts and output value from the PID is now -1%. Speed given to drive is 49%.

Feedback from loadcell is getting worse because the drive decreases his speed. calculated feedback from loadcell is now 45%. PID reacts and output value from the PID is now +5%. Speed given to drive is now 55%.

And so on.

When we miscalulate the theoretical speed, the result is even worse.

It shouldn't be....We'll take the case of no bias for now.

Input: Wanted 50% (theoretical)
Actual Measured: 50%
PID Output: 0%
Drive Output: 50%

Input: Wanted 50% (theoretical)
Actual Measured: 47%
PID Output: 3% (Actually could be anything, and I think this is where your misunderstanding of PIDs lies).
Drive Output: 53% (adding the 3% to your 50%)

Now, the total output to the drive is 53% and the error is 0%.
Input Wanted 50%
Actual Measured: 50%
PID Output: 3%!!! It stays at 3% because it has driven the error to 0. It DOES NOT go back to 0%.
Drive Ouput: STILL 53%

What you are describing is a linear offset, NOT A PID.

Now, the case W/ Bias.

Input: Wanted 50%
Actual Measured 50%
Bias: 50%
PID Error Calc: 0%
PID Ouput (w/ Bias): 50%
Drive Speed: 50%

Input: Wanted 50%
Actual Measured: 47%
Bias: 50%
PID Error Calc: Whatever it takes to get the error to 0, but we'll assume 3% for consitency.
PID Error Calc: 3%
PID Output (w/ Bias): 53%
Drive Speed: 53%

Input: Wanted 50%
Actual Measured: 50%
Bias 50%
PID Error Calc: STILL 3% (the KP and Ki are no longer changing the last equation as the error is ZERO)
PID Output (w/ Bias): 53%
Drive Speed: STILL 53%

Now, you change the setpoint.

Input: Wanted 60%
Actual Measured: Should jump to 60% or so if linear
Bias: 60%
PID Error Calc: STILL 3%
PID Output (w/ Bias): 63%
Drive Speed: 63%



If you do understand the PID operation (I re-read your post, and can't determine one way or another), then the over/under shoot is a matter of tuning and inherent in any PID operation.

 

[Banker]

Why don't you use a flowmeter to control the VFD using PID or Fuzzy from Siemens, and just use the load cells connected to a Siemens Siwarex module just to help you verify that the system is working ok. We have lots of dosing systems working this way.

What kind of flowmeters work for granulate.

Banker

 

[pietervdb]

Thats a refill every 2 minutes at 1500kg/u. What do you do when the refill is active. Stop the calculations and start over again when the refill is finished?

 

[Banker]

Thats a refill every 2 minutes at 1500kg/u. What do you do when the refill is active. Stop the calculations and start over again when the refill is finished?

A refill take from 8 to 10 sec. I this time a hold the PID ( man. mode ) and the calculation.

Banker

 

[pietervdb]

Flowmeters won't work for (precise) gravimetrical dosing.

 

[pietervdb]

Wrong, wrong....We'll take the case of no bias for now.

Input: Wanted 50% (theoretical)
Actual Measured: 50%
PID Output: 0%
Drive Output: 50%

Input: Wanted 50% (theoretical)
Actual Measured: 47%
PID Output: 3% (Actually could be anything, and I think this is where your misunderstanding of PIDs lies).
Drive Output: 53% (adding the 3% to your 50%)

Now, the total output to the drive is 53% and the error is 0%.
Input Wanted 50%
Actual Measured: 50%
PID Output: 3%!!! It stays at 3% because it has driven the error to 0. It DOES NOT go back to 0%.
Drive Ouput: STILL 53%

What you are describing is a linear offset, NOT A PID.

Now, the case W/ Bias.

Input: Wanted 50%
Actual Measured 50%
Bias: 50%
PID Error Calc: 0%
PID Ouput (w/ Bias): 50%
Drive Speed: 50%

Input: Wanted 50%
Actual Measured: 47%
Bias: 50%
PID Error Calc: Whatever it takes to get the error to 0, but we'll assume 3% for consitency.
PID Error Calc: 3%
PID Output (w/ Bias): 53%
Drive Speed: 53%

Input: Wanted 50%
Actual Measured: 50%
Bias 50%
PID Error Calc: STILL 3% (the KP and Ki are no longer changing the last equation as the error is ZERO)
PID Output (w/ Bias): 53%
Drive Speed: STILL 53%

Now, you change the setpoint.

Input: Wanted 60%
Actual Measured: Should jump to 60% or so if linear
Bias: 60%
PID Error Calc: STILL 3%
PID Output (w/ Bias): 63%
Drive Speed: 63%

I used to do it like you are saying but i never got a good result because when i reaches my wanted setpoint the output of the PID drops down. I'm using the Siemens bloc FB41 (CONT_C).

 

[robertmee]

I used to do it like you are saying but i never got a good result because when i reaches my wanted setpoint the output of the PID drops down. I'm using the Siemens bloc FB41 (CONT_C).

Sorry, can't help you too much there, as most of my PID PLC experience has been with TI (now siemens) and AB. I edited my above post as I went back and re-read your post again, and determined maybe you do understand a PID operation, but perhaps are having difficulties with tuning. That may be a matter of the process itself (slow reacting), the PID coefficients, or both, BUT, almost all PIDs will have some degree of hunt (over and under correction). The chore is to get that to as small as possible.

There's always the possibility that a PID is not the right application here. I've done similar operations where the feedforward time was too great (change vs measured result) and so I went with a straight offset routine. In a nutshell:

Take your measurement and compare it to your desired. If not where you want it, make a small incremental change to your drive setpoint. Wait. Measure again. If not there, make another change. Repeat until the desired is within your deadband of acceptable error.

 

[nonuke]

To Banker,
"What kind of flowmeters work for granulate".
I read the first post from Pieter, and he Never mentions if the product is or isn't a granulate so I assumed it was a liquid.

 

[pietervdb]

Biggest problem is that i don't have a continuous feedback. It's always calculated (and delayed).

There will always be an error where the PID reacts.

 

[robertmee]

Biggest problem is that i don't have a continuous feedback. It's always calculated (and delayed).

There will always be an error where the PID reacts.

That's why I'd probably recommend the approach I posted above. Do away with the PID and just go with a timed offset function.