Fired-Heater Temperature Loop Cycling After a Control-Valve Replacement
A fired crude preheat furnace discharges at 285 °C into a distillation column. The outlet temperature PID (4 s scan, derivative enabled) has oscillated with a 4-minute period and ±6 °C amplitude since a new control valve was installed. In manual, a 5 % step in controller output produces no measurable flow change for about 3 s, after which the flow jumps by 8 %. A separate step test on the fuel-gas flow gives a 2 % step, a final outlet temperature change of 6 °C, 63 % of the response in 45 s and a transport delay of about 15 s. Operators have been putting the loop in manual during feed changes and adjusting the valve by hand. The furnace has a 15-minute loss-of-flame trip, the coil outlet limit is 320 °C, and the refinery is considering a model-predictive controller.
1. Diagnose the oscillation from the evidence and state what tuning alone can and cannot achieve.
Read the worked answer
The valve is the prime suspect, not the tuning.
A 5 % output step that produces no flow for ~3 s and then an 8 % jump is the classic signature of stiction/hysteresis: the actuator must build force until static friction is overcome, the stem then slips and overshoots, and the loop repeats. The apparent deadband is of the order of 3–5 % of stroke.
A PID cannot remove a mechanical nonlinearity. With integral action, the controller keeps winding the output while the valve is stuck; when the valve breaks free, the accumulated integral plus the overshoot drives the process past the setpoint, the controller reverses, and a limit cycle is established. The period is set by the process time constants and the controller integral time, and the amplitude by the deadband plus the loop gain. Reducing the controller gain will shrink the amplitude but will also slow disturbance rejection, and the cycle will persist.
Distinguish this from a tuning problem: in manual the process is stable; the cycle exists only in automatic; and the controller output has a square-ish, jumpy character rather than a smooth sinusoid. All three point to the final element.
Also check: the derivative action on a noisy thermowell with a 4 s scan. Derivative amplifies measurement noise and interacts badly with valve deadband; it is probably contributing even if it is not the root cause.
What tuning can achieve: a safe, slow loop that does not amplify the deadband, buying time for the repair. What it cannot achieve: tight temperature control and the disturbance rejection the furnace needs.
2. Assuming the valve is repaired, propose the tuning approach and initial settings, and say whether MPC is justified.
Read the worked answer
Model from the step test:
- Process gain K = 6 °C / 2 % = 3 °C/%.
- Time constant τ ≈ 45 s; dead time θ ≈ 15 s (delay plus thermowell lag).
- θ/τ = 0.33, a delay-dominant loop that must be tuned conservatively.
IMC/PID settings (series form, gain in %/°C):
- Choose λ ≥ θ, e.g. λ = 15–30 s.
- Kc = τ / [K(λ + θ)] = 45 / [3 × 30] = 0.5 %/°C for λ = 15 s; halve it if λ = 30 s.
- Ti = τ = 45 s.
- Td = 0 initially. Only add derivative if the noise is filtered, and then Td ≈ θ/2 = 7 s with a filter of 8–10 s.
Practical steps:
- Verify the valve signature first: a 1 % staircase open and closed in manual, plotting flow against output, to prove the deadband is below about 1 %.
- Check the fail-safe action, the positioner calibration, the packing friction and the actuator sizing at the available air supply pressure.
- Tune in automatic with the loop on a safe setpoint, then step-test at the lowest expected load, since furnace gain changes with firing rate and feed flow.
- Enable anti-reset windup and bumpless transfer, and add output limits consistent with the flame safeguard.
MPC: a single temperature loop is not an MPC problem. MPC becomes justified only if the furnace is coordinated with the column, the feed preheat train and the firing-rate constraints in one multivariable controller. Fix the valve and tune the PID first; otherwise MPC will simply issue smoother commands into the same deadband.
3. What safety, mechanical-integrity and operating-procedure issues does this cycling raise?
Read the worked answer
Mechanical integrity: a ±6 °C swing at a 4-minute period is roughly 360 thermal cycles per day on the coil, the tube supports and the outlet transfer line. Over months this is a low-cycle fatigue and creep-fatigue concern at the tube-to-header welds and at any restraint. It also drives cycling of the burner tiles and refractory. The oscillation should be treated as a mechanical-integrity threat, not just a quality issue.
Safety: the coil outlet limit is 320 °C and the loop is running at 285 °C with a ±6 °C swing, so normal operation sits 29 K below the limit with excursions reducing that. If the loop is in manual during feed changes, the operator becomes the control system; the time available to react to a firing upset must be compared with the process safety time, and the loss-of-flame trip must remain effective throughout. Any credit taken in the LOPA for operator response to high outlet temperature is invalid while the loop is routinely in manual.
Operating procedure:
- Stop the practice of manual operation during feed changes; instead implement feed-flow feedforward into the fuel-gas controller, so the main disturbance is rejected before it reaches the temperature loop.
- Define a maximum time in manual and require supervisor authorisation.
- Add alarming for valve output versus flow divergence (stiction detection) in the historian, so the fault is caught early next time.
- Record the valve repair in the maintenance strategy and add a periodic partial-stroke or signature test.
Change control: the new valve changed the dynamics of a safety-relevant loop. The step test, tuning and trip verification should be documented as a management-of-change close-out.