We are controlling discharge pressure from a reciprocating compressor by recirculating compressed and cooled gas through a control valve. The valve is a cage style with linear trim and a diaphragm-spring actuator, which opens on air failure. We are not satisfied with our pressure control. If we tune the controller for fast load response, recovery from upsets is satisfactory, but the occasional large upset can cause the loop to swing until the valve is driven from limit to limit -- manual intervention is needed to stop the cycling. We have traced this behavior to the slow stroking time of the valve -- 15 sec. for full stoke.
So, we inserted a volume booster between the pneumatic positioner and the valve motor. At first, this reduced the stroking time to 2 sec., but motion was very erratic. The booster has a bypass needle valve, which we adjusted to stabilize the motion and this left the stroking time at 4 sec.. We now, however, see a slow, triangular cycle in the pressure loop. Reducing the controller gain reduces its amplitude and extends its period, but at the cost of poorer load response. What's causing the cycle and how do we fix it?
-- From November's Chemical Processing
Remove the booster
F. Greg Shinskey, process control consultant
North Sandwich, N.H.
Consider a surge tank
Joe C. Ranshaw, E/I engineering specialist
Solutia, Anniston, Ala.
Use an unloader
Paul E. Noges, staff engineer
U.S. Oil & Refining Co., Tacoma, Wash.
Integral time is too fast
The integral time must be larger than the product of the % upset, controller gain, and time to stroke, that is: Ti > Eo x Kc x Tv/100%
Installing a booster added a dead band that will cause cycling in a gas pressure loop. The pneumatic positioner has few tuning options. You are lucky if it still has held its calibration.
A better solution would be a smart digital positioner with advanced tuning capability and if necessary, one of the better doubling-acting pistons that are designed to respond to small changes in signal almost as well as the diaphragm actuator. Either way, you must still make sure the integral time is not too fast compared to whatever stroking time you end up with.
Greg McMillan, professor
Washington University, St. Louis
Is the downstream system high gain?
You don't indicate what the downstream system looks like. It sounds like it might be quite high gain, having rapid changes, which you are controlling with a valve with slow dynamics. Have you tried integral control action to recover faster from large upsets? Can you install a pressure receiver to smooth out rapid system changes? If this is a sizeable compressor, a variable-frequency drive should offer a very good payout, since a significant letdown and recycle stream is certainly an energy waster.
Matt Beasley, sr. process engineer
GEO Specialty Chemicals, Cedartown, Ga.
Look for a limit
The only way I can think of to get a true triangular cycle is to have some kind of discrete limit in the control algorithm or the system. Normally, a control oscillation should be a sine wave unless the process runs into limits. If there actually is a triangle wave, you need to determine what limits are causing this, either in the control loop or mechanically. Look hard for a limit, that's a huge clue.
Improving the control valve stroke by adding a volume booster is ok and 10 years ago was a very good idea. Conventional positioners would cause instability in a system that had to be very fast; so they were avoided like the plague in critical service. Instead, we would use volume boosters just as you describe to get the valves to move quickly. Lately, though, we have become big fans of the newer digital positioners. They have allowed us to eliminate the boosters on these sensitive services. The results are impressive, and we have been using them for control loops that need to be real fast. Besides moving the valve fast, you want to move it accurately and the newer positioners give you this. We have seen an improvement in both speed and accuracy of the valve position. This is somewhat contrary to logic, as you would expect an analog positioner to be more stable as you do not introduce the steps that occur in a digital system.
I'm not sure a linear trim makes sense. It suits situations where the upstream pressure and downstream pressure do not change greatly when the valve moves, i.e., the pressure drop across the valve is a relatively small portion of the available pressure drop. When the pressure drop changes appreciably across the valve, an equal-percentage trim normally makes more sense. This would probably have nothing to do with the triangle wave unless the valve was hitting a mechanical limit.
As far as an upset causing the control algorithm to drive the valve limit to limit, there normally will be a gain in every system that can accomplish this. It just gets easier when you can add things like slow control response, e.g., the slow valve stroke. The more lag time you can take out of a system, the harder it is to get it to oscillate. Given that this is a gas system, you never will get rid of the capacitance caused by the compression and expansion of the gas; so, it has some lag built into it. To expect it to never oscillate or expect to get it so that it cannot oscillate is probably unreasonable for all process disturbance cases. Again, this should not cause a triangular oscillation but a sine wave.
Jason L. Noe, process specialist
UOP, Des Plaines, Ill.
Follow a three-step procedure
I assume that the cycle in the loop is not the same frequency as the piston frequency of the compressor; if it is this simple, dampen the pressure transmitter so you don't see the compressor cycle.
I would follow the sequence below and quit at the step that solves the problem:
1. Is the valve "tight?" Play between the positioner and the valve can be the culprit. If you are going to Step 2, make the valve as fast as you can.
2. Install a flow meter in the recycle, cascade the flow set point with a pressure control loop. With proper tuning, the valve characteristics are lost in the flow control loop, which should be tuned fast with PI. Tune the cascade as a slow loop with PID heavy on the derivative. If you use auto tune, apply it only on the first loop; manually tune the second loop.
3. Use multiple PID values in the cascade loop depending upon the deviation from the set point. Close to the set point, the loop is slow; far from the set point, the loop is fast. Another variation on the multiple PID is to select PID values based on rate of change of the input.
Bill Wood, P.E.
Texas Brine Co., Houston, Texas
Check for a sticky valve
It sounds like a sticky valve is causing the triangular results. The valve moves a bit too much and sticks. Then the reverse takes place. Check the valve action and then replace that pneumatic positioner with a new digital one with a greater air volume.
Larry Richardson, technical specialist
UOP, Des Plaines, Ill.
Get back to basics
Does the controller have integral or derivative control and are they being used? Is the valve sized correctly? How much is the valve opening and closing for each cycle? Is valve movement causing a pressure change or is pressure change causing the valve to move? How much must the valve move to see a change in pressure? Essentially, I am asking about the gain of the valve. If the gain is too small or too big, the valve is the wrong size. The slow triangular response in the loop sounds like the PV is ahead of the control valve. If you are using PI rather than derivative control, then adding derivative control may slow or speed up the valve response enough for the process to respond. Other items to check are the calibration of the valve and controller output.
Rick Hildebrand, engineer
Amgen, Thousand Oaks, Calif.