Fault diagnosis of wave sensor based on wavelet packet and improved RBF neural network

This paper presents a method to diagnose faults of wave sensor. Firstly, the simulated fault signal of wave sensors is decomposed by wavelet packet, then the feature vector of the fault signal is constructed, the feature vector is inputted into the improved RBF neural network. The improve RBF neural network adopts a mean shift algorithm to solve the problem of exactly determining the center vector of a traditional RBF neural network. Finally the results are visualized by t-sne dimensionality reduction method. Experiments show that this method can effectively determine the center vector of RBF neural network, and the accuracy of fault type recognition can be boosted.