October 2, 2026
The Feedback Loop in One Picture
A feedback controller does one thing: it measures a process variable, compares it with the value you want, called the setpoint, and adjusts something it can move, usually a valve, to shrink the difference. That difference is the error, e = y_sp − y_m. Everything interesting about the controller lies in how it turns the error into a valve position.
Written as transfer functions, the loop has a controller G_c, the process G_p, and a measurement G_m feeding back. Reducing the block diagram gives the closed-loop response to setpoint changes, Y/Y_sp = G_cG_p/(1 + G_cG_pG_m). The 1 + G_cG_pG_m in the denominator is the characteristic equation, and whether its roots sit in the left half of the complex plane decides whether the loop settles down or oscillates out of control.
Proportional Action and the Offset Problem
The simplest controller moves the valve in proportion to the error, with the controller gain K_c setting how hard it reacts. Proportional control is fast and intuitive, but it has a built-in flaw. When a sustained disturbance arrives, such as a colder feed into a heater, the controller needs a new, permanent valve position to compensate. A proportional controller can only hold that new position if a permanent error is driving it. The result is offset, a steady-state gap between setpoint and measurement that never closes.
Raising the gain shrinks the offset, but it also makes the loop more aggressive. Past some gain, the loop starts to oscillate, and past the ultimate gain, those oscillations grow instead of dying out. Proportional control alone is therefore a compromise between offset and stability, and it is acceptable only where a small offset does not matter, such as many level loops on surge tanks.
Integral Removes Offset; Derivative Anticipates
Integral action adds a term proportional to the accumulated error over time, scaled by the integral time τ_I. As long as any error remains, the integral keeps growing and keeps pushing the valve, so the only possible steady state is zero error. That eliminates offset, which is why proportional-integral control is the workhorse of process plants. The cost is a slower, more oscillatory response, and a failure mode called integral windup: if the valve saturates fully open, the integral keeps accumulating, and the controller overshoots badly once the valve can finally respond. Industrial controllers include anti-windup logic for exactly this reason.
Derivative action responds to the rate of change of the error, scaled by τ_D, so it reacts to where the error is heading rather than where it is. That damps oscillation and allows more aggressive tuning on slow processes with large lags, like temperature loops on big vessels. It also amplifies measurement noise, since noise has large rates of change, so derivative action is usually filtered and is left off fast, noisy loops like flow control. Surveys of industrial plants consistently find that most loops run as PI, with derivative used selectively.
Ziegler-Nichols Tuning, and Why It Is Only a Starting Point
In 1942 John Ziegler and Nathaniel Nichols, engineers at Taylor Instrument Companies, published tuning rules that are still taught everywhere. Their closed-loop method is simple to describe: switch off integral and derivative action, raise the proportional gain until the loop sustains a steady oscillation, and record that ultimate gain K_u and the oscillation period P_u. Then set K_c = 0.6K_u, τ_I = P_u/2 and τ_D = P_u/8 for PID, or K_c = 0.45K_u and τ_I = P_u/1.2 for PI.
The rules were aimed at a quarter-decay response, where each oscillation peak is a quarter of the one before. That is quite aggressive by modern standards and gives little margin if the process changes. Many plants also cannot tolerate deliberately driving a loop to the edge of instability. Model-based methods have largely taken over in practice. Internal model control rules, such as Skogestad's SIMC method published in 2003, fit a simple first-order-plus-dead-time model to a step test and compute settings with a single tuning parameter that trades speed against robustness.
References
- Ziegler, J. G., & Nichols, N. B. (1942). Optimum settings for automatic controllers. Transactions of the ASME, 64(8), 759–765. doi:10.1115/1.4019264
- Seborg, D. E., Edgar, T. F., Mellichamp, D. A., & Doyle, F. J. (2016). Process Dynamics and Control (4th ed.). Hoboken, NJ: Wiley.
- Åström, K. J., & Hägglund, T. (2006). Advanced PID Control. Research Triangle Park, NC: ISA.
- Skogestad, S. (2003). Simple analytic rules for model reduction and PID controller tuning. Journal of Process Control, 13(4), 291–309. doi:10.1016/S0959-1524(02)00062-8