PID loop troubleshooting: symptom → cause → check
A table of typical control loop problems: oscillation, sluggish response, steady-state error, overshoot after saturation, a trembling output, different behaviour at different loads. For each symptom — likely causes and how to check them.
In short. Before changing the settings, find out whether tuning is to blame. Put the loop in manual with a constant output: if PV oscillation stops, the loop itself causes it (tuning, valve, sensor); if it continues, the source is outside (a disturbance, a neighbouring loop). Then use the table.
First check: manual mode
| What you see in manual | Conclusion |
|---|---|
| PV calms down | the loop causes the oscillation — see the "oscillation" rows below |
| PV oscillates the same way | an external source: a neighbouring loop, load, supply pressure |
| PV does not respond to output changes | actuator, a manual valve, a bypass, wrong channel |
| PV moves the opposite way | wrong direction of action (direction) |
Symptom → cause → check
| Symptom | Likely cause | How to check |
|---|---|---|
| Oscillation, period ≈ 3–5 θ, sinusoidal output | Kc too large | reduce Kc by 1.5–2; the period hardly changes, the amplitude drops (manual tuning) |
| Slow oscillation, PV "floats" past the set point | Ti too small | double Ti |
| Oscillation with a "sawtooth" output and a "square" PV | valve stiction | manual, output steps of 0.5–1 %: PV does not move, then jumps |
| Small oscillation around the set point with any settings | actuator backlash, dead band, quantisation | small steps both ways; compare signal and valve position |
| Slow approach to the set point without oscillation | Kc too small or Ti too large | response metrics; compare with a model-based calculation |
| Steady-state error | no I (P controller), or the output at a limit | is Ti set? is the output at a limit? |
| Large overshoot after a long saturation | integrator windup | time at the limit; enable anti-windup |
| Output "trembles" | PV noise + D, or a large Kc | turn D off, look at the PV noise; D filter |
| Output jump on a set-point change | P and D on error | D on measurement; set-point weighting |
| Jump when switching manual → auto | no tracking | bumpless transfer |
| Good at one load, oscillates at another | valve or process non-linearity | step tests at several points; compare K |
| Loop got worse after a sensor or valve replacement | range, characteristic or K changed | normalisation; a new test |
| Oscillation after a task period change | jitter, wrong Ts in the block | discretisation |
| PV runs away, output at 0 % or 100 % | wrong direction, or the actuator cannot cope | direction; is the valve range sufficient |
Example: how dangerous a K error is
The loop is SIMC-tuned for the model K = 1.8 °C/% (Kc = 1.042, Ti = 45 s). If the real process gain is twice as large (K = 3.6 °C/%), the product Kc·K doubles: 55 % overshoot and 184 s settling instead of 5.5 % and 74 s. Hence a practical rule: for a non-linear process tune on the region with the largest K.
What to record before changing anything
- Trends of PV, SP, output (and valve position, if available) over several oscillation periods.
- Current block parameters: form, units, Kc/Ti/Td, filters, limits, D mode, task period.
- Changes on the plant: replaced equipment, new operating modes, loads.
Limits
The table is a guide to frequent causes. One symptom can have several causes at once (e.g. valve stiction and a large Kc). Mechanical and process problems are not fixed by tuning.
Connections
Open in Knowledge Graph →Loading the map…
Sources
Updated 2026-10-12