Study of Rotating Machine Fault Diagnosis System Using Neural Network

• Main Authors: Jhih-Wuei Gwu, 辜志偉
• Other Authors: 洪瑞鴻
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/04811851187522357960

碩士 === 國立臺灣海洋大學 === 機械與輪機工程學系 === 92 === Machine fault diagnosis system was to be respected for industry as a result of producing automatically with less and less operators. It is not easy to design fault diagnosis system for a complex and non-linear system with traditional mathematical module analysis. Neural network has the learning ability from training and tolerance for a slight change, and therefore it is more suitable for the complex and non-linear system especially. In this thesis, the network parameters trained from Levenberg-Marquardt method are to be used to classify the condition of rotating machine by back propagation algorithm. Simulates bi-directional acoustic emission like human hearing and employs multi-microphones system to acquire vibration sound signal brought on rotating machine. The spectrum analysis technique can extract signal features from frequency domain and these features were to be taken as input of neural network that can diagnose fault accurately. The diagnosis system in this study can detect motor condition on-line whether in normal state or not by using graphical user interface. It can diagnose not only single motor condition on-line, but also extend to several motors conditions at the same time.