Surge Tank Level Loop Cycling After Pump Change
A surge tank level is controlled by a PID loop that modulates a 100 mm butterfly valve on the outlet. The transmitter range is 0–4.0 m and the setpoint is 2.0 m. After a pump speed change, level cycles between 1.5 and 2.5 m with a period of about 22 minutes. The controller output is smooth and sinusoidal, and the valve position follows the output with no obvious stick-slip. A bump test with the controller in manual showed that a 10% step in output moved level 0.80 m; the response reached 63% in 6 minutes and showed about 1 minute of dead time. The controller gain is currently 40 %/m, integral time is 1.5 minutes, and derivative is 0. The valve is fail-closed and the level is critical; operators have been adjusting the setpoint to stop alarms.
1. Using the evidence, decide whether the cycle is more likely caused by controller tuning, valve stiction, or a process upset. Show your reasoning.
Read the worked answer
The controller output is smooth and sinusoidal, and the valve follows it without obvious stick-slip. Valve stiction usually produces a flat or stepped output while the valve stays stuck, followed by a sudden jump; that pattern is not described. The 22-minute period is also much longer than the 1-minute dead time, which is typical of an unstable feedback loop rather than a fast valve limit cycle.
The bump test gives a process gain of Kp = 0.80 m / 10% = 0.08 m/%. The present controller gain is 40 %/m, so the loop gain is: Kc × Kp = 40 × 0.08 = 3.2 A loop gain well above 1, combined with an integral time of 1.5 minutes versus a 6-minute process time constant, is aggressively tuned. The cycle is most likely a tuning problem, but confirm with a valve signature and check whether the pump change altered the process gain.
2. Calculate conservative PI settings using an IMC/lambda approach with lambda = 3 minutes. State the model assumptions.
Read the worked answer
Use the bump-test FOPDT model: process gain Kp = 0.08 m/%, time constant tau = 6 min, and dead time theta = 1 min. For a PI controller, a common IMC rule is: Kc = tau / [Kp × (lambda + theta)] Ti = tau
With lambda = 3 min: Kc = 6 / [0.08 × (3 + 1)] = 6 / 0.32 = 18.75 %/m Ti = 6 min Derivative = 0
This reduces the gain from 40 to about 19 %/m and slows the integral action from 1.5 to 6 minutes. Lambda = 3 min gives a reasonable margin over the 1-minute dead time. If the level is safety-critical and the process is noisy, a larger lambda, such as 5 minutes, would be more conservative and would give a lower gain. Because level loops are often integrating rather than self-regulating, verify the model against a second bump test before finalizing.
3. How would you implement and validate the new tuning safely?
Read the worked answer
Place the controller in manual, confirm the valve fail-safe action, and verify the level is within safe operating limits. Enter the new gain and integral time, then transfer to auto bumplessly. Make a small setpoint change and trend PV, controller output, and actual valve position. If the response is too slow, reduce lambda slightly; if it is oscillatory, increase lambda or reduce gain. Do not add derivative unless the measurement is clean and the process justifies it.
Check anti-reset windup, output limits, and alarms. Confirm the valve is not sticking or suffering from deadband, because a mechanical problem can mimic poor tuning. Coordinate with operations, avoid tuning during a critical batch, and document the old and new settings. Retest at the normal operating level and flow before leaving the loop in automatic.