State-dependent seismic fragility functions for bolted flange joints on special-risk industrial substructures

Seismic vulnerability assessment of industrial plants and process equipment has gained attention lately. However, the complexity of the problem and its modelling, combined with a general scarcity of available data on industrial systems, contributed to limiting or developing risk assessment methods based on extremely simplified models. On these premises, a new methodology that combines limited data from finite element (FE) models with cutting-edge machine learning techniques is developed to generate state-dependent fragility functions for critical components mounted on archetypical industrial substructure modules. Specifically, referring to the classical forward uncertainty quantification scheme, a physics-informed FE model, calibrated on the results of a comprehensive shake table test campaign of a real-scale industrial braced-frame (BF) steel substructure, is considered. Time histories of seismic event sequences generated by a site-based ground motion model compose the input provided to the FE model. The uncertainty is then propagated through Monte Carlo analysis on inexpensive-to-run polynomial chaos expansion (PCE) metamodels set for each starting damage level condition measure. As a result, empirical state-dependent fragilities are evaluated, assuming the lognormal distribution.