A Real-time Vibration Diagnosis System by Using an Artificial Neural Network

• Main Author: 張顯盛
• Other Authors: 黎文龍
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/32740462693413530540

碩士 === 國立臺北科技大學 === 機電整合研究所 === 86 === 　　A vibration diagnosis system based on the back - propagation neural network (BPN) is developed in the present research. The building procedures of the system are described in the research thesis step by step. In addition, there are four different input features are utilized. They are the frequency response function (FRF), the power spectrum (PS), the orbits and the new statstical moments. The main objective of the study is to find the optimal representation set such that the conditions such as mass unbalance in rotation machines. 　　The study includes three parts. They are (1) to build the monitoring BPN system by Borland C++ programming language;(2) to train the neural network system such that it can adaptively identify the machine conditions; and (3) to test the system by feeding the vibration signals. By changing of unbalanced mass, the locations of the unbalance and the rotating speed, etc, the machine condition signals are generated through the rotor kit in the Lab. After pre - processing and transforming into the mentioned representations, there exist fifteen combinations of input features which are then fed to the system as the trainning and testing set. 　　After hundreds of the experimental works, the combination of FRF and PS is found to be the best input feature for the mass unbalance conditions. However, it also reveals that the system can successfully identify the mis - alignment if the condition signals are also included. Nevertheless, the study has showed that the monitoring system based on BPN and vibration signals may be the best way to resolve the on - line machine diagnoses.