Automated modal analysis of an arch dam using a hybrid SSI framework based on error analysis and fuzzy clustering

This study investigates the recent status of Pacoima arch dam using an automated covariance-based stochastic subspace identification method. Since existing OMA techniques rely on the assumptions of white noise and stationary inputs, they often introduce uncertainties. To address this, the proposed approach integrates multiple approaches to enhance accuracy and minimize these uncertainties. The first approach uses sensitivity analysis to determine the Toeplitz matrix dimensions and maximum system order. A figure of merit analyzes noise indicators in the SSI method, showing convergence dimensions of the Toeplitz matrix. Also, this sensitivity analysis evaluates the condition number of observability matrix to determine the maximum order for identifying dynamic models and plotting a stabilization diagram. The second approach introduces a distance function to identify stable primary dynamic models and plot them in the stabilization diagram. The count of axes in the diagram defines the number of stable modes. The third approach validates mode shapes identified in the previous step using the modal complexity factor and determines their imaginary degree. The final approach employs the fuzzy c-means clustering algorithm to remove spurious modes and extra noises from initial models, extracting the final dynamic models. The findings indicate that the natural frequencies of the identified dynamic models closely match those of the FEM and previous research, with an average difference of less than 8%. Despite the closely spaced nature of the modes in this dam, the identified damping ratios show minimal differences compared to other studies, particularly the 2002 forced vibration test.