Automatic modal parameter identification and associated uncertainty quantification for civil structures via SSI-COV and Deep Learning

Accurate estimation of modal parameters and associated uncertainties is essential for achieving reliable structural health monitoring. Conventional identification techniques often depend on subjective judgment and lack systematic uncertainty quantification, leading to inconsistent results and limited reproducibility, which ultimately hampers automation. An automated post-processing framework is developed by integrating covariance-driven stochastic subspace identification (SSI-COV), clustering-based uncertainty label generation, and YOLOv10s-based modal-cluster detection. The clustering-based method is used to generate bounding-box labels during training, whereas the trained YOLOv10s detector directly predicts modal centers and uncertainty-related bounding boxes from unlabeled normalized frequency-damping candidate-pole maps during inference. Numerical simulations and field measurements on a high-rise building and a long-span bridge demonstrate accurate automated modal identification with reliable uncertainty quantification. The proposed method enhances automation and reliability in uncertainty-aware modal analysis for civil infrastructure monitoring.