Wavelet network for in-loop identification and control of frictional nonlinear systems

Abstract

We propose a novel wavelet network for closed-in-loop identification and control of systems containing nonlinear coulomb friction. Usually an actual system contains both linear and nonlinear dynamics, hence my wavelet network structure yields two parts: (a) nonlinear connection from hidden (middle) layer neurons, (b) linear direct connections from input regressors to output neuron. The wavelet network identifier has appropriate dynamical functional links to identify the friction of the nonlinear system. Moreover, wavelet network controller employs the identified parameters for on-line training to compute appropriate control signal. The simulation results reveal that with this control strategy, the system can track even a periodic input reference signal successfully. In addition, in the presence of external disturbance and plant parameters variations, the compensated system operates facilitative. Furthermore, this identification and control strategy can eliminate chattering in spite of coulomb friction.