Fault Diagnosis of Robot Based on Deterministic Learning

A rapid fault diagnosis scheme based on deterministic learning was proposed for robot with disturbance. The system dynamics underlying normal and various fault modes were locally accurately approximated through deterministic learning, and the obtained knowledge of system dynamics was stored as constant neural weights to build out a mode bank. In the diagnostic process, the learned knowledge was reused through the dynamic pattern recognition method so that the fault can be detected and isolated quickly without training neural networks again. Simulations were included to demonstrate the effectiveness of the proposed approach.