Blog · 2026-08-23
Ziegler-Nichols tuning: the 1942 method everyone still names
How the two Ziegler-Nichols methods work, why both aim at quarter-amplitude damping, and why a method from 1942 is a poor default for a loop you intend to keep running.
The method older than the transistor
Ziegler-Nichols was published in 1942, five years before the transistor existed. John Ziegler and Nathaniel Nichols worked at the Taylor Instrument Company in Rochester, tuning pneumatic and early electric controllers, and their two papers defined how process control was taught for the rest of the century.
The first paper described the open-loop reaction-curve method: bump the process in manual, fit a first-order-plus-dead-time model to the response, and read the tuning straight off three numbers. The second described the closed-loop method, in which you raise the gain until the loop oscillates and record what is now called the ultimate gain and ultimate period.
Both survive because both are teachable. Neither needs a computer, a model in software, or anything more than a recorder and arithmetic — which is exactly why they filled every textbook that followed.
The closed-loop method asks you to court instability
The closed-loop method's input is the one condition you would normally pay a great deal to avoid. To find the ultimate gain you switch integral and derivative off, put the loop in automatic, and raise the gain until the process variable settles into a sustained constant-amplitude oscillation.
That oscillation is the stability boundary. The controller gain you record there, and the period of the cycle, feed correlations that assume quarter-amplitude damping — each successive peak a quarter of the one before — which is itself closer to that boundary than most operating loops should ever live.
On a production plant the method is therefore more than inconvenient. Driving a loop into sustained oscillation means upsetting the process deliberately, alarming the board, and explaining why the disturbance was self-inflicted. PIDSnap never asks for the ultimate gain, for precisely this reason: a number found by destabilising the loop is not worth the risk on something you intend to keep running.
The open-loop method needs a process that settles
The open-loop method is gentler in execution but stricter in assumption. You step the output in manual, wait for the measurement to respond and settle, and fit the three FOPDT numbers — gain, dead time, time constant — to the curve.
The correlations then give the tuning as multiples of those three numbers, with reset and derivative set as fractions of the dead time. All of it presumes a self-regulating process that actually settles; an integrating level, or a response with extra lags, quietly invalidates the numbers the rules hand back.
What the method is actually good for
Judged as a production tuning practice, Ziegler-Nichols is aggressive. Its quarter-amplitude target leaves little stability margin for the changes in process gain that ordinary load movement brings, and a loop tuned that hard on paper is frequently detuned on the plant.
What the method is genuinely good at is an order-of-magnitude estimate. On a new self-regulating loop with no history, the open-loop version gives a defensible first value faster than anything else, provided you treat it as a starting point and verify from the response. As a recipe for a loop you intend to keep running, it is a poor default.
What PIDSnap does instead
PIDSnap has no connection to any control system, and no feature that drives a loop to its stability boundary. A session starts with two photographs — the trend and the faceplate — and a pre-flight check that catches structural problems before any constant is touched.
From there the guidance works the way a bump test does: step the output, watch what comes back, and adjust by the response actually observed, one iteration at a time. It never originates a tuning value from a formula, because a number computed without having measured your loop is a guess.
Questions that come up
Does PIDSnap use Ziegler-Nichols tuning?
No. PIDSnap works from the measured response of your loop, not from the 1942 correlations. It never drives a loop into sustained oscillation to learn a number, which the closed-loop Ziegler-Nichols method requires.
Is the open-loop Ziegler-Nichols method safe?
It is gentler than the closed-loop method, but its numbers assume a self-regulating FOPDT process and a quarter-amplitude target. On a loop that fits those assumptions it is a defensible starting point; on anything else it is a guess with a formula attached.