Output-only operational modal analysis for cumulative seismic damage tracking in reinforced concrete structures

This thesis investigates output-only dynamic identification for monitoring damage evolution in full-scale, three-dimensional, single-bay, three-story reinforced-concrete (RC) buildings subjected to shaking table tests and evaluates the effectiveness of seismic retrofitting. Two nominally identical RC frames, one as-built and one retrofitted, were evaluated under controlled laboratory conditions. The frames were subjected to repeated low-level white-noise excitations, alternating with earthquake records of increasing intensity, using dense accelerometer instrumentation on both structures. The data were then reoriented to a common frame, time-aligned, uniformly resampled, and anti-alias filtered to make them suitable for modal analysis. Modal parameters at each stage were estimated using Frequency-Domain Decomposition in combination with covariance-driven Stochastic Subspace Identification as part of the operational modal analysis. To evaluate robustness and reproducibility, the complete identification workflow in Python (PyOMA2) and MATLAB (OoMA Toolbox) for cross-validation. Reference properties for comparison were provided by a finite-element baseline implemented in the Scientific Toolkit for OpenSees (STKO) software. The results reveal clear, progressive stiffness degradation with increasing seismic excitation in both specimens. In the case of the retrofitted building, a stiffer response was observed, which highlights the effectiveness of the retrofit. Damping estimates continued to fall within the ranges reported for similar RC frames, and agreement between experimental results and numerical models remained strong. Although mode-shape estimates did not perfectly match the finite-element predictions, the Modal Assurance Criterion (MAC) showed well-separated principal modes and stable consistency across tests. That suggests the identification was physically meaningful even if some local differences were present. Residual differences are possibly due to sensor layout, mild nonlinear effects under stronger inputs, and modeling simplifications. The work presents a reproducible workflow from sensing and preprocessing to identification and validation and provides practical recommendations for monitoring cumulative earthquake damage in buildings, including guidelines on sampling, decimation, and model stabilization choices for robust OMA-based tracking.