Two-stage DBSCAN clustering method for online automated operational modal analysis of arch dams under ambient excitation

Continuous identification of modal parameters of large civil structures under ambient excitation is essential for structural health monitoring (SHM). In practice, however, automated operational modal analysis (OMA) rarely delivers clear and directly applicable mode-evolution curves in real-time online. In this paper, a two-stage clustering method is proposed for online OMA. The method first applies covariance-driven stochastic subspace identification (SSI-COV) in short time windows to extract modal points. On this basis, density-based spatial clustering of applications with noise (DBSCAN) is performed at two different scales (single and cross windows), define as two stages in this study. The mode symmetry coefficient (MSC) is further used as a geometric constraint for symmetrical structures. A one-year dataset of vibration measurements from a concrete gravity arch dam is employed to systematically investigate the effects of clustering control parameters on the performance of the proposed method. Moreover, a quality evaluation metric is developed to jointly evaluate the bandwidth and density of the identified mode-evolution curves, leading to the determination of practical parameter ranges. The transferability of both the method and the parameter recommendations is then demonstrated on a concrete arch dam with different excitation levels and closely spaced modes. The results show that the proposed method can automatically produce clear mode-evolution curves in real-time online, significantly suppress spurious modes, and reliably separate closely spaced modes, providing a robust and reproducible solution for online modal parameter tracking in long-term structural monitoring.