control//feedback control//open-loop control

Open-loop control is control that sends a precomputed command to the plant without measuring the result, and it is used wherever the process is repeatable enough, and the disturbances small enough, that checking would cost more than the error: a toaster heats for a fixed time and never looks at the bread, a washing machine runs its programme, an irrigation timer opens the valves at six. A thermostat is the closed-loop counterpart: it measures the temperature, compares it with the setpoint and decides, without knowing how many windows are open.

Open-loop control is control that sends a precomputed command to the plant without measuring the result, and it is used wherever the process is repeatable enough, and the disturbances small enough, that checking would cost more than the error: a toaster heats for a fixed time and never looks at the bread, a washing machine runs its programme, an irrigation timer opens the valves at six. A thermostat is the closed-loop counterpart: it measures the temperature, compares it with the setpoint and decides, without knowing how many windows are open.

Open loop works only if the model is exact and nobody pushes. To hold a small oven at a temperature rrr, the open-loop command is u=ar/bu=ar/bu=ar/b, computed from how fast the oven loses heat (aaa) and how much each watt heats it (bbb). A 10 % error in aaa gives a 10 % error in the temperature, and an open door gives an error the controller never learns about. Closing the loop divides both by the loop gain, without knowing what they were; that is the whole case for feedback control.

It fails silently. The stepper motors of a desktop 3D printer run open loop, counting steps instead of measuring position; when a nozzle catches on the print and the motor skips steps, every later layer is shifted and nothing in the machine knows. The failure mode of open loop is an undetected error, which is why it suits processes where an error is cheap or visible to a person.

It cannot go unstable on its own. With no return path there is no loop to oscillate, no sensor whose noise reaches the actuator, and nothing to tune; it is also as fast as the actuator allows, since it does not wait for an error.

Inside a closed loop it is called feedforward: the open-loop part does the bulk from the model, the feedback cleans up what the model got wrong. Most well-designed loops combine both, and pure open loop is the special case where the feedback was judged not worth its sensor.

Choosing between them is a question of cost and consequence. Open loop wins on a repeatable process whose error is cheap and tolerable. Closing the loop pays when the plant drifts, disturbances are large or an unseen error is expensive, and it brings along a sensor whose errors the loop inherits and a stability question the open loop never had.