A Comparison of Real-time NARX Models With Feedback From Real and Estimated Output

The process of creating a model that delineates a certain dynamic system by defining a link between the input and output is known as system identification. In real life, all systems engage in nonlinear behavior that can range from absurd to enormous. In this study, real output feedback and estimated output feedback NARX networks for biface loaded DC motors are compared with respect to backlash and dead zone nonlinear effects of gears and bearings. Recursive least-squares (RLS) algorithm is exploited to train the two networks in real-time system identification. To choose the optimum model that represents the system of loaded DC motor characteristics. The goodness of fit (gof), using normalized mean square error (NRMSE) as an objective function, is used as a cross-validation technique. In comparison to an oscillated angular velocity output, the best gof is around 94.86% for the true output feedback NARX model and 87.27% for the estimated output feedback NARX.