Supervision of adaptive control algorithms (Q1576503)

It is often argued that a difference between rule-based adaptive controllers and model-based adaptive controllers is that the rule-based controllers perform the adaptation after major disturbances, whereas model-based controllers perform a continuous adaptation. The authors show that this is not true. Even the model-based controllers have to ensure that adaptation only is performed when the excitation is sufficiently high. The key robustness issue is to ensure that adaptation only is performed when the variations of the measurement signal are caused by the control signal. In this paper, it is suggested that this is ensured in the following way. First, an excitation detection procedure is used to detect when the excitation is high enough. Several additional supervisory functions are then added to ensure that adaptation is performed only when \(y(t) = y_u(t),\) where \(y(t)\) is the process output and \(y_u(t)\) is caused by the control signal. Procedures to improve the robustness of adaptive controllers are presented.