A Graph Signal Processing Technique for Vibration Analysis with Clustered Sensor Networks

The modal analysis of large structures, because of spatial and electrical constraints, generally requires cluster-based networks of sensors. In such solutions, dedicated procedures are required to reconstruct the global mode shapes of vibration starting from the local mode shapes computed on individual groups of sensors. Commonly adopted strategies are based on overlapped schemes, in which at least one sensing position is shared among neighbour clusters. In this paper, a non-overlapping monitoring approach is proposed. It relies on the intrinsic capability of graph signal processing to encode structural connectivity on edge weights and exploits the maximization of the global graph signal smoothness to define the best set of scaling factors between adjacent networks. Experiments on a pinned-pinned steel beam in condition of free vibrations proved that the proposed method is consistent with respect to numerical predictions, showing great potential for distributed monitoring of complex structures.