Fault Detection and Isolation in a Puma 560 Manipulator Via Neural Networks

摘要

In this paper a Multilayer Perceptron trained with Backpropagation is employed to reproduce the manipulator dynamical behavior. The Perceptron outputs are compared with real measurements generating the residual vector. Then, a RBF Network isutilized to classify the residual vector generating the fault isolation. Two different algorithms have been employed to train this Network. The first employs subset selection to choose the radial units from the training patterns. The second utilizesregularization to reduce the flexibility of the model. Simulations employing a two link manipulator and a Puma 560 are presented demonstrating that the system can detect and isolate correctly faults that occur in nontrained trajectories.