Process control is about continuously monitoring a process variable and using feedback to keep it at a desired value. In a closed-loop system, a sensor measures the variable and a controller compares it to a setpoint. The resulting error drives an actuator to adjust the process output, so the variable stays near the target even when disturbances occur. This ongoing regulation distinguishes true process control from static actions like a one-time calibration, which doesn’t continuously regulate; from designing piping layouts, which is about physical arrangement; or from manual data logging, which records information without automatic control.

In this tuning method, you determine the point where the closed-loop system just starts to sustain oscillations and measure those two values from that condition. The gain at that point is the ultimate gain, and the period of the sustained oscillation is the ultimate period. These two numbers are then fed into the Ziegler–Nichols rules to set the PID gains (for example, Kp ≈ 0.6 Ku, Ki ≈ 2Kp/Pu, Kd ≈ Kp Pu/8 for the full PID version). That’s why the two measured values used to compute the PID gains are the ultimate gain and the ultimate period. The other options refer to different stability metrics or analysis methods (open-loop gain and phase margin, system delay and dead time, or gain/phase margins) that aren’t the measurements used in the classic Ziegler–Nichols PID tuning.

The main idea is the difference between a deterministic distortion of the sensor’s response and random variability in measurements. Nonlinearity means the output is not proportional to the input; the sensor’s transfer function bends or saturates, so the relationship between input and output deviates from a straight line in a systematic, repeatable way. This kind of distortion can be characterized and corrected with a calibration curve or compensation, and it depends on the input level. Random noise, on the other hand, is unpredictable fluctuations around the true value caused by various stochastic processes in the sensor and electronics. It does not form a fixed curve with input and can be reduced by averaging or filtering, but it won’t produce a consistent non-straight transfer relationship. So the distinction is: nonlinearity is a systematic deviation from linearity; noise is random variability.

Integral windup happens when the integral term keeps accumulating error while the actuator is saturated and cannot respond. The integral term sums past error to eliminate steady-state error, but if the actuator hits its limits, the output can’t increase any further even though the error remains, so the integrator continues to grow. Once the system leaves saturation, that large integral value drives the output beyond what’s needed, causing overshoot and a long recovery toward the setpoint. This degrades performance by increasing settling time and can lead to sustained oscillations in some cases. Anti-windup techniques—like limiting the integrator to the achievable output range, using back-calculation based on saturation, or stopping integration when saturation occurs—help prevent this. The issue is tied to the integral term, not the derivative term, which is more about noise sensitivity; saying windup has no impact on performance isn’t correct.