Synthesis of a robust neural input-state feedback controller for nonlinear systems

Abstract

This paper presents a design approach to hybrid control systems, combining analytical feedback linearization control techniques with neural networks. Such a mixed implementation leads to a more effective control design with improved system performance and robustness. The main objective of integrating neural networks is to overcome the problems with uncertainties in the plant parameters and structure encountered in the analytical model-based design. Shunt DC motor is characterized by complex nonlinear and time-varying dynamics and inaccessibility of some model parameters for on line measurements, and hence can be considered as an important challenging engineering topic. The input-state feedback linearization technique is known for its good results locally in a neighborhood of an operating point. However, these results are sensitive to model parameter variations and so performances may deteriorate. Neural network-based controllers are considered as candidates for this parameters sensitivity. In a first step, an algorithm for analytical exact input-state linearizing control is formulated. The following step is dedicated to the robust neural feedback controller design. A simulation study of these methods is presented. The effectiveness of the neural controller with respect to the analytical one is demonstrated for a large armature resistance variation.