DI Water Centrifugal Pump Flow PID control

[Bit_Bucket_07]

I wasn't referring to how fast the PID updates. I was referring to how short the integrator time constant can be and not oscillate. If the integrator time constant can be 3 to 5 times shorter or faster than the time constant for the disturbances then the error should never exceed 1%.

The update time of 100 millisecond seems to be plenty fast enough. I am assuming there is 12 second between divisions on the trend. The current integrator time constant is 10.5 seconds. The disturbance seems to occur over about 6 seconds before the integrator winds up to correct. What if the integrator time constant is set to 2 seconds? Now the integrator may be able to wind up faster than the error can deviate 1%. The valve should be able to change that fast.

Yes, well I wasn't referring to this particular PID's tuning whatsoever, but to the fact that a PID controller (in general) is capable of being tuned to respond quickly enough to react to such extreme transient deviations. That said, generating the appropriate response mathematically means very little if one's control equipment is incapable of responding as quickly as does one's microprocessor based PID algorithm.

 

[Camcanpro]

More recorded data

Thanks for all the help and suggestions.

I recorded the PV,SP and CV value for 20 min when system running, please see attached file. Hope to find more useful info for troubleshooting for the recorded data.

Tomorrow I will get chance to tune two flow loops again. The PID generated by AB auto tuning software can't give good result. I need to generate P gain and I gain ( independent gain) manually. What is the procedures shall I follow to get PI value which finally archive smooth flow control? Can you give me step by step instruction? Thanks again.

 

[daba]

A proven method for tuning (Ziegler-Nichols), starts by increasing P (with I and D at zero), until the PV starts to "hunt" or oscillate. The period of oscillation is then used to calculate the *final* P, I, and D terms to "close the loop" completely.

I have had success with this method, but usually end up "tweaking" the values manually afterwards.

There are many resources on the web, Wikipedia has a good write-up.

 

[Peter Nachtwey]

Thanks for all the help and suggestions.

I recorded the PV,SP and CV value for 20 min when system running, please see attached file. Hope to find more useful info for troubleshooting for the recorded data.

This data says a lot. It is also much different than the previous data.

Tomorrow I will get chance to tune two flow loops again. The PID generated by AB auto tuning software can't give good result.

I am not so sure. The tuning looks pretty good to me. It is the feedback that is noisy but if you were too filter the feedback the tuning would look great. Look at about time 230. The set point changes and the average value of the PV changes quickly to the new level.

There is a HUGE spike at time 200. Does that look real to you? Is that noise or is something down stream causing that? You must find out.

I think you can waste a lot of time trying to tune something that is already tuned.

I need to generate P gain and I gain ( independent gain) manually. What is the procedures shall I follow to get PI value which finally archive smooth flow control? Can you give me step by step instruction? Thanks again.

Try it. I would bet that no matter what you do the results will not change much.


ZN is garbage. If you mentioned ZN on the LinkedIn Automation and Control group it would be flagged as an inappropriate comment. ZN tries to tune a system with no knowledge of the plant. How is Camcanpro going to get any useful data for ZN with all the spikes in the PV?

I know how auto tuners work. The auto tuner will try to fit a model to the data that minimizes the sum of squared errors. Even with all the noise the auto tuner can find the best fit. Here is another example
http://www.designnews.com/document.asp?doc_id=229767&dfpPParams=ind_182,aid_229767&dfpLayout=article
The velocity data I had is very noisy but my model of the system is not. It would be a waste of time to try to tune a system like this to reduce the velocity error to some small error. I am pretty sure the Rockwell guys use the same techniques I do to find a model that basically filters out the feed back noise.

Camcanpro, if you want better results you need to find the cause of the fluctuations. Are they due to the pump, measurement noise, or changing back pressures.

BTW, how are you measuring the flow?

Can the data from that trend be exported to a CSV file? I would look at the relative phase of the PV and the CO to see if the CO is causing the PV to spike or the spike in the PV is causing the CO to change rapidly.

 

[Lancie1]

Are they due to the pump, measurement noise, or changing back pressures?

I bet on changing back pressures.

No one so far seems to care what the downstream load IS. I think Peter is the only other one to even mention it. Why should I be surprised about the same old phenomenon of solving the problem without knowing what the solution should look like.

 

[Peter Nachtwey]

Maybe, but if so, then why?......

I bet on changing back pressures.

No one so far seems to care what the downstream load IS. I think Peter is the only other one to even mention it. Why should I be surprised about the same old phenomenon of solving the problem without knowing what the solution should look like.

