Research On Acoustic Emission Recognition Based On Sparse Representation Theory

With the development of science and technology,rotating machinery plays an important role in large-scale equipment manufacturing,energy,metal processing,civil projects and so on.So it is very important to detect the rubbing fault of rotating machinery.The acoustic emission testing technology is a kind of method with great potential for the fault diagnosis of rotating machinery.Due to the diversity of acoustic emission sources and the strong noise interference,it is difficult to extract and identify the characteristic signals.The above-mentioned problem has been studied in this paper.(1)Sparse representation is introduced into the acoustic emission identification technology.The experimental results show that the sparse representation can effectively improve the performance of the acoustic emission identification system.(2)Using K-SVD dictionary learning algorithm on the grinding acoustic emission signal denoising.In the training phase,the use of K-SVD algorithm for training the acoustic emission signal overcomplete,then according to the estimation of noise variance,the orthogonal matching pursuit algorithm is used to decompose the noisy acoustic emission signal in the dictionary,rubbing.The acoustic emission signal and noise are separated.The simulation results show that the acoustic noise reduction algorithm based on K-SVD dictionary learning algorithm can improve the signal to noise ratio(SNR).(3)Using the MFCC coefficient and the Gauss mixture model is the first step to construct the signal.Then based on the K-SVD algorithm,this paper introduces the acoustic emission identification system based on D-KSVD,which adds the discriminant to improve the recognition rate.The accuracy of the identification results is verified by experiments,and the requirements of the state information monitoring and fault diagnosis for the grinding operation of rotating machinery are achieved.(4)In order to overcome the shortcoming of the sparse representation dictionary construction,the Fisher discriminant dictionary learning algorithm is used to construct a discriminative dictionary.A detailed description of the Fisher discrimination dictionary learning algorithm and dictionary update and characteristics is given.Based on the identification of acoustic emission,an acoustic emission identification system based on Fisher discriminant dictionary learning algorithm is proposed.The experimental results proved that the algorithm effectively improves the recognition result.(5)Classification algorithm for the identification of acoustic emission is based on deep learning.Deep learning model is sparse autocoder network.The front part is constructed using a sparse auto encoder,followed by Softmax regression as a classifier The algorithm uses first-order differential MFCC as classification feature.The experimental results show that the algorithm not only has high recognition rate,but also has strong robustness.