Blind separation of vibration components: Principles and demonstrations

Blind source separation is the issue of recovering the various independent sources exciting a system given only the measurements of the outputs of that system. It has recently become the focus of intensive research work due to its high potential in many applications. However when it comes to vibration signals, blind source separation faces a number of difficulties which seriously hinder its feasibility. In this paper we first draw a list of these difficulties. We then propose to somehow lessen the ambitions of BSS to more realistic goals, and in particular to separate vibration signal into contributions of (1) periodic, (2) random stationary, and (3) random nonstationary sources. Although not ideal, such a separation can provide substantial information in many practical cases of interest. Towards this aim we propose two robust separation techniques based on the short-time Fourier transform. They are perfectly suited to handle convolutive mixtures of sources and they lead to fast and efficient implementation.