An overview of anti-windup techniques for saturated systems - Anti-windup control

A controlled process involves updating a control input, calculated using the error between an output measurement and its setpoint, through an actuator. In the ideal case, the actuator can be represented by a "wire" connecting the controller output and the system input. In practice, the actuator may have a dynamic behavior, but above all an operating range limited between two limits, lower and upper. These limits on the actuator's linear operating range are reflected in the notion of actuator saturation. This is a non-linear phenomenon which has the undesirable effect of degrading the nominal behavior of the system, in the best of cases, or even destabilizing the controlled system in extreme cases.

To avoid the undesirable effects of actuator saturation, a conventional solution is to oversize the actuator, which entails an additional cost in terms of the load carried, the energy expended, and the price of the actuator itself. A second classic solution is to limit the controller's performance (speed) to avoid, as far as possible, approaching the actuator's linear operating limits.

Anti-windup control is an alternative way of making the best possible use of the actuator's capabilities, by allowing saturation while preserving the stability (and performance) of the controlled system within the system's intended operating range. The principle is to feed back the actuator's error to the control system's input. This controller feedback is only active when the actuator is saturated. In this way, the nominal behavior of the controlled system, originally designed without regard to actuator non-linearities, remains unchanged. In its simplest version, the synthesis of an anti-windup control can be likened to the synthesis of a simple additional gain, and presents no more difficulty than the synthesis of a PID corrector.

By extension, anti-windup control can also be applied to counter speed saturation or higher actuator dynamics, or even other types of actuator non-linearities, although this is not covered in this article.