Bayesian model uncertainty analysis of fluid viscous dampers and its influence on retrofitted reinforced-concrete frame fragility

Fluid viscous dampers are widely used to mitigate the seismic response of civil structures, but constitutive-model choice, parameter uncertainty, and residual modelling error may affect structural fragility estimates. This study develops a Bayesian workflow for calibrating and comparing multiple damper constitutive laws and for examining the influence of selected damper-parameter uncertainty on frame-level fragility. Nine representative damper models, ranging from algebraic velocity laws to differential hysteretic formulations, are calibrated using cyclic test data from mass-produced 300 kN dampers. The comparison considers predictive agreement, residual statistics, parameter identifiability, posterior uncertainty, physical interpretability, parsimony, and feasibility in nonlinear time-history analysis. A Bouc–Wen-type LAW6 model is retained as the representative propagation model for a three-storey reinforced-concrete frame retrofitted with fluid viscous dampers and subjected to 150 selected ground-motion records. The fragility study distinguishes record-to-record variability from epistemic sources related to damper calibration, likelihood assumptions, model-form uncertainty, and structural modelling. The frame-level uncertainty bands are interpreted as LAW6 joint-posterior parameter-uncertainty results. Complete LAW6 parameter vectors are sampled directly from the full-data TMCMC posterior and propagated through the 150-record nonlinear time–history workflow, thereby preserving posterior dependence among the constitutive parameters. Across 100 complete vectors, the 95% ranges of the median-intensity shift relative to the posterior-mean reference remain within −0.37% to ＋0.36% when all 150 records are used. A conservative 84-record diagnostic screen changes the absolute PSDM and fragility parameters, but the corresponding shifts remain within −0.44% to ＋0.40%. Relaxing the upper support of 𝛽 from 200 to 400 m −1 likewise leaves the all-record shifts within −0.42% to ＋0.38%. Under the adopted calibration, LAW6 parameter uncertainty therefore has a smaller influence than record-to-record variability on the drift-based fragility estimates; the absolute fragility parameters remain conditional on the PSDM-processing choice. These results do not represent model-form uncertainty propagation across all nine candidate laws.