Abstract

Abstract: In the last few years, thanks to progress in computational power, optical methods have found extensive application in the field of structural dynamics; the main advantages reside in high-resolution or even full-field measurements of displacement or velocity, enabling the capture of modal vibration patterns of the structure, and, since they entail non-contact data acquisition, in the ability to test under operational conditions where conventional sensors are difficult to deploy. However, current implementations still face limitations due to measurement noise, illumination variability, and camera-motion artifacts, as well as the computational cost of high-resolution processing, leaving room for significant improvements in robustness and efficiency.

This talk will present state-of-the-art computer vision approaches and our recent work to estimate resonant frequencies and extract operational deflection shapes via Video Motion Magnification (VMM). The proposed methods, developed in part during my research visit to the University of Cambridge, are based on Dynamic Mode Decomposition and a deep learning VMM model combined with signal processing; they represent a step forward in Modal Identification via Computer Vision, although some limitations and open questions concerning reliability in challenging conditions remain to be addressed.