Quasi-instantaneous Operational Modal Analysis of Bridges Through AI-Driven Blind Source Separation

Managing aging infrastructure has become a top priority on the political agenda due to alarming incidents like the Genoa Bridge collapse in 2018, underscoring the critical need for efficient structural maintenance strategies. To address this challenge, vibration-based structural health monitoring (SHM) approaches and, in particular, operational modal analysis (OMA)-based systems are becoming especially popular owing to their non-invasive and non-destructive attributes. OMA systems exploit ambient structural response data to extract modal characteristics (e.g., resonant frequencies, damping ratios, and mode shapes), which serve as damage-sensitive features. In this context, new advances in the realm of artificial intelligence (AI) have opened exciting new possibilities for the development of next-generation SHM systems. Despite still in its early stages, a few pioneering studies in the literature have hinted at AI’s potential to enhance OMA methods, allowing for superior damage detection while minimizing human involvement. This research introduces an innovative AI approach based on blind source separation (BSS) with quasi-instantaneous OMA capabilities. Unlike previous research in the literature, the architecture of the developed network is defined in a multi-task fashion to extract both the real and imaginary parts of the independent modal components from ambient acceleration data.