Reflux Drum Level Loop: Stiction or Tuning?
A depropanizer reflux drum (1.8 m ID, 2.4 m tangent-to-tangent) splits reflux from distillate with a level controller. At normal rates the drum's residence time is about 2.5 minutes, so the level behaves as a fast integrator. For three weeks the level has cycled ±7% around setpoint with a period near 3 minutes; operators keep the loop in automatic, but the reflux pump cavitates at the bottom of each cycle. A trend capture shows the controller output as a series of slow ramps followed by abrupt 1.5–3% jumps, with the output stationary for 40–60 seconds at the top and bottom of every cycle. The DCS was replaced two months ago; the PID is gain 2.0, reset 0.6 minutes. The 4-inch globe valve has a pneumatic positioner and was last serviced four years ago. Maintenance proposes cutting the gain to 0.8.
1. The maintenance supervisor says the loop is simply tuned too aggressively. Using the trend signature, decide whether the oscillation source is the controller or the final element, and describe the field test that confirms it along with the pass/fail criterion you would apply.
Read the worked answer
Diagnosis: the final element, not the tuning.
The discriminator is the shape of the output, not the amplitude of the PV.
- In a gain-induced closed-loop instability, the OP is a continuous, smooth sinusoid. The controller is always computing a new output; the output never parks.
- Here the OP ramps, then sits flat for 40–60 s, then jumps 1.5–3%. That is stick-slip: the controller ramps the output, static friction holds the plug, the accumulated force releases suddenly, and the PV follows. No tuning change can produce flat spots in the output.
- A 3-minute period is the right order of magnitude for this loop's ultimate period (~4× the valve-plus-transport dead time of roughly 40 s), so the loop is near its stability limit - but the stiction is what is sustaining the cycle, not the gain.
Confirming field test
- Steady the plant; put the level controller in MANUAL.
- At the normal operating point, step the output in 0.5% increments every 30 s, in both directions.
- Record actual stem travel (positioner feedback, or a strap-on displacement gauge if the positioner is not trustworthy) against command.
The command change required before the stem moves is the total dead band plus backlash.
Acceptance criterion
For a positioner-equipped globe valve on a slow integrating loop: dead band ≤0.5% of stroke, hysteresis ≤1% over a full stroke. A measured dead band above 1% is unacceptable; above 2% fully explains the observed jumps. Also record supply pressure (should hold steady at 1.4 bar / 20 psig under stroke - a drooping supply has the same signature) and confirm the valve is not running below 20% travel, where packing friction dominates.
2. Evaluate maintenance's proposal to halve the gain. Quantify the effect on cycle amplitude and on operability, given the 2.5-minute residence time and a pump that cavitates near the current low point of the cycle.
Read the worked answer
Halving the gain does not remove a stiction limit cycle - it stretches and enlarges it.
For a stick-slip cycle in an integrating loop, the peak-to-peak amplitude scales roughly as (dead band) ÷ (controller gain). Dropping Kc from 2.0 to 0.8 therefore takes the current 14% peak-to-peak swing to something on the order of 25–30% peak-to-peak, and lengthens the period.
Consequences against this specific drum:
- The drum cannot absorb it. With a 2.5-minute residence time, a 25–30% swing is a large fraction of the working volume. The low end will go below the level the reflux pump needs - the cavitation problem gets worse, not better, and a low-level trip becomes likely.
- Disturbance rejection collapses. A step in distillate draw will integrate the level away from setpoint faster than a Kc of 0.8 can correct. You trade a fast, small cycle for a slow, large one plus poor regulation.
- Phase margin is already thin. The short 3-minute period says the loop is close to its stability limit, so raising gain is not an attractive alternative either.
Verdict: reject 0.8 as a fix. A modest reduction (to roughly 1.2) is defensible only as one element of an interim package that also moves the level setpoint up and adds a suction-pressure alarm, and only with a firm date for the valve work. Detuning alone converts a visible nuisance into an availability and equipment-damage problem that is harder to diagnose later.
3. Define the permanent fix and the interim operating plan for the six weeks before the next maintenance window. Include what you would specify for the rework and how you would accept it.
Read the worked answer
Permanent fix
- Pull the valve: inspect plug and seat for wiredrawing, check stem straightness, replace packing with low-friction packing to the correct torque.
- Check the trim against the sizing sheet at min/normal/max flow. If normal flow parks the plug below ~20% or above ~80% travel, the valve is oversized for the duty and is operating where packing friction dominates - resize the trim. This is the single most common root cause of a stiction-dominated level loop.
- Fit a smart positioner with automatic friction/dead-band compensation and diagnostics (travel deviation, friction trending).
- Acceptance: as-found and as-left signature tests; step response at 25/50/75% travel; dead band ≤0.5% and hysteresis ≤1% of stroke, documented. Do not accept "it moves freely by hand."
Interim plan (six weeks)
- Raise the level setpoint into the upper part of the drum so the ±7% swing stays clear of the pump's NPSH limit. Cheap, immediate, and directly addresses the cavitation.
- Add a pump suction pressure low alarm and a low-low level alarm so the risk is visible between rounds.
- Reduce gain only modestly (to ~1.2), not to 0.8, and accept the resulting amplitude increase knowingly.
- Keep the loop in automatic. Manual operation on a 2.5-minute integrator is not sustainable across shifts.
- Trend review daily, with the valve work scheduled at the earliest opportunity - do not let a six-week window become a six-month one if the pump is at risk.
- Finally, compare the new DCS output module's resolution, filter settings and scaling against the old controller. A new DCS that adds output filtering or coarse resolution can create or aggravate stick-slip; rule that out before blaming only the valve.