Application of Radical Basis Function Neural Network in Fault Diagnosis of Rectifier

Abstract

The Radical Basis Function (RBF) neural network is a kind of three-forward neural network, which can approximate any continuous functions to arbitrary precision, particularly suited to solve classification problems. In this paper, according armored vehicle power system silicon rectifier prophase fault is single diode and diode short-circuit fault in a very short period of time turned into the situation of the open-circuit fault, make full use of the characteristics of the RBF network classification to determine the fault to a special diode of the rectifier model. It achieves the aim at fault diagnosis of rectifier. Comparing with BP neural network, RBF neural network has better classification ability. Index Terms - RBF neural network, Rectifier, Fault diagnosis. I. Introduction At present, army armored vehicles mostly adopt brushless silicon rectifying generator, which consists of a three-phase ac low voltage generator and an air-cooled three-phase bridge rectifier with silicon rectifier diode. Among them, the three- phase alternating current coming from the generator rectifier outputs into direct current to supply the power device and generator self excitation, and also for battery charging. As the constant voltage power supply for the whole vehicle electrical and electronic equipments, the rectifier has played an important role in guaranteeing the reliability and quality of the power system. Therefore, it has important practical application value to carry out fault diagnosis method research on rectifier. There are many literatures for the rectifier device fault diagnosis is studied. Among them, the literature (2) is proposed based on support vector machine (SVM) fault diagnosis method of rotating Rectifier Bridge, literature (3) is proposed a rectifier based on SOM and ELMAN neural network fault diagnosis methods, literature (4) puts forward a kind of fault diagnosis methods based on artificial immune algorithm brushless excitation generator. They all obtain a good diagnosis effect from the simulation results. RBF neural network belongs to the type of feed forward neural networks, in which the underlying layer uses nonlinear optimization strategy to adjust the parameters of the activation function. And the output layer adopts linear optimization strategy of the linear power adjustment. It can approximate any continuous function with arbitrary precision, especially suitable for solving the problem of classification. Overall, the network mapping from input to output is nonlinear, but the network output for adjustable parameters are linear. The adjustable parameters of the network can work out directly by the linear equation, which greatly accelerates the learning speed and avoids local minimum problem (5). This paper takes full advantage of the characteristics of RBF neural network and flexibly uses in the rectifier fault diagnosis. It is of great significance for intelligent fault diagnosis. II . RBF Neural Network (5) The neuron structure of Radical Basis Function (RBF) neural network is shown in figure 1. The basis function of the activation function uses radial basis function, which usually defined as the space of Euclidean distance between any points into the center of a monotonic function.