Development of Refined Data-Driven Stochastic Subspace System Identification for Buildings and Bridges

Abstract

Frequent large-scale earthquakes, climate changes, manmade hazards, and the duration of the service are the possible origins of structural damage in Taiwan. To detect the changing features and damage states of the structures, the demand for understanding the unknown system models of the operating structures has risen. The accuracy of structural health monitoring has become a significant issue. Therefore, four kinds of refined data-driven stochastic subspace system identification (SSI-DATA) methods, namely the mode-by-mode methods, are proposed in this research. Because the mode-by-mode methods only extract a single mode per iteration, the “mode elimination” and “signal reconstruction” steps are added to the traditional SSI-DATA. The mode elimination is realized by removing the singular components that have been exploited in the identified mode. Meanwhile, the signal reconstruction employs a similar approach used in the singular spectrum analysis after the Hankel matrix is regenerated with the removal of identified modes. Moreover, the effective projection operations and modification of the singular value decomposition process are employed in the refined methods. A unified analysis procedure is also introduced to automatically extract all the concerned modes one by one using the methods, while the errors between the reference frequencies and identified frequencies and the calculated frequency resolutions are the criteria for selecting modes. To verify the proposed methods, cases of a simulated eight-story frame and the actual operating bridge structure are studied using the proposed system identification methods. Consequently, the identification results show that the refined methods can yield slightly more accurate modal parameters of the structures. Moreover, the computational time of the second, third, and fourth methods is much less than the traditional SSI-DATA.