Why can'y Cancampro correlate the spikes to something happening down stream like a valve suddenly opening or closing. Those spikes are pretty severe so the ramping the down stream pump or changing a valve slowly wouldn't be the problem.

I doubt that shortening the integrator time constant can remove that noise. I also wonder would would happen if the sample rate where higher in the trend. Would we see more noise? If so then you know darn well that things don't change that fast in a process system on a continual basis.

 

[Camcanpro]

More test and info for troubleshooting

By following Peter's suggestion, I did several tests to collect process data

(1) Camcanpro, if you want better results you need to find the cause of the fluctuations. Are they due to the pump, measurement noise, or changing back pressures.

Today I use similar loop as an example and disconnect the pipe after the flow control valve. The tank size is about 4000 Liter. Water flow out at bottom of the tank and then pass through centrifugal pump, an E+H Magnetic flow meter and and a Spx flow control valve and then go to drain directly. I collect two group of data.
<1> with pump stop,valve at 100% and only use tank pressure ( full of water) to push water out at rate of 6.4 to 7.4 L/min.
<2> with Pump run,valve at 100%,and the flow rate is 15 to 16.3 L/Min.

From the two attached trend graphic, the flow rate is flutuating. Does it mean the flow transmitter has problems?
The magnetic flow meter is 120VAC power supply and I found power cable and signal cable are in one flex conduit to control panel.

(2) BTW, how are you measuring the flow?
We use E+H mass or magnetic flow meter ( Promag H )

(3) Can the data from that trend be exported to a CSV file? I would look at the relative phase of the PV and the CO to see if the CO is causing the PV to spike or the spike in the PV is causing the CO to change rapidly.

Please see attached xls file.

Thanks for continous support.

 

[Bit_Bucket_07]

From the two attached trend graphic, the flow rate is flutuating. Does it mean the flow transmitter has problems? The magnetic flow meter is 120VAC power supply and I found power cable and signal cable are in one flex conduit to control panel.

I suspect that your PV consists of more noise than signal. Were you able to visually inspect the fluid flow? If so, did it appear to actually fluctuate in the manner indicated by the flow meter?

The electrical installation of this flow meter was obviously not done properly. I'd check all of the analog wiring to ascertain whether the system is properly grounded and I'd also look at whether the physical installation of the flow meter was handled as cavalierly as was the electrical installation. This meter is not installed near a 90 degree turn in the piping, is it?

 

[Peter Nachtwey]

Do you have a place where you can put two flow meters in series? It would be interesting to see if they read the same thing at the same time. If not, why not? Is there noise induced on the signal or are the flow sensors simply that noisy?

 

[Bit_Bucket_07]

I've used Endress+Hauser magnetic flowmeters myself, and if they are properly installed then they are not this noisy. I suspect that this meter isn't reading laminar flow within the pipe.

 

[MATT116]

Did I read in an earlier post that your pumping DI water? If so Mag meters usually require the water to be greater than 20µS.

 

[Bit_Bucket_07]

Did I read in an earlier post that your pumping DI water? If so Mag meters usually require the water to be greater than 20µS.

A good point. Truly ion-free water would be transparent to a magnetic flowmeter. I merely assumed that this issue would have been discussed with the instrument vendor. In this application, it appears that nothing should be assumed.

That being said, there are mag flowmeters capable of sensing DI water with a minimum conductivity of 0.008 mS/cm.

 

[MATT116]

Quote from Camcanpro
We use E+H mass or magnetic flow meter ( Promag H )

What Promag H model are you using?
From E+H website Promag H fluid requirements:
H200 >20uS/cm
10H >50uS/cm
23H >50uS/cm
50H >5uS/cm
53H >5uS/cm
55H >5uS/cm

 

[Andybr]

Quote from Camcanpro
We use E+H mass or magnetic flow meter ( Promag H )

What Promag H model are you using?
From E+H website Promag H fluid requirements:
H200 >20uS/cm
10H >50uS/cm
23H >50uS/cm
50H >5uS/cm
53H >5uS/cm
55H >5uS/cm

Actually the minimum conductivity for DeMin Water is always 20us/cm even if lower values are OK for other liquids. This minimum can also increase significantly if you have a remote head when you need to take cable lengths into account. You will need to check the E&H Manual for the values for your meter.

 

[daba]

I've always held the view that mag.flow meters should be in a vertical pipe with the flow going upwards, and placed more than xxx pipe diameters from bends, tees, and anything else that may affect the flow pattern.