An integral-plus-state adaptive neural control of mechanical system

Two integral-plus-state (IPS) direct adaptive neural control schemes (with one or two I-terms), appropriate for mechanical systems applications, are proposed. Both schemes contain recurrent trainable neural network (RTNN-1) as a plant parameter identificator and state estimator, and a neural controller (RTNN-2) with one or two additional I-terms, so to form the IPS action. Simulation results, obtained with a 1-DOF mechanical plant with friction, corrupted by noise, confirmed the good dynamic performance of both schemes.