BLIND SEPARATION OF EXCAVATOR NOISE BASED ON INDEPENDENT COMPONENT ANALYSIS

In order to identify excavator noise sources under non-laboratory environment, noise signals in frequency domain were separated based on Independent Component Analysis (ICA). Firstly, experiments were carried out in a manufacture plant and excavator noise signals were acquired, which had been interfered with by drastic echo and background noise. Secondly, signals in time domain were transformed into frequency domain via Fourier transform (FT), so that convolution mixtures were turned into linear mixtures. Thirdly, these linear mixtures were separated into principal components by Fast fixed-point independent component analysis (FICA). Finally, a comparison of pricipal components and the result of Ansys modal analysis was conducted. Research shows that seperation of excavator noise signals based on ICA in frequency domain is effective, and noise sources can be identified properly by comparing basic frequencies of independent components with the result of modal analysis.