Level loop cycling on a CIP balance tank - tuning, stiction or measurement?
A beverage bottling plant in Tsuen Wan reports a persistent level oscillation on a 10 m3 CIP balance tank. Level is measured by a 24 GHz radar transmitter (4-20 mA, 2 s damping) and controlled by an S7-1200 PID driving a 2-inch equal-percentage globe valve with a smart electro-pneumatic positioner. The tank level cycles plus or minus 6 % around setpoint with a period of about 35 s. Operations confirm the amplitude is identical whether the line runs at 12 000 or 20 000 bottles per hour, and that the level is steady when the controller is in manual at a fixed 30 % output. The loop was re-tuned last month: increasing the integral time from 20 s to 60 s lengthened the period to about 50 s without reducing the amplitude, and halving the gain did much the same. Trend review shows the controller output moving in steps of roughly 4 % separated by flat dwells of 8-12 s. The valve Cv is 63, while the maximum required flow needs only 22 % of it. The radar has been checked against a sight glass.
1. The plant manager wants the loop re-tuned again. Using the evidence, identify the most probable root cause and explain how each observation supports or contradicts your diagnosis, including why the loop is stable in manual.
Read the worked answer
Most probable cause: valve stiction (stick-slip) in the final element, aggravated by a grossly oversized trim. The controller, the radar and the process are not the problem.
Evidence and what each item rules in or out:
- Stable in manual. A valve that sticks holds a fixed position, so no disturbance is injected. A genuine process upset or a noisy measurement would still show in manual. This isolates the fault to the closed loop, and specifically to a non-linearity inside it.
- Amplitude independent of throughput. A tuning-driven instability scales with process gain and load; a friction-driven limit cycle is set by the stick band and is largely load-independent. This points away from tuning.
- Re-tuning changed the frequency but not the amplitude. With stick-slip, reducing gain or lengthening the integral time slows the rate at which the controller output ramps, so the period lengthens - exactly what was observed. The amplitude stays tied to the stick band. A well-tuned linear loop simply cannot produce a fixed-amplitude cycle.
- Output in 4 % steps with 8-12 s dwells. This is the signature: the controller integrates error until the demanded change exceeds static friction, then the valve breaks free and jumps, overshooting because kinetic friction is lower.
- Valve at about 22 % of required Cv. Normal operation sits at 20-25 % of travel on an equal-percentage trim, near the seat, where the friction-to-thrust ratio is worst and positioner resolution is poorest.
Confirming test: place the controller in manual and step the output in 5 % increments, trending the positioner travel feedback against the output. Travel that lags 3-4 % and then jumps confirms stiction. Travel that follows the output exactly moves suspicion to positioner tuning or the process. A 0-100-0 hysteresis test in 5 % steps gives the stick band as a number you can compare before and after repair.
2. Set out the corrective actions in priority order, stating for each the expected effect and the risk if it is skipped.
Read the worked answer
Overhaul the valve first (packing, stem, guide, seat, actuator springs). Re-torque or replace the packing to the manufacturer's friction specification, check stem straightness and guide clearance. Effect: removes the stick band. Skip it and any tuning or positioner work is compensating for a mechanical defect that will return.
Verify the actuator and bench set. Confirm the spring range matches the positioner output range, measure the air supply at the actuator while stroking (a sagging supply under demand is a common cause of a slow, sticky stroke), and check the fail-safe direction against the loop's safety requirement. Effect: ensures the thrust available exceeds the packing friction with margin. Skip it and the valve may still stick at the top of the stroke only, which is harder to diagnose later.
Right-size the trim. Required Cv is roughly 14 (22 % of 63). Fit a Cv 16-20 trim so normal flow sits at 50-70 % open, away from the seat. Effect: restores controllability and linearity at normal load. Skipping this is the most likely reason the fault recurs, because the valve still operates in its worst region.
Re-commission the smart positioner. Run its auto-tune, set the zero and span, reduce the deadband only after the mechanics are correct, and export the diagnostic baseline. Effect: gives reliable travel feedback for the verification trending.
Re-tune last, using an integrating-loop method (lambda tuning, closed-loop time constant about three times the dominant lag). Do not attempt to tune out stiction: a de-tuned loop only lengthens the period, which is precisely what the previous attempt achieved.
Consider removing the non-linearity altogether - a variable-speed drive on the transfer pump with the valve used only for isolation - if the plant can accept the change. Effect: eliminates throttle loss and the friction problem, and saves energy.
3. What records would you require before and after the repair to demonstrate the loop is fixed, and what would you change so the problem does not recur?
Read the worked answer
Before and after, at the same production rate and the same setpoint:
- A 24-hour trend at 1 s resolution of PV, SP, OP and valve travel feedback, exported from the historian, not screenshots.
- A 0-100-0 hysteresis test sheet in 5 % steps with as-found and as-left values, signed by the technician and witnessed.
- Positioner diagnostic export (friction index, travel deviation, deadband) taken before and after the overhaul.
- A 5 % open-loop step test in manual with the travel feedback overlaid on the output, to quantify the residual stick band.
Acceptance criteria stated up front: PV within plus or minus 1.5 % of setpoint, OP moving in increments smaller than 1 %, no sustained oscillation in the OP autocorrelation, and the standard deviation of PV reduced by an agreed factor.
Changes to prevent recurrence:
- Put the valve on a critical-loop register with a quarterly hysteresis check, not just a generic annual 'functioning' check on the CMMS.
- Record the Cv, range, spring range, fail-safe action and required stroking time in the loop description so the next engineer can see that the valve is oversized before re-tuning anything.
- Amend the loop-check and SAT procedure to require an open-loop step test with travel feedback, so stiction is caught at commissioning rather than after a month of production.
- Train technicians to distinguish 'loop unstable' from 'valve sticking', and make re-tuning require a documented step test as justification.
- If the valve is retained, add it to the spares holding with a service kit, since a repeat overhaul will be needed.