An oscillating loop is rarely a mystery once you know what to look for. The symptom is much the same each time — a steady hum, a slow hunt, an actuator chattering against its limits — but the cause is almost always one of six faults. Work through them in order and you will find it faster than by twiddling gains at random.
1. Excessive loop gain
The classic. Push proportional gain too high and the loop crosses over with too little phase margin: it corrects quickly, overshoots, then keeps ringing because every correction arrives slightly late.
- Oscillation amplitude shrinks when you reduce gain and grows when you raise it.
- Ringing after a setpoint step, at roughly the frequency where open-loop gain crosses unity.
- A steady tone from a motor or valve that changes pitch with the gain setting.
- Almost no oscillation when the loop sits in manual.
Halve the gain. If the oscillation continues unchanged, stop — gain is not your problem, and you are about to make the loop sluggish for nothing. If it shrinks, halve again, then bring it back up until you see the first hint of ringing and back off by a factor of two or three. That headroom is margin, not wasted performance.
2. Sensor lag and filters you forgot you added
Every low-pass filter in the feedback path buys noise reduction and pays for it in phase. A filter with a time constant of half a second contributes noticeable lag at 1 Hz, and that lag comes straight out of your phase margin.
The giveaway is history. The loop was stable, someone added smoothing to calm a noisy signal, and now it hums. Or the oscillation frequency drops as you increase the filter time constant. Thermowells, long pneumatic sensing lines, slow RTDs and averaging windows in software all behave the same way.
To confirm it, temporarily bypass the filter and feed the raw signal back. Take care: raw signals can be noisy enough to wear out an actuator. If the oscillation frequency rises or disappears, you have found it. Removing the filter is not always the answer — sometimes you should slow the loop down, or filter less and clean up the signal at source.
3. Accidental positive feedback
This one does not sound like tuning trouble; it sounds like a fault. A sign error anywhere in the loop — reversed motor polarity, an encoder counting the wrong way, an inverting amplifier where you expected a buffer — turns negative feedback into positive feedback.
The signature is unmistakable:
- The output slams to one rail or limit and stays there.
- Any oscillation is a square wave between limits, not a smooth sine.
- Reducing gain does not help.
- The frequency is set by actuator slew rate or a saturation limit, not by loop dynamics.
Test it in open loop: disconnect the feedback, apply a small manual step, and watch the sensor. If the measured value moves the wrong way, the sign is wrong. Fixing polarity is usually a five-minute job that saves a week of tuning.
4. Too much integral action (and windup)
Integral action removes steady-state error and adds up to 90 degrees of phase lag while it does so. That is a large bill, and paying it too early in the frequency range makes the loop hunt slowly and persistently.
Look for a period of oscillation in the same ballpark as the integral time — often many seconds or minutes. Setting the integral term to zero stops it. Setpoint changes produce long, lazy overshoots that take several cycles to settle.
Windup is the cousin of this problem. When the actuator saturates, the integrator keeps accumulating error it cannot act on. The loop eventually comes out of saturation, over-corrects and oscillates between limits. Anti-windup clamping or back-calculation fixes that, but tuning the integral term sensibly matters first.
5. Dead time and mechanical resonance
Dead time — transport delay on a conveyor, fluid travel in a long pipe, network latency in a remote loop — is pure phase loss with no compensating benefit. It punishes aggressive tuning harder than almost anything else. If your process does nothing for a second and then moves, you cannot tune past it; you can only slow down.
Resonance looks different. A flexible mounting, a stretched belt, a long shaft or a lightly damped structure adds a sharp peak to the plant's frequency response. The symptom is oscillation at a fixed frequency that barely changes when you adjust gains, and that shifts when you change the mechanical setup — different belt tension, an added brace, a heavier load.
A bump test or a slow frequency sweep shows both. Dead time appears as steadily rotating phase lag; resonance appears as a peak with a rapid phase drop alongside it.
6. Coupling, ground loops and sampling artefacts
Some oscillations are not in the loop at all — they get into it. Motor cables routed alongside sensor wiring, a shared ground carrying load current, a switching supply dipping the reference, crosstalk between adjacent channels: all can inject a periodic disturbance that the controller then dutifully amplifies.
