Fault detection in robots using stochastic filtering

A monitoring scheme for detecting and isolating abrupt and tendentious faults is developed for industrial robots executing the same sequence of motion with the same load repeatedly. The first part of the monitoring system is an innovation-based fault detection system identifying the joint where the fault occurred. The second part is a fault isolation system which uses tests based on variance ratios to determine whether or not significant differences exist among the statistics of groups of observations, obtained by the realization of the cycle. The monitoring system is improved by using a better reference trajectory for linearization in each working cycle. The performance of the monitoring system was partly evaluated by computer simulation, using a two-joint robot manipulator