Processing sensor data, from multiple, non simultaneously recorded measurement setups, for structural analysis is often achieved by merging identification results obtained from records corresponding to different sensor pools. Since pole matching and eigenvector glueing may raise difficulties in some cases, it is tempting to try merging data first and then processing them globally. In this paper we present such an approach, for subspace-based identification algorithms. Subspace identification algorithms have proven useful for the output-only identification of the eigenstructure of a linear MIMO (Multiple Input - Multiple Output) system subject to uncontrolled, unmeasured, nonstationary excitation, a situation encountered in structural identification of vibrating structures subject to natural excitation.  This paper investigates how subspace algorithms must be adapted to cope with the use of multiple measurement setups, under nonstationary excitation. We present the theory and discuss realistic examples.

Keywords : output-only structural identification, subspace algorithms.