Instability in pressure control

You could use a Dome Loaded Pressure Regulator, I installed a number of Kaye & MacDonald Regulators in that type of service in the past .. it acts as a one - to - one repeater ... 5 psi to the dome ... 5 psi regulated etc.

snipped-for-privacy@hotmail.com (Anita Richards) wrote in news: snipped-for-privacy@posting.google.com:

Briefly:

Changing the void space above the liquid level changes the process gain. Note that this is an integrating process.

Make the gain to be proportional to the void space. You tuned with 200 gallons of water in the vessel, the vessel is 500 gallons, so you've got reasonable tuning for a void space of

500 - 200 = 300 gallons. Use gain scheduling:

K = K' * current void space / 300

...where K' = the gain from your earlier tuning test.

If your controller doesn't have gain scheduling as a built- in function, then check if a batch program can change the gain on the fly.

In more detail:

Think of a flow meter and a valve on a line. Open the valve from, say, 0% to 20%. The flow makes a step increase, then stops. Flow is not integrating.

Think of a level indication and a valve on a line going into a vessel. Open the valve from 0% to 20%. The level starts ramping up. The step increase happened on the rate of change of the level. Level is integrating.

Pressure is integrating. (It's also self-stabilizing, so the rate of change tends to level out, but ignore that.)

Consider your pressure vessel when there's no liquid, and the void space is 500 gallons. Crack open the N2 valve until the pressure hits 10 psig, then close it again. Open the vent valve from 0% to

20%. The pressure starts dropping at, say, 0.01 psi/sec.

Let's try that again, only with the vessel almost full of water. Let the void space be only 0.5 gallons this time. Such a small volume will depressure very quickly. The pressure will drop

1000 times as fast.

From the viewpoint of the controller, the process now reacts

1000 times more strongly to an output change (AKA "the process gain is 1000 times higher"). To get the same response as before, the controller has to make moves 1000 times smaller: it has to lower its own gain in proportion.

The process gain had increased by a factor of three and it went unstable.

Not a surprise.

Not a problem under gain scheduling.

Anita,

A suggestion: It is possible to preload some regulators pneumatically so as to vary their set point. Feed the output of your pressure controller to the regulator. Now you have a pressure/pressure cascade with a very fast inner loop and an operator accessible outer loop. This setup is exactly as you had it except that the control valve is replaced with a regulator.

Walter.

Anita,

This is a classic example of controllability. I even demonstrated it to a class using a laboratory version of your process.

The stability of a process is affected by the ratio of the longest lag to the sum of the smaller lags (including dead time) in a process, even though the process gain is not affected. If you calculate gain using the Ziegler-Nichols open loop or any of several reaction curve techniques, you will see that the gain you calculate is based on that ratio.

The smaller lags in the process include a dead time equal to one half of the scan time of the loop, the time it takes to move the valve, and perhaps a lag in the pressure transmitter. These times are not affected by the level.

The predominate dynamic element in the process is the volume in the space above the liquid level. A 500 gal. vessel filled with 200 gal. of liquid leaves a gas space of 300 gallons. Increase the liquid to

400 gal. and the space for N2 is reduced to 100 gallons. Therefore, while the process gain remains the same, the controllability of the process decreases, and you must use a smaller controller gain.

If you can increase the scan rate of the loop, you will reduce the dead time and increase controllability. That may help. You should tune the pressure with the level at its highest, and see if the control is acceptable the lower level. If not, you may be able to adjust the gain based on the level, with a higher gain at lower liquid levels.

I hesitate to suggest replacing hardware; that costs more money. But, if you can replace the pressure valve with a smaller valve with a smaller diaphragm you might be able to reduce the lag caused by the valve reaction time.

Let us know what happens.

John Shaw Process C> I have been working with a 500 gal. stirred vessel. It has an N2

Anita, Based on my work with similar configurations, increasing the sample rate of your loop - or removing time quantisation competely - could well help, if you can do it. For this sort of application, good old pneumatic control with short lines can produce impressive results. A local regulator with an adjustable setpoint is worth investigating.

Most of the other posters have provided a close enough answer.

Really the simplest way you could have done this is to have purchased an I-P or V-P pressure regulator valve from a pneumatics company from someone like Festo, Norgren etc.

These convert a 0 to 10V or 4 to 20mA signal into a pressure. They are widely used. You can get rid of your PID.

As far as you tunning a response problems I would say they relate to the compressibillity of the residual air space in your vessel changing with changes in level and perhaps lags due to sampling rate. You should be able to get around it by tunning for the worst case of adaptive tunning whereby you adjust the PID parameters according to the level in the vessel. Sort of like autopilots that adjusted their PID parameters according to altitude or Indicated Air Speed.

Thanks, everybody, for the help.

John, I checked on your theory about the controllability. I ran some tests, making a step change in the output with the controller in manual and recording the pressure. Sure enough, the process gain (eventual pressure increase divided by output change, was about the same for all liquid levels. However, the shape changed. The first part of the S curve was the same, the second, as the pressure approached the final pressure, was longer for the lower level.

I speeded up the loop scan time. That reduced the effect of level on stability. I then added a gain schedule to reduce the gain at the high levels. That eliminated the problem.

I forgot to add that one of the reasons for keeping the PID in the DCS is that at times the control program, or operator working through the display, will put the loop in manual to close the valve fully and draw a very slight vacuum, at other times the valve will be held fully open. Only during certain operations will we be changing the pressure of an N2 blanket. Also its cheaper just to keep all loops in the DCS since we have to have a direct output to the valve and have to measure and record the pressure.

Thanks for all the help from everyone.

Anita

snipped-for-privacy@jashaw.com (John Shaw) wrote: