Manual tuning from trends

In short. Change one parameter at a time, by a factor of 1.5–2, and watch the response to a small set-point step. Fast oscillation — reduce Kc; slow long oscillation — increase Ti; a sluggish return without oscillation — increase Kc or reduce Ti. A model (if you have one) gives a better starting point: IMC and SIMC.

The model is not the plant

The numbers below come from the section model (K = 1.8 °C/%, τ = 45 s, θ = 12 s) in the simulator. They show the direction of the effect. A plant has noise, non-linearity, saturation and disturbances, so the same settings give a similar but not identical picture.

Effect of the parameters

PI controller, set-point step:

Open the baseline in the simulator — change Kc and Ti and compare with the table.

Telling too much P from too much I

• Too much Kc: relatively fast oscillation — a period of about 4–5 θ (58 s with θ = 12 s in the example), output and PV almost in antiphase.
• Too much integral (Ti too small): slower oscillation (99 s in the example), PV smoothly "floats" past the set point, the output lags PV by about a quarter period.
• Ti too large: PV approaches the set point quickly, but the last percent of error takes very long.

Order of steps

1. Starting values. From the model; without one — a Kc such that a 2–5 % set-point step changes the output by roughly the same percentage, Ti ≈ the estimated process time constant, Td = 0.
2. Small set-point steps (or wait for natural disturbances) — in both directions.
3. Kc first, then Ti, then (if needed) Td. Change by a factor of 1.5–2.
4. Write down each "setting → result" pair; evaluate with the response metrics.
5. Stop at an acceptable result. The last few percent of speed are usually bought with stability margin.

Safety

• Changing settings on a running plant is a process operation. Agree on it, define acceptable deviations and be ready to switch to manual.
• Switching modes and changing parameters must not bump the output (bumpless transfer).
• An oscillating loop that "seems to have calmed down" may start oscillating again at a different load.

Typical mistakes

• Changing several parameters at once — you cannot tell what worked.
• Curing oscillation by reducing Ti (more integral) — only makes it worse.
• Tuning with a saturated output — the controller is not in control and the settings do not show in the response (integrator windup).
• Tuning around a mechanical problem: valve backlash or stiction cause oscillation that tuning cannot remove (troubleshooting).

Limits

The period-based signs are guidelines for self-regulating processes. For integrating processes (level) slow oscillation is more often caused by too small a Kc together with a small Ti.