Choosing P, PI, PD or PID for a Loop

… more less You do not have to turn on all three terms. Pick the fewest terms that do the job. Start with the process. If a steady offset is not okay, you need integral action, so most loops get at least PI. Flow and pressure loops are fast, and their signals are noisy, so PI is the usual pick there. Derivative mostly adds chatter to them. Temperature loops are slow and smooth, with some dead time. PID often helps them, since D brakes the rise before it overshoots. A tank level keeps ramping on its own, so it is called integrating. P only or a gentle PI often works, and too much I makes a slow, wide swing. Position loops, like a motor moving to an angle, are also integrating. They often use PD, since D adds damping, which means it calms the ringing. Now check the measurement. If it is noisy, drop D or filter it, because D turns noise into output jumps. If the dead time is long compared with the time constant, D helps little, so use PI with a modest gain. If setpoint steps make the output spike, keep D but have it act on the PV, not the error. These are starting points, not laws. A loop that does not fit the pattern is a reason to test, not to guess. After this Concept you can look at a loop's speed, noise and process type, and choose P, PI, PD or PID with a reason.