Abstract: The BP neural network,used in the study of fault diagnosis for rotating machinery,is easy to fall into local minima,and then the global optimal solution could not be obtained,that results in a poor classification and recognition rate of couple rub impact fault.Based on the analysis of an Empirical Mode Decomposition (EMD) method and a BP-AdaBoost method, puts forward a new algorithm of fault identification combined with the advantages of them above.First of all,an Empirical Mode Decomposition(EMD) method is used to decompose vibration signal from the rotor,for the aim of removing background noise signal,so the fault feature of rub impact signal of rotor system is obtained.Then a BP-AdaBoost model is used for the recognition of three kinds of different conditions.The analysis results on the experiments data show that the recognition rate of this new method is better than that of a single BP neural network.

Key words: coupling rubbing, fault features, Empirical Mode Decomposition(EMD), BP-Adaboost