Sensitivity Analysis and Optimization of User-Defined Parameters in Covariance-Driven Stochastic Subspace Identification

An accurate identification of structural modes is considered crucial for operational modal analysis, and stochastic subspace methods have been extensively employed for this purpose. In stochastic subspace identification, the number of block rows ([Formula: see text]) and the model order ([Formula: see text]) are two important parameters that affect the quality of the stabilization diagram. An index is proposed for optimizing these parameters based on the Covariance-driven Stochastic Subspace Identification (SSI-Cov). The stabilization diagrams obtained with the optimized parameters are characterized by fewer spurious modes. The influences of different parameter settings on the identification are then studied by automatically examining the stabilization diagram using hierarchical clustering. The variations in the estimated modal frequencies, damping ratios, and mode shapes are minimized with the optimized parameters. A computational scheme for the optimization of parameters in SSI-Cov is presented, with a reduction in subjective intervention. The SSI-Cov with optimized parameter settings does not require an excessively large model order for estimation, which also results in improved computational efficiency.