Robust Preprocessing: Denoising and Whitening in the context of Blind Source Separation of Instantaneous Mixtures

In this paper the feasibility and performance of two different preprocessing schemes for blind source separation (BSS) problems have been investigated. BSS is a promising technique for non-destructive machine condition monitoring by vibration and acoustic analysis. BSS methods offer non-invasive and inexpensive solution by restoring the specific signature of each machine from a mixture of signals which can be used for monitoring and fault diagnosis. Signals recorded by sensors in an industrial application are often disrupted by ambient noise and other mechanical systems. The presence of noise in the observation poses difficulties in the BSS process. Robust preprocessing involves denoising followed by whitening. Eigenfilters as well as wavelets have been used prior to the traditional independent component analysis (ICA) algorithm to test the separation ability for the instantaneous mixtures of synthetic signals, speech and acoustic noise from electromechanical systems. A comparative assessment with and without the denoising schemes have been presented.