Content area
Modeling structural damping remains one of the most challenging aspects of dynamic analysis, as damping cannot be directly derived from geometry or material properties and is commonly represented through simplifying assumptions. In practical applications, damping is often modeled using classical or proportional formulations, despite evidence that real structures may exhibit more complex dissipation behavior. At the same time, the growing availability of seismic response data from instrumented buildings raises the question of whether damping properties can be inferred directly from measured responses rather than prescribed a priori.
The objective of this thesis is not to characterize or improve physical models of damping in real buildings, but to evaluate whether a data-driven identification procedure can synthesize useful and dynamically consistent estimates of the mass-normalized damping matrix from partial seismic measurements. The central question addressed is whether such an inverse approach can recover damping models that reproduce global structural behavior, and whether these models remain practically interpretable.
The methodology estimates the damping matrix directly from the equations of motion by eliminating stiffness effects through a null-space projection. Missing response data due to limited sensor placement are reconstructed using cubic spline interpolation, and numerical stability is ensured through singular value decomposition–based regularization. The approach is first validated through controlled numerical simulations of eight-story and sixteen-story shear buildings subjected to multiple earthquake records. Because the true damping matrices are known in these simulations, the accuracy of the method is quantified using modal properties, cumulative energy dissipation, and peak displacement responses.
Results demonstrate that the method reliably reproduces global system behavior, with energy dissipation accuracy typically exceeding 85–90% and peak displacement errors remaining within acceptable engineering tolerances, even under sparse instrumentation. The framework is subsequently applied to an instrumented seven-story building using recorded seismic data, where physically plausible and numerically stable response predictions are obtained.
Overall, the study establishes that the proposed data-driven methodology can synthesize practically useful damping estimates from partial measurements, providing a reliable basis for response prediction in structural health monitoring applications.