|Budget Amount *help
¥1,100,000 (Direct Cost : ¥1,100,000)
Fiscal Year 1999 : ¥1,100,000 (Direct Cost : ¥1,100,000)
Among the many problems in industries, the most important factor is for each machine systems to work in the normal state. In order to maintain a normal condition of a machine system, a fault prediction and diagnosis system are necessary, especially for those such as electrical generators, which run continuously for long time. The failures should be detected as soon as possible when failures occur, because if these machines run at abnormal condition continuously, it may result in heavy loss and even loss of human lives. Up to date, though many kinds of diagnosis methods have been developed, much of them have been based on traditional method of establishing mathematical model, analyzing variety of parameters and judging the operating condition of machine system. However, because of the complication of machine system, the uncertainty of operating condition and many nonlinear factors, it is, in most cases, very difficult to establish mathematical model of machine structure and to know the
operating conditions of the machine. Also it is that, in some cases, even impossible to detect failures occurred when it is running. So that, many researchers are recently attracted in untraditional approaches.
This study develops a method of fault diagnosis based on machine fault diagnosis system using neural networks. For approached of fault diagnosis, perhaps the oldest and the most wide used method is hearing sound of operating machine by human ear. Because, it can be thought that sound signal of operating machine includes important information of machine condition. Therefore, it can be considered that sound signal is suitable for fault diagnosis of machines. However, time series data of sound signal is very complex, and it is influenced by noise. Here, the power spectrum data of sound signal is used as fault diagnosis signal. In this method, the diagnosis system has been constructed by neural network. The diagnosis neural network learns the already obtained data of both normal and abnormal operating condition of machine system. This diagnosis system diagnoses the fault based on learned condition. Through experiments, the effectiveness of proposed diagnosis system is verified.