PID loop oscillating
An oscillating loop is giving you more information than most people take from it. The shape of the oscillation, its period, and whether it decays all point at different causes, and reading them costs nothing.
The three questions worth answering, in order: does it decay, how long is one cycle, and does it survive putting the loop in manual. Between them they narrow the field to one or two candidates before you touch a single constant.
Does the oscillation decay?
A decaying oscillation — each peak smaller than the last — is a tuning signature. The loop is over-correcting but the correction is still stabilising, so it settles eventually. This is fixable from the faceplate.
A constant-amplitude oscillation that never decays is a limit cycle, and limit cycles come from nonlinearities, not from tuning. Stiction, backlash, and a valve hitting a travel limit all produce them. Tuning changes the period and does not touch the amplitude.
A growing oscillation means the loop is genuinely unstable. Put it in manual now and reduce the proportional action before putting it back.
How long is one cycle?
The period tells you roughly where in the loop the problem lives. A period of a few seconds on a flow loop is the loop's own dynamics — tuning territory. A period of many minutes on the same loop is not: nothing in a flow loop is that slow, so something external is driving it.
Compare the period against the loop's natural response time, which you can measure by stepping the output in manual and timing how long the process variable takes to settle. An oscillation much slower than that is coming from outside.
What to change first
If it decays and the period matches the loop's own dynamics, reduce the proportional action and observe. If it decays but the period is several times the loop's response time, the reset is too fast — slow it down before touching gain.
A rule that saves a lot of wasted effort: change one thing at a time and capture a trend after each change. Two changes at once and you have learned nothing about either.
What causes it, most likely first
Proportional action too high
most common tuning can fix this
The controller responds to error more strongly than the process can absorb, overshoots, and then over-corrects in the other direction.
- How to confirm it
- Decaying oscillation with a period close to the loop's natural response time. Output swings as hard as the process variable.
- What to do
- Reduce proportional action and observe the overshoot reduce.
Reset too fast for the dead time
most common tuning can fix this
Integral action repeating faster than the process can respond. It keeps adding correction while the last one is still in transit.
- How to confirm it
- Slow decaying cycle, period several times the loop's response time, output drifting steadily in one direction then the other.
- What to do
- Slow the reset substantially before adjusting anything else.
A nonlinearity in the final element
common tuning cannot fix this
Stiction, backlash or a valve sitting against a limit. These produce cycles that do not decay because the nonlinearity is the same size regardless of how small the error gets.
- How to confirm it
- Constant amplitude across many cycles. Stops in manual.
- What to do
- Deal with the valve. Tuning will not remove the cycle.
Derivative acting on a noisy measurement
common tuning can fix this
Derivative amplifies rate of change, and noise has a large rate of change even when the process is steady.
- How to confirm it
- High-frequency activity on the output that mirrors noise on the process variable, with the output noisier than the measurement.
- What to do
- Set rate to zero. On flow, pressure and speed loops it should almost always be zero anyway.
Questions that come up
Is quarter amplitude damping a good target?
It is a traditional one and it is more aggressive than most operating plants want. Quarter amplitude damping leaves a loop close to the edge — a modest change in process gain, which happens whenever load changes, can tip it into a much less comfortable place. Aim for a response that settles inside two peaks.
Why does the loop only oscillate at certain production rates?
Because the process gain changes with load, and a tuning set that suits one end of the range is too aggressive at the other. This is very common on flow loops with equal-percentage trim, and on any loop with a nonlinear process.
Related
Last reviewed 2026-08-01.