BLIND SEPARATION OF NOISY HARMONIC SIGNALS USING ORTHOGONAL TECHNIQUES FOR ROTATING MACHINE DIAGNOSIS

P. Fabry, Ch. Servi`ere
e­mail : Christine.Serviere@lis.inpg.fr

We present a robust­to­noise technique for the blind source separation of harmonic signals, using temporal correlations. The mixtures are convolutive and cor­ rupted with spatially correlated noises. Their distribu­ tions and correlation functions are unknown. As the usual Principal Components Analysis cannot provide a correct estimate of the signal subspace in this situa­ tion, we introduce a new estimator of the decorrelation matrix and the signal subspace. It exploits the model of harmonic sources and uses non­hermitian interspec­ tral matrices of delayed observations. After whitening, the delayed observations are used again to determine the Givens plane rotations that complete the separa­ tion process. A Wiener filtering is then applied for denoising the outputs. The proposed method is effi­ cient for low Signal to Noise Ratio and illustrated with simulations and rotating machine vibration data.