Process control

A plant at steady state on paper is constantly pushed off it in reality: feed changes, the weather, fouling. Control keeps it where it should be. Each control loop has a controlled variable (what you want to hold, such as a tank level), a set point (the value you want), and a manipulated variable (what you adjust, usually a valve on a flow).

In feedback control a sensor measures the controlled variable, a controller compares it with the set point, and the difference (the error) moves a valve. A heater outlet too cold? Open the steam valve a bit. It's simple and copes with disturbances you never predicted, but it only acts after something has already gone wrong.

Most controllers are PID. The P (proportional) part moves the valve in proportion to the error now. The I (integral) part adds up error over time, so a small error that lingers is eventually removed. The D (derivative) part reacts to how fast the error is changing, damping overshoot; it's often left out because it amplifies noisy signals. Tuning means picking the three settings so the loop is quick without swinging back and forth.

Feedforward control measures a disturbance before it hits and corrects for it in advance: if the feed flow to a heater jumps, open the steam valve at once, before the outlet temperature falls. It's usually combined with feedback, which cleans up whatever feedforward misses.

Valves are chosen to fail safe when air or power is lost. A cooling water valve on a reactor should fail open (cooling stays on); a fuel valve to a furnace should fail closed (fire goes out). Safety instrumented systems, separate from normal control, shut the plant down if a reading passes a trip limit.