Normalized Characteristic Parameters Extraction And Fault Diagnosis Of Gearbox Under Variable Operating Conditions

The gear box is widely used as a main transmission component in various rotating machinery.With the improvement of automation and intelligence,higher requirements have been put forward for intelligent fault diagnosis of gearboxes.The key to fault diagnosis under variable conditions is that the feature parameters are only related to the type and degree of the fault.However,the commonly used feature parameters can not achieve satisfactory results.As a kind of channel compensation technology,the Nuisance Attribute Projection(NAP)is earliest applied to speech processing.NAP can eliminate the effect of operating conditions on the feature parameters to the greatest extent,and then they can effectively identify faults after the projection.Therefore,based on the basic theory and application of Ensemble Empirical Mode Decomposition(EEMD),this paper proposes an improved EEMD method and combines NAP,transfer learning to diagnose gearboxs under variable conditions.The main research contents of this paper are as follow:(1)An improved EEMD method is proposed.The parameters of EEMD method is not adaptive,based on this,an improved EEMD method is proposed.Analyzed the influence of the maximum frequency of the white noise on the signal envelope.the improved EEMD method regards the amplitude and the maximum frequency of white noise as the parameter.This method fixes the overall average number to 2,and then determines the analysis range and step size of the noise amplitude and maximum frequency.After traversing,calculating the orthogonality of the decomposition results under different parameters.The decomposition result has minimum orthogonality is the finally result.The simulation and experimental signals prove that this method can suppress mode mixing better.(2)Presented a fault diagnosis method based on NAP in variable conditions.In this method,all the signal are decomposed by improved EEMD method and a feature matrix is formed by the feature parameters of decomposition results,and then a new feature matrix is obtained by NAP which has removed the influence of operating conditions and it is uesd to train the neural network to fault identification.The simulation and experimental signals verify the effectiveness of the method in fault diagnosis under variable conditions.(3)Presented a fault diagnosis method based on operating conditions in variable conditions.The method regards the operating conditions as the characteristic value for model training and establishes the relationship model between operating conditions,characteristic values,and the faults for fault identification.The effectiveness of the method was verified by the gear box vibration signal.(4)Presented a fault diagnosis method for small sample based on transfer learning.This method can improve the accuracy of the model trained through few target samples by adding many other working conditions’ auxiliary samples into the training.Moreover,aiming at the negative transfer,an improved method is proposed.It calculates the transfer degree of the auxiliary samples and keeps the samples have the large transfer degree.The effectiveness of this fault diagnosis method for small sample is proved by the gear box vibration signal.