Fault diagnosis of electric apparatus component using improved RBFNN

Radial basis function neural network (RBFNN) approach is investigated and applied for fault diagnosis of electric apparatus control system under working state, the aim is to achieve accurate fault type identification when component failures. After fault occurrence relationships among fault types, fault feature and fault cause are analyzed, non-linear mapping relationship between fault feature and fault cause is extracted based on engineering viewpoint, in which 5 significant measure parameters is treated as network input, and 11 typical fault type is treated as output. In order to reduce training time and accelerate convergence speed, K-mean clustering and adaptive learning method is adopted to improve RBF neural network performance. Simulation and test result is shown, and comparison between RBF network and BP network is also discussed to validate the method.