Neural Netowrk Based Fault Diagnostics of Industrial Robots using Wavelt Multi-Resolution Analysis

Abstract

A multi-resolution wavelet analysis coupled with a neural network based approach is applied in the problem of fault diagnostics of industrial robots. The multi-resolution analysis implements discrete wavelet transforms with filters and decomposes the signal in various levels. The approximate and detailed coefficients of the decomposed signals are then used for training a feedforward neural network whose output determines the state (faulty or normal) of the robot. The neural network classifier was then implemented and monitored in a Matlab-Simulink environment using a state-flow model. Validation of the method was performed offline using experimental data obtained from an industrial robot manipulator used in the semi-conductor industry.