A neural network application to fault diagnosis for robotic manipulator

Abstract

This paper illustrates a new approach of performing fault detection and isolation for robotic manipulators. The neural network fault isolation monitor utilizes a non-linear observer to generate a residual set. This residual set is presented to an artificial feedforward neural network with full connectivity. The neural network extracts specific characteristics which correlate to the operational mode of the system during the off-line training session. Once trained, the network performs efficiently in detecting and isolating faulty modes of the system. Although a robotic manipulator is used to illustrate the effectiveness of this approach, we believe that it can also be applied to other non-linear systems.