A stable neural network observer with application to flexible-joint manipulators

A stable neural network based observer for a general multivariable nonlinear system is considered in this paper. A linearly parameterized neural network is employed for approximation of the unknown nonlinearities of the system. The recurrent network configuration is obtained by a combination of feedforward network architectures with dynamical elements in the form of stable filters. The weights of the network are updated according to a novel approach based on the modification of the backpropagation algorithm. The stability of the system is shown using Lyapunov direct method. No SPR assumption is imposed on the output error equation. The proposed observer is applied to a flexible-joint manipulator to evaluate its performance. The simulation results show the effectiveness of the proposed observer.