The PID controller is the workhorse of industrial control. It runs temperature loops, motor drives, flow valves, and flight surfaces. The algorithm itself is three terms and a line of arithmetic. The hard part has never been the controller — it is choosing the three gains. Set them too low and the system crawls toward its target. Set them too high and it oscillates, or worse, runs away. "Tuning" is the craft of finding the gains that land in between.
This article covers two complementary ways to tune a PID loop: the empirical Ziegler-Nichols method and the graphical root locus view — and shows how they describe the same boundary from different sides.
Why this calculation matters
An untuned loop is not a minor inconvenience. A sluggish temperature controller wastes energy and product. An oscillating position loop hammers the mechanics and shortens equipment life. An unstable loop is a safety incident. Between those failure modes sits a band of gains that gives a fast, well-damped response — and tuning is how you find that band on purpose rather than by trial and error at the plant.
The two methods matter because they suit different situations. Ziegler-Nichols needs only a simple experiment on the real system and no model. Root locus needs a model but, in return, shows you the whole stability picture at a glance. A capable engineer reaches for whichever the situation allows.
The two methods
Ziegler-Nichols, ultimate-sensitivity form. Disable the integral and derivative terms, leaving proportional control only. Raise the proportional gain until the loop sits in a steady, sustained oscillation — neither growing nor decaying. Record two numbers: the gain that produced it, the ultimate gain Ku, and the period of the oscillation, the ultimate period Pu. The classic PID settings then follow from a small table:
Kp = 0.6 * Ku
Ti = Pu / 2 (integral time)
Td = Pu / 8 (derivative time)
In the parallel form the integral and derivative gains are Ki = Kp/Ti and Kd = Kp·Td.
Root locus. This is the model-based companion. The root locus is a plot of where the closed-loop poles travel in the complex plane as the proportional gain increases from zero. Poles in the left half-plane mean a stable, decaying response; poles crossing into the right half-plane mean instability. The exact gain at which the locus crosses the imaginary axis is the ultimate gain Ku, and the frequency at that crossing gives the ultimate period through Pu = 2 pi / omega. The two methods are looking at the same stability boundary — Ziegler-Nichols finds it by experiment, root locus finds it by geometry.
A worked example
Suppose the ultimate-sensitivity experiment on a process gives an ultimate gain Ku = 8 and an ultimate period Pu = 2 seconds. Apply the Ziegler-Nichols PID table:
Kp = 0.6 x 8 = 4.8
Ti = 2 / 2 = 1.0 s -> Ki = Kp / Ti = 4.8
Td = 2 / 8 = 0.25 s -> Kd = Kp x Td = 1.2
So a starting controller is Kp = 4.8, Ki = 4.8, Kd = 1.2. Note the word starting. Classic Ziegler-Nichols deliberately targets a fairly aggressive response — roughly quarter-amplitude decay, meaning each overshoot is about a quarter of the one before. That is lively. For a loop that must not overshoot, back the proportional gain down by 20 to 40 % from the table value and re-check. Ziegler-Nichols gets you into the right neighbourhood fast; the final trim is yours.
The root locus view of the same system would show the closed-loop poles starting on the open-loop poles, sweeping outward as gain rises, and crossing the imaginary axis exactly at gain 8 — confirming Ku — at a frequency of pi rad/s, which gives Pu = 2 pi / pi = 2 s. Same two numbers, reached without ever oscillating the real plant.
Common mistakes
Running the Ku experiment on a fragile plant. Driving a real system into sustained oscillation is not always safe — think of a large thermal mass or a machine with travel limits. When oscillating the plant is risky, use a model and root locus, or a relay-feedback test that bounds the swing.
Leaving the gains at the textbook values. The Ziegler-Nichols table is a first guess tuned for disturbance rejection, not for a smooth setpoint response. Treat its output as iteration zero.
Over-using the derivative term. Derivative action amplifies measurement noise. On a noisy sensor a large Kd makes the actuator chatter. Filter the derivative, or reduce it, before blaming the loop.
Forgetting the sample rate on a digital controller. A discrete PID is not the continuous one. If the loop runs too slowly relative to the process dynamics, the phase lag from sampling erodes the stability margin the tuning assumed.
Try the interactive NovaSolver calculator
Tuning is much easier to learn when you can see the step response change as you move the gains. The PID tuning tool on NovaSolver lets you apply Ziegler-Nichols settings and then drag Kp, Ki, and Kd to watch overshoot, settling time, and steady-state error respond in real time.
Related calculators
- PID controller response — to see how each of the three terms shapes the output on its own.
- Root locus — the graphical stability picture, including the imaginary-axis crossing that defines Ku.
- Digital PID discretization — how sample rate and discretization affect a controller running in software.
The full set is in the PID and controls tools hub.
Closing note
PID tuning has a reputation as a black art, but the two methods here remove most of the mystery. Ziegler-Nichols gives you a fast, model-free starting point from one experiment. Root locus gives you the whole stability map if you have a model, and it pinpoints the exact boundary the experiment is feeling for. Use them together: locate the stability edge, back off to a sensible margin, and trim against the response you actually want. That is tuning done deliberately rather than by luck.
Top comments (0)