Noise source identification and localization of mechanical systems based on an enhanced independent component analysis

In this paper, an enhanced independent component analysis (EICA) is comparatively studied with the traditional fast independent component analysis algorithm, and a noise source identification and localization method based on the EICA and spectral correlation analysis is proposed. The EICA selects the optimal separations using clustering analysis from multiple source separations, and the robustness and effectiveness of the EICA are validated by a numerical case study. The proposed noise source identification and localization method firstly separates the mixed noise signals measured outside of a mechanical system, which guarantees an easy and complete measure of all the source information and an accurate source separation. Secondly, it evaluates the separating performances by time and frequency feature analysis and waveform correlation analysis. Finally, it adaptively identifies and localizes the noise sources by spectral correlation analysis and priori information of the mechanical system. The effectiveness of the proposed method is validated by experimental studies on a test-bed, and this study can be beneficial for vibration and noise monitoring and the control of mechanical systems.