Recurrent neuro-fuzzy modeling and fuzzy MDPP control for flexible servomechanisms

This paper considers the nonlinear system identification and control for flexible servomechanisms. A multi-step-ahead recurrent neuro-fuzzy model consisting of local linear ARMA (autoregressive moving average) models with bias terms is suggested for approximating the dynamic behavior of a servomechanism including the effects of flexibility and friction. The RLS ( recursive least squares) algorithm is adopted for obtaining the optimal consequent parameters of the rules. Within each fuzzy operating region, a local MDPP ( minimum degree pole placement) control law with integral action can be constructed based on the estimated local model. Then a fuzzy controller composed of these local MDPP controls can be easily constructed for the servomechanism. The techniques are illustrated using computer simulations.