Research On Bearing Acoustic Emission Feature Extraction And Sound Source Localization Technology

The working environment of aero-engine spindle bearing is bad,and the early weak fault will quickly evolve into a serious fault.The early weak fault diagnosis of aero-engine spindle bearing is the key technology of aero-engine safety and predictive maintenance.Aeroengine bearing is large in volume and complex in structure,which is inconvenient to disassemble directly to check the damage location and severity.In this paper,acoustic emission technology is used to realize in-situ detection and location of bearing fault.Firstly,multiple features of bearing acoustic emission signal are extracted to train neural network to complete rough location of fault source.In order to further realize accurate location of fault source,a real-time fault detection system is built The array sparse representation method is studied to realize the accurate location of AE fault source.The main results are as follows:1.Aiming at the rough location problem of AE fault source,this paper adopts the intelligent location method of combining AE feature extraction with BP neural network to extract the amplitude domain,frequency domain and time-frequency domain feature parameters of AE signals of different fault bearings under actual working conditions as network inputs,and realizes the fault source location classification.In order to improve the positioning effect of rolling element fault and cage fault,this paper proposes to use IMF component with larger energy value to calculate energy entropy and local average frequency to expand the feature vector sample library.The experimental results show that the fault classification BP network can converge quickly,and the accuracy of fault source classification reaches 84.7%.2.Aiming at the problem of accurate location of acoustic emission fault source,this paper constructs a nested array model applied to the near-field environment,proposes IMUSIC L1-SVD method for near-field acoustic emission source location,and realizes the acoustic source location with the number of sources greater than or equal to the number of array elements.In this paper,Gabor wavelet transform is used instead of CSSM to process wideband signals,which avoids the pre estimation errors of the number of AE sources,incident angle and distance,and reduces the time cost of constructing focusing matrix.In this paper,IMUSIC method is used to construct the weight matrix,so that the larger value in the source signal can accept the smaller weight,and the smaller value can accept the larger weight,which reduces the error between L1 model and L0 model,reduces the correlation between signals,and eliminates the interference of noise subspace to signal subspace.The experimental results show that the method can accurately locate the fault source in an average time of 4.3s,and the distance estimation error is less than 2cm.