Recognition of pump state by RBF neural network

On the basis of the traditional BP and RBF neural network, a new algorithm-user-defined step radial basis function was developed to monitor and recognize the three state (normal, wear and abnormal state) of vacuum air press pump ZYB03-60, then diagnosed the fault of it. The feature was extracted by sorting, learning and discriminating of the vibration signal. The frequency factor was taken into account in the new method. Therefore, the entire initial central node was adjusted and moved to the data area. This avoided the "dead center" which came from the traditional method with the principle of the nearest distance. In addition, the algorithm punished the competitor of the winning node and realized the choice of node automatically. Results indicate that the three states are presented in different area. The node reaches steady state with training 400 times and this is much faster than the traditional (1000 times of traditional).