Why does a PID loop oscillate?

Sources of oscillation

Oscillation starts when the loop cannot balance its gain against its delay. With high gain, a small error produces a large correction, and the correction overshoots. If the integral time (Ti) is much shorter than the process lag, the I term reacts to information that arrives late, and the system hunts. Dead time, the delay between a change in output and the change in measurement, adds phase lag.

Valve stiction and backlash make the valve move in jumps. The loop may then settle into a sustained oscillation known as a limit cycle. Output saturation adds to the problem through integral windup, and noise in the measurement is amplified by the derivative term.

Example: starting values with Ziegler–Nichols

On a temperature loop, only P is active and the gain is increased. At an ultimate gain Ku = 4, the output oscillates steadily at constant amplitude. The period of that oscillation is measured as Pu = 20 s (hypothetical). Using the classic Ziegler–Nichols closed-loop rules:

• Kp = 0.6 × Ku = 0.6 × 4 = 2.4
• Ti = Pu / 2 = 10 s, so Ki = Kp / Ti = 2.4 / 10 = 0.24 (1/s)
• Td = Pu / 8 = 2.5 s

These are only a starting point. If the loop still oscillates, lowering Kp further or increasing Ti may be needed.

Diagnosis and common mistakes

The first step is to measure the oscillation period. A short period often points to high gain or a noisy derivative; a long period suggests a slow integral or a large process delay. The period alone is not enough, so plot the controller output and the measurement on the same chart. With valve stiction, the controller output can look like a square or triangular wave. A common mistake is trying to remove the oscillation by changing tuning alone. Check the valve, positioner and sensor first.

Follow-up questions