Temperature Loop Hunting After Packing Was Tightened
A preheat-train outlet temperature controller on a 12-inch line has hunted continuously for three weeks. The PV is a smooth sinusoid of about ±6 °C with a period near 90 s. The controller output also cycles, but with a distinctive shape: it sits nearly flat near 26% for several seconds, then jumps quickly to near 38%, sits flat again, then jumps back. The loop is a PID with Kc = 2.5, Ti = 60 s, Td = 0. The valve is a 3-inch equal-percentage globe with a positioner; maintenance records show the stuffing-box packing was tightened during the last outage - the same outage in which the bump test was run. That test gave process gain 1.8 %PV/%, dead time 12 s, and time constant 45 s. Production tolerates no more than ±2 °C. The instrument technician wants to halve the controller gain; the process engineer says the tuning is fine.
1. Using the shape and period of the oscillation, decide which of the two is closer to right and justify the diagnosis.
Read the worked answer
The technician's fix will not work; the process engineer is closer, but 'tuning is fine' is also too strong.
Evidence from the data:
- Output shape. A pure excess-gain limit cycle produces a roughly sinusoidal OP that leads the PV by a consistent phase. Here the OP has flat segments and fast transitions - the classic staircase/sawtooth signature of a valve that must build up force to break static friction, then slips. That is stiction (or positioner deadband/backlash), a nonlinearity, not a tuning problem.
- Timing. With 12 s dead time and a 45 s time constant, a loop hunting purely from high gain would cycle with a period on the order of a few dead times plus dynamics - tens of seconds. A 90 s period is slow for this process and is typical of a friction-driven limit cycle.
- History. Packing was tightened in the same outage. Increased packing friction is the most probable root cause, and no tuning change occurred to explain a new oscillation.
Why halving the gain is the wrong move: in a stiction loop, detuning usually makes the cycle slower and larger in PV amplitude, because the controller takes longer to unwind the accumulated error before the valve breaks free. It buys quiet at the console and hides a mechanical fault.
2. Compute IMC PI settings from the bump-test data and compare them with the settings in service. What does the comparison tell you?
Read the worked answer
FOPDT data: Kp = 1.8 %PV/%, τ = 45 s, θ = 12 s.
IMC-PI rules: Kc = τ / (Kp(λ + θ)), Ti = τ. The closed-loop speed is set by λ, and a common robust default is λ = θ.
- λ = θ = 12 s → Kc = 45 / (1.8 × 24) = 1.04, Ti = 45 s
- λ = 3θ = 36 s (conservative) → Kc = 45 / (1.8 × 48) = 0.52, Ti = 45 s
In service: Kc = 2.5, Ti = 60 s.
Interpretation
- Current gain is roughly 2.4× the aggressive IMC value and ~5× the conservative one, with a slower integral time. The loop is tuned hot - that is a real, secondary problem, and it will amplify whatever disturbance the valve introduces.
- But the observed PV amplitude (±6 °C) and OP staircase are not consistent with gain alone. The honest position: fix the valve first, then retune to λ ≈ θ (or λ = 2θ if robustness matters), because retuning a stiction loop just relocates the oscillation.
3. What actions do you take this week, in order, given that production wants ±2 °C now and the valve cannot be pulled from service?
Read the worked answer
Immediate (today)
- Take the loop off automatic and run on manual at a fixed output while the temperature stays within limits, if downstream can accept it; a steady manual output is better than a ±6 °C cycle. Otherwise reduce Kc toward 1.0 and increase Ti, accepting slower rejection, purely as a holding action.
- Log the cycle with a valve-position vs. controller-output trend - the signature is the evidence for the work order.
This week
- Valve technician: verify air supply and positioner zero/span, run a step test (5% steps up and down), and measure the deadband and the breakaway output. Deadband above roughly 1–2% of stroke confirms friction.
- Re-tension the packing to the manufacturer's minimum, not 'tightened until it felt right'; if the service allows, use low-friction packing.
After the valve is free
- Re-run a short bump test and set Kc ≈ 1.0, Ti ≈ 45 s, then verify the response to a 3 °C setpoint change against the ±2 °C tolerance.
- Record the failure mode in the maintenance history so the next outage does not repeat it.