A Real-Time Nonlinear Hammerstein Model For Bidirectional DC Motor Based on Multi-Layer Neural Networks

Abstract

System identification is finding a model that can describe the dynamic characteristic of the examined system and predict the next output depending on the collected input/output data for that system at previous times. All the real dynamic systems have a nonlinear behavior, but this non-linearity is graduating from a simple to a brutal degree; Mechatronic systems are not spared from this rule. This article presents a real-time nonlinear model for bidirectional DC motor based on block-oriented Hammerstein to avoid the Coulomb friction and its dead zone nonlinear effect with the viscous friction. The recursive weighted least squares (RWLS) method is used to train the Hammerstein network. The mean square error for the system’s closest model is about 9.5 relative to fluctuated output speed from 1870 to -1035 (rpm).