Research On Feature Extraction Of Noisy Acoustic Emission Signals Based On Independent Component Analysis

Acoustic emission detection technology is a kind of safe and efficient dynamic nondestructive detection technology, which is widely used in industrial and agriculture areas.With multiple acoustic emission sources in detection, the results of the measurement are mixed signal of multiple acoustic emission sources, and the noise are common existence in nature, which makes the detection results more complicated and not easy to analyze and deal with later. In this paper, the blind source separation of the acoustic emission signals with noise is carried out by using the method of independent component analysis, and the characteristic frequency is extracted and analyzed. In experiments,the acoustic emission signals from different parts fault of rolling bearing from rotating machinery fault simulation test platform are used for major acoustic emission signal source and the SAEU2 S acoustic emission system is used for the detecting instrument.Firstly, using the FAST-ICA algorithm based on maximum negative entropy, selecting the similar coefficient, secondary residuals and performance index(PI) constitute a set of evaluation indicators to analyze separation results and evaluate algorithm performance from different angles.The blind source separation for sine signal, simulation of acoustic emission signal which are added random white noise and measured rolling bearing acoustic emission signals are carried out respectively, and the characteristic frequencies of separation results are extracted by spectrum analysis. From the point of view of simulation results and evaluation index, the FAST-ICA algorithm basically realizes the blind source separation of acoustic emission signals with noise, and retains the characteristic frequencies of the signals.Secondly, considering the influences from random noise and the time difference between acoustic emission signals from different sensor, the high order statistics method are introduced to blind source separation process, can probably reduce the influence which caused by Gaussian noise. In this paper, joint approximate diagonalization of eigen-matrice(JADE) algorithm based on fourth-order cumulant matrix is used for blind source separation and the joint approximate diagonalization of eigen-matrice process can probably reduce the influence from time difference. As can be seen by contrasting with FAST-ICA algorithm, the acoustic emission signals separation results of JADE effectively retains the characteristic frequencies from rolling bearing on different work state, and restrains the noise influence, which can facilitate the state analysis and fault diagnosis.