Instability on header system, any practical suggestions?

Not quite a whole practical suggestion, but some observations:

I've dealt with this tendency for control loops to interact with their environment (in the mechanical, not the mass-flow realm) by solving for the "impedance" at the interface. In my case it was rotation vs. torque, in your case it'll probably be flow vs. pressure.

I'm not talking about just at DC: I mean you want to be able to make a Bode plot. Look for places where the real part of the impedance is negative: this tells you that in an environment that presents an impedance with a sufficient phase shift, you'll see oscillation. (In my case it was a device which, when mounted on a sufficiently compliant and undamped platform, would oscillate).

Figure it out for your feed to the header (flow into vs. header pressure) and for your five outflows (flow out of vs. header pressure), then add them up and see if inspiration dawns. It could be that you can maintain tuning on your proportional but need to relax the integral, or it could be that relaxing the tuning on just the feeder loop but not the outflow loops would work.

Can you de-tune the feeder valve loop, but use feed-forward from the five outflow loops to get snappy response?

And having said all the above: you know that your simulation isn't adequate or it would have found the problem. Are you sure that it was taking into account the fact that all the loops are "talking" to one another via the header pressure? Getting the simulation working may give you more chances to play with solutions.

Just another observation, thinking about this from first principles:

Unless your header has some good volume of air in it, any change in any of the six valves is going to have an immediate effect on the pressure seen by the other six valves. Presumably the "correct" tuning for such a case would be integrator action only -- but practically that means that with six valves all talking to each other you basically have six integrators in a heated discussion about who's on first.

Again, I dunno if it's any help -- but it does make me wonder why the simulation didn't see problems.

The simulation includes the cross-interactions due to pressure in the header. One of your points I'll follow up, that is a possible vapour lock in the pipe, I've seen that cause problems numerous times in the past. I didn't consider it because the geometry here doesn't seem to allow a vapour lock, but I may not have done a thorough walkround, things like drains and startup lines can harbour non-condensibles as well (actually makes the modelling easier, no tear variables required).

As you say, noncompliance between the model and reality may well be pointing to something physical that we're not seeing. I'll spend some time working on the code.

Feedforward here would be a b***** to implement, due to the valve nonlinearities, it's an approach that I've considered and avoided thus far. You don't know how lucky you are dealing with your motion stuff, it must be a dream dealing with dependable (to a degree) transfer functions.

Anyway, why am I complaining, this sort of problem is the best sort of fun for a control engineer. If I find anything, I'll post some more info in case it helps anyone else.

Thanks for the thoughts, Tim .... Cheers

You need to spend some quality time with something that has gears and a lot of friction.

You control the pressure downstream the control valve. Where do you have pr essure measurements apart from the header volume?

How does the position of the flow control valves change if you compare cont rol valve in manual and in controlled mode? Did you try different control v alve positions? Are the five flow controllers in control mode all the time or did you put them to manual mode, too?

Is the mean pressure value of the oscillation your target?

The control valve is about 30% open. What's the amplitude of the oscillatio n of the control valve?

Try to speed up the centrifugal pump so that your control valve operates in slightly more closed position.

Assume all valves are in a steady-state condition with no oscillations (i.e ., equilibrium), and one of the offtake flow control valves opens because t he load downstream increases and more flow is required.

The increase in flow causes the header flow to increase and the correspondi ng pressure drop goes up, therefore the header pressure goes down. The pres sure controller will try to open the header valve to raise pressure back up again. Meanwhile, the flow has lessened through all of the offtakes due to lower dp across them, and the flow controllers will cause all of the flow valves to open more. As the pressure in the header rises again due to the a ction of the pressure controller, the flow will go up through all of the of ftakes due to the rising dp, and their flow controllers will begin to shut each valve. As the offtake flow decreases, the header pressure will rise ag ain, resulting in the pressure controller beginning to shut the header valv e to lower pressure. A limit cycle is born. This often happens when you are trying to control both the across variable (i.e., pressure) and the throug h variable (i.e., flow) in the same general location, something I try to av oid whenever possible. The loop interaction causes the cycling.

Why are you trying to control the pressure? Are there other loads on the pi pe besides the offtakes you mentioned that require a constant pressure feed ing them? Will the pressure swing significantly, causing flow valves to cyc le, if you don?t control the pressure? If it isn?t critical, you can ge t by with looser tuning on the pressure loop. Some people even put the loop into manual and leave it there.

A quick web search yielded a couple of links that you might want to examine . The first has a section on pressure-flow interactions beginning on page 4

Assume all valves are in a steady-state condition with no oscillations (i.e., equilibrium), and one of the offtake flow control valves opens because the load downstream increases and more flow is required.

The increase in flow causes the header flow to increase and the corresponding pressure drop goes up, therefore the header pressure goes down. The pressure controller will try to open the header valve to raise pressure back up again. Meanwhile, the flow has lessened through all of the offtakes due to lower dp across them, and the flow controllers will cause all of the flow valves to open more. As the pressure in the header rises again due to the action of the pressure controller, the flow will go up through all of the offtakes due to the rising dp, and their flow controllers will begin to shut each valve. As the offtake flow decreases, the header pressure will rise again, resulting in the pressure controller beginning to shut the header valve to lower pressure. A limit cycle is born. This often happens when you are trying to control both the across variable (i.e., pressure) and the through variable (i.e., flow) in the same general location, something I try to avoid whenever possible. The loop interaction causes the cycling.

Why are you trying to control the pressure? Are there other loads on the pipe besides the offtakes you mentioned that require a constant pressure feeding them? Will the pressure swing significantly, causing flow valves to cycle, if you don?t control the pressure? If it isn?t critical, you can get by with looser tuning on the pressure loop. Some people even put the loop into manual and leave it there.

A quick web search yielded a couple of links that you might want to examine. The first has a section on pressure-flow interactions beginning on page 400 and talks about methods of analysis and prevention. The second is for if you really need to control the pressure, and it gives some really good suggestions. It also implicitly suggests you should check how you are measuring each parameter, i.e., make sure you have the proper instrumentation properly installed.

Thanks for going to the trouble, that second reference is really useful. Amazing how specific the info on the web can be, if you know where to look.

The situation here is that the units downstream of the flow valves are running close to serious pressure limits, and when one comes offline, the pressures in the others surge to an undesirable degree (the whole thing is liquid filled so there is no cushioning of responses anywhere). We've been asked to 'make the pressure loop faster', but I'm at the point now where I doubt that will be possible. All the evidence points to a combination of loop interactivity and dynamic influence - in particular the 1-second DCS processing frequency, which we can't speed up.

The answer looks like being to take direct control of the valves when a unit is coming off, and ramp things as required to keep the pressure within limits. That's actually not trivial either, the valves are subject to serious erosion and their characteristics are pretty unpredictable over time. Tim, I note your comment regarding friction and gears, I still think I'd trade my job for yours.

Thanks again to everyone who's come back with comments.