Fusing physics-based and machine learning models for rapid ground-motion-adaptative probabilistic seismic fragility assessment

Abstract

Performance-based earthquake engineering (PBEE) has asserted probabilistic seismic fragility assessment (PSFA) as the main research content in light of its irreplaceable significance for seismic decision-making in recent decades. Among the multiple approaches of PSFA implementation, the classical linear regression method (LRM) dominates over practice regarded as one of the most widely-used. However, the general LRM adopts quantile regression method (QRM) on the group of fragility curves to approximate a deterministic probability density distribution (PSD) of structural fragility against certain intensity measure (IM) of potentially confronting earthquake. Consequently, the QRM-derived fragility representation might not be credible enough while evaluating a newly-occurred seismic event owing to its neglect of specificity of stochastic ground motion. To address this issue, a fusing physics-based and machine learning models towards rapid ground-motion-adaptative probabilistic seismic fragility assessment (GmaPSFA) is proposed in present study. With sophisticated framework design and novel fragility hyperparameters estimation, the involved design philosophy and mechanism translating are both elaborated. To validate the method, both the LRM and GmaPSFA were conducted on a six-story frame structure, where a novel fully-automatic batch processing approach fusing APDL and coding languages was propounded