Abstract
This research develops a method of machine fault diagnosis system using neural networks and spectral analysis. Generally, it is difficult to diagnose the fault of a machine by the conventional mathematical method. In this research, normal and fault spectral data, which are obtained from operating machine, are learned by the diagnosis neural network. Next, when a fault occurs in the machine, the fault is detected using this fault diagnosis system. This diagnosis system can diagnose not only a known fault that was learned but also an unknown fault. And, it can also find the unknown fault pattern based on the known fault pattern. This diagnosis system diagnoses the fault based on only the change of sounds or behaviors obtained from a operating machine. Therefore, the characteristic of this method is that, as neural networks can identify machine system, a mathematical model is not need.
