Power spectrum density transmissibility (PSDT) functions have attracted widespread attention in operational modal analysis (OMA) because of their robustness to excitations. However, the selection of the peaks and stability axes are still subjective and requires further investigation. To this end, this study took advantage of PSDT functions and support-vector machines (SVMs) to propose a two-stage automated modal identification method. In the first stage, the automated identification of peaks is achieved by introducing the peak slope (PS) as a critical index and determining its threshold using the SVM classifier. In the second stage, the automated identification of the stability axis is achieved by introducing the relative difference coefficients (RDCs) of the modal parameters as indicators and determining their thresholds using the SVM classifier. To verify its feasibility and accuracy, the proposed method was applied to an ASCE-benchmark structure in the laboratory and in a high-rise building installed with a structural health monitoring system (SHMS). The results showed that the automated identification method could effectively eliminate spurious modes and accurately identify the closely spaced modes. The proposed method can be automatically applied without manual intervention, and it is robust to noise. It is promising for application to the real-time condition evaluation of civil structures installed with SHMSs.
Enhancing two-stage modelling methodology for loss given default with support vector machines
