Experimental validation of system identification techniques for structural health monitoring by means of cameras and accelerometers.

In recent years, the development of computer vision technology has led to its implementation for structural identification. This thesis investigates the performance offered by the Phase Based Motion Magnification (PBMM), an algorithm developed by researchers at the MIT in Boston, that uses video acquisitions to estimate the displacements of target pixels within frames and to amplify the vibrations occurring within a desired frequency band. Thus, by means of a simple experimental set-up, it is potentially able to replace the pointwise measurements provided by sensors directly attached to structures, achieving high efficiency. In the current document, the results of this approach, applied to videos recorded with commercial-grade cameras, were compared with those obtained from an array of physically-attached accelerometers for the dynamic characterization of a three-storey, square-based aluminium frame. The main novelty of this thesis is the validation of PBMM-based structural identification procedure for structures with variable stiffness. For this purpose, the frame was stiffened by diagonal bracings, resulting in six different configurations. An impulsive excitation was provided by means of hammer tests. First, the modal identification of the different structural configurations was carried out on the data acquired from the accelerometers’ recordings, after which several observations on the effects of the bracing installation emerged. Subsequently, the PBMM was applied for the modal identification of the first two bending modes of the frame, taking advantage of the time series of virtual sensors extracted from the video acquisitions. The accuracy of the PBMM in the reconstruction of the vibrations occurring at these points and in the estimation of the modal parameters for all the configurations was then evaluated. For the modal detection, the Eigensystem Realisation Algorithm (ERA) was applied in Matlab environment. In addition, the influence of the pixel intensity on the quality of the estimates was also investigated, taking advantage of different brightness conditions within the videos. Finally, accelerometers and video-extracted recordings were used in combination to directly estimate the masses, stiffness coefficients and damping ratios of the 3-degrees-of-freedom frame; this was tested for the configuration without bracings. Specifically, the PBMM displacements were exploited to set the filtering parameters of the accelerometer recordings, reducing the error introduced by the accelerations double integration and optimising the estimations of the investigated quantities.