An automated signal quality–driven framework for decomposition-based modal identification using ambient vibration data

Accurate identification of modal parameters from ambient vibration data is critical for the assessment and monitoring of large-scale civil structures. However, long-term structural health monitoring data often suffer from noise, non-stationarity, and low-information intervals, which can compromise the reliability of operational modal analysis (OMA). To address these challenges, this study proposes an automated signal quality–based framework that integrates an aggregated statistical indicator, including standard deviation, skewness, kurtosis, root mean square value, crest factor, and total frequency energy, with multivariate variational mode decomposition to enable robust modal identification using short data segments. Multichannel acceleration data are segmented and ranked based on multiple statistical and signal-based features to automatically select the most informative segment across all sensors. The selected segment is then analyzed to extract natural frequencies, damping ratios, and mode shapes while preserving inter-channel dependencies. The proposed framework is validated using full-scale ambient vibration data from three large civil structures (Milad Tower, Canton Tower, and Hardanger Bridge). The results demonstrate stable and accurate modal identification from short data records, with consistent modal parameters across different segment durations. Comparative evaluations against established OMA methods (FDD, EFDD, SSI-COV, and SSI-DATA) and conventional signal processing techniques (VMD, EMD, PSD, and FFT) confirm that the proposed approach achieves superior performance, particularly in identifying higher-order modes, while maintaining stable and reliable performance under varying environmental and operational conditions in long-term monitoring scenarios.