Direct iterative tuning via spectral analysis (Q5926265)

A model-free scheme is proposed for control tuning, which exploits the frequency-domain properties of the closed-loop signals. The tuning procedure is formulated as a minimization problem for a frequency-weighted linear quadratic cost function, with respect to the controller parameters. Both first and second derivatives of the cost function can be computed by using the time-domain data collected from closed-loop experiments. The iterative adjustment of the controller parameters according to the directions resulting from minimization guides the system towards an optimal-regulation state, such that the tuning method circumvents the need for an explicit model of the plant. The scheme incorporates a mechanism that computes stability margins and ensures closed-loop stability during the tuning procedure. A simulation experiment illustrates the applicability of the procedure.