1 ECCM 2010 IV European Conference on Computational Mechanics Palais des Congrès, Paris, France, May 16-21, 2010 Opportunities and challenges of Blind Source Separation techniques for dynamic parameter identification and monitoring C. Rainieri 1 , G. Fabbrocino 2 1 Structural and Geotechnical Dynamics Lab « StreGa », University of Molise, Termoli, Italy, carlo.rainieri@unimol.it 2 Structural and Geotechnical Dynamics Lab « StreGa », University of Molise, Termoli, Italy, giovanni.fabbrocino@unimol.it An accurate knowledge of modal properties is helpful for a number of engineering applications, such as structural design, validation of numerical models and Structural Health Monitoring (SHM). Output-only techniques are very effective to this aim, in particular for applications where a known input is not available. It is the case of large civil structures, heritage structures, satellite and spacecraft structures in operation. Several output-only methodologies are proposed and several successful applications are reported in literature. Beside the classical time domain and frequency domain techniques, a recent proposal in this field is related to the use of the Blind Source Separation (BSS) techniques for output-only modal analysis purposes. Blind Source Separation consists in estimating a set of signals, called sources, using observations of some mixtures of these signals [1]. Only fairly general assumptions are made about the sources and the mixing process. BSS techniques can be classified as linear [2] or non-linear [3], according to the type of combination of the sources. Moreover, linear simultaneous (static) mixing [2] and convolutive mixing [4] can be considered. In recent years an increasing number of applications of BSS to structural dynamics has appeared in the literature. They included damage detection [5], condition monitoring [6, 7], discrimination between pure tones and sharp-pointed resonances and applications in the field of output-only modal analysis [8]. However, use of BSS techniques in structural dynamics is still a challenge [9] mainly due to the fact that the time response of structures is related to the excitation through a convolutive mixture, which is much more difficult to treat than a static mixture. Nevertheless, different BSS algorithms are being applied in the field of output-only modal analysis, pointing out, in some cases, promising results. In fact, about Applicability of BSS to vibration data, it has been proved by taking into account that the physical responses {x(t)} are related to the modal responses {q(t)} by the modal matrix [Φ]: () { } [ ] ( ) { } t q t x Φ = (1) Even if the dynamic response of a system [x(t)] is given by the convolution product of the impulse response function [h(t)] and the external forcing vector [f(t)], the response of a dynamic system can be interpreted as a static mixture of sources [10]. In fact, assuming linear and static mixtures, the noisy model can be expressed in matrix form as: () { } [ ] ( ) { } ( ) { } t t s A t x σ + = (2)