Automated mode identification and tracking via frequency domain decomposition

• A novel algorithm for automated mode identification and tracking is introduced. • The method is suitable for a wide range of structural vibration monitoring tasks. • Based on FDD, tracking uses a robust correlation-driven framework. • A unified solution eliminates the need for separate AOMA stages. • Performance is validated through numerical tests and long-term field data. A novel algorithm for automated mode identification and temporal tracking is proposed for long-term applications in operational modal analysis and monitoring. The methodology, based on the frequency domain decomposition technique, provides a fully automated, unified solution that circumvents the limitations of conventional two-stage mode tracking routines relying on automatic operational modal analysis processes. The algorithm integrates a robust initialization scheme, allowing reference modal parameters to be defined either via manual input or through an innovative automatic modal identification procedure based on HDBSCAN clustering. Tracking is performed using a correlation-driven framework that recursively updates the reference modal parameters by assessing their alignment with the principal vectors derived by singular value decomposition of the cross-spectral density matrices computed from sequential acceleration response segments. The algorithm is validated through both numerical simulations and experimental testing on long-term vibration data from a truss timber bridge in Norway, a structure characterized by complex dynamics and substantial environmental and operational variability. The method demonstrates high detection rates, precise tracking of modal evolution, and strong resilience to temporary mode inactivity and abrupt shifts in modal characteristics. These results prove the algorithm’s potential as a powerful tool for long-term modal identification and tracking, suitable for large-scale data generation, statistical modelling, and fully automated structural health monitoring. The tracking phase requires only two user-defined parameters, basic information about the structure, and offers adaptability across diverse monitoring scenarios, substantially reducing the need for expert supervision.