Abstract:

To overcome intrinsic shortcomings of back propagation(BP)neural network, including slow convergence rate and easy trapping in local minimum, an empirical mode decomposition (EMD)genetic neural networks method is proposed. Firstly, EMD is used to decompose the signals with noise to obtain each intrinsic mode function, each intrinsic mode function corresponding to a frequency band with different energy or a fault feature, and feature vector of each frequency band is used as input sample to optimize neural network. Secondly, the genetic algorithm is used to optimize the weights and thresholds of BP neural network. This method is applied in a simulating experiment for the rolling bearings multiple fault signal analysis, and the ability of fault identification is therefore improved by this method.

Key words: empirical mode decomposition, intrinsic mode function, neural network, genetic algorithm