Convolutive blind source separation in the frequency domain of mechanical noise for gas turbines based on bounded component analysis

Abstract

Noise source identification of gas turbines can provide the basis and guidance for vibration and noise reduction of gas turbines. Independent component analysis (ICA) is one of the most popular techniques for blind source separation (BSS) widely used in vibration and noise source separation in mechanical systems. ICA is suitable for independent source signals. However, in order to identify dependent mechanical noise sources in gas turbines, a convolutive BSS in the frequency domain based on bounded component analysis (BCA) is proposed. First, the basic theory of BSS and BCA is introduced in detail. The convolutive mixing in the time domain is transformed into an instantaneous mixing in the frequency domain by short time Fourier transform (STFT), and complex BCA is performed at each frequency bin. Second, a permutation alignment method based on local and global optimization is proposed to solve the problem of the permutation ambiguity. Finally, the accuracy and robustness of the proposed method are comparatively studied through typical numerical and experimental studies on a three-rotor experimental bench. The results show that the proposed method can effectively separate and identify independent and dependent source signals.