Fault diagnosis system for machines using neural networks

This research is concerned with the fault diagnosis for machines using neur al networks. Generally, it is difficult to diagnose machine's fault by the conventional technique. In this research, two new fault diagnosis systems a re proposed which one diagnoses a fault based on behavior of the object sys tem, and another diagnoses a fault based on power spectrum of the object sy stem. In the former, when an object system is a normal state, the system id entification is performed by the neural networks. The diagnosis system dete cts a fault by finding the behavior's gap between the state of the real sys tem and the identified normal one, and also the fault part is specified by fault diagnosis neural network. In the latter, neural network learns power spectrum of both the normal and fault states for the object. When a fault o ccurs, fault part is diagnosed by fault diagnosis neural network based on p ower spectrum. Finally, through simulation and experiment, the effectivenes s of proposed fault diagnosis systems is verified.