PSI - Issue 44

Marialuigia Sangirardi et al. / Procedia Structural Integrity 44 (2023) 1602–1607 M. Sangirardi et al./ Structural Integrity Procedia 00 (2022) 000 – 000

1605

4

The semi-automated post-processing can involve two different parameters, such as the grey intensity and the motion of a set pixels, appropriately selected in a region of interest (ROI). They were both deduced from the video recording and, in the presented analyses, were examined simultaneously having the intensity variation tracking provided more consistent results. In both cases, the procedure follows the following steps: the video is first recorded, paying attention to light conditions (which need to be as constant as possible) and motion of the camera (which should be, in principle, null so to identify pure structural vibration). Then, the video is pre-processed through a motion magnification algorithm, which consists in a fictitious amplification of the movements within a pre-determined range of frequency and through an amplification factor whose value is fixed a priori. In this stage, some engineering judgment is needed to properly enhance the dynamic motion of the structure to be analysed. To this end, in this case, the outcomes of analytical analyses and numerical simulations were used to select a motion magnification frequency window between 4 Hz and 8 Hz. The amplification factor was set equal to 40. After magnification, the video is post processed referring to n frames, in which either the intensity or the motion of selected pixels is tracked, obtaining m time-histories, being m the number of pixels. Fig. 3 reports a sample frame of the recorded video (a), the entropy map (b) according to which the ROI for intensity variation analysis is selected (Fioriti et al. 2018, Sangirardi et al. 2022) and the pixels whose motion is tracked (c).

Fig. 3. Video processing stages: sample frame (a), entropy map (b) and selected pixels for motion tracking (c) .

Recorded digital videos can indeed contain a number of irrelevant information, i.e. related to noise, which can bias the results of the frequency analysis. These unwanted components can be originated by the acquisition instrument, by the resolution of the camera, or by light conditions in the test environment and can render the identification of the most relevant frequency components very cumbersome. Hence, the time histories representative of the motion of the structure are then independently correlated and analysed via a Principal Component Analysis, whose purpose is that of extracting the most relevant information which, in this case, are finally translated to the frequency domain, through a Power Spectrum Analysis. 4. Analyses and comparisons with accelerometric measurements In order to monitor the velocity of the structure during the activities of a neighbouring construction site, the structure was equipped with two MEMS (Micro Electro-Mechanical System) triaxial accelerometers at two different elevations, such as 15.50 m and 20.0 m (Fig. 1a). The accelerometers were powered by a photovoltaic panel, had  2g range and recorded data at 320Hz sampling frequency. Only the former one was continuously operating while the latter one only starts registering when triggered over a pre-determined threshold. For the sake of the study presented herein, only one of the two sets of registrations was analysed, namely the one continuously operating. Acceleration time-histories of 2 minutes duration were extracted, filtered through a 2nd order baseline correction