| description abstract | Most of the vibration-based ambient modal identification methods in the literature are structured to process vibration data collected from a dense array of sensors centrally to yield modal information. For large systems, for example bridges, one of the main disadvantages of such a centralized architecture is the cost of dense instrumentation, predominantly consisting of the sensors themselves, the data acquisition system, and the associated cabling. Recent advances in wireless smart sensors have addressed the issue of sensor cost to some extent; however, most of the algorithms—with the exception of very few—still retain an essentially centralized architecture. To harness the full potential of decentralized implementation, the authors have developed a new class of algorithms exploiting the concepts of sparsity (using wavelet transforms) within the framework of blind source separation. The problem of identification is cast within the framework of underdetermined blind source separation invoking transformations of measurements to the wavelet domain resulting in a sparse representation. Although the details of these decentralized algorithms have been discussed in other articles, in this paper, for the first time, these algorithms are studied experimentally on a full-scale structure using wireless sensors. In a truly decentralized implementation, only two sensors are roved along the length of a pedestrian bridge, and the performance of the proposed algorithms is studied in detail. A pedestrian bridge located in Montreal, Quebec, Canada, is chosen primarily to highlight the methodology used to address modal identification under low-sensor density and for pedestrian loading. Issues arising from several modes being excited on this bridge and the presence of narrowband pedestrian excitations are addressed. The accuracy of modal identification that is achieved using the proposed decentralized algorithms is compared with the results obtained from their centralized counterparts. | |
