Research and development of intelligent diagnosis based on neural network

Abstract

The intelligent diagnosis technology has got a rapid development in the past few decades. Some new theories and methods have been applied to practice successfully, and obtained fruitful achievements. The artificial neural network leads to a new way for the research on fault diagnosis. It also provides a new scientific theory and thinking way for fault diagnosis. The intelligent fault diagnosis technology based on neural network has a great application potential and practical value. At present, it plays an irreplaceable role in monitoring, prediction, diagnosis and control of mechanical equipment condition, especially in the study of nonlinear and non-stationary signal processing. It shows that the intelligent fault diagnosis technology based on neural network is worthy of serious study. However, each diagnosis method has both advantages and disadvantages. Because of the inherent mechanism of the artificial neural network, this method also has some weaknesses inevitably. It is difficult to find an almost perfect way for the fault diagnosis. Only the intersection of a variety of methods, theories and technologies, and the introduction of new technologies and thoughts, could get a more accurate result of the fault diagnosis. The fault diagnosis based on neural network mainly includes two processes, training and diagnosis matching. The research focuses on three aspects: considering the neural network as a classifier from the angle of pattern recognition, considering the neural network as a dynamic forecasting model from the point of fault prediction, and establishing an expert system based on the neural network from the perspective of knowledge processing. This paper introduces the two processes of fault diagnosis, analyzes the present research situation from the three different angles above, put forward their advantages and disadvantages respectively, and discusses the development trends in the future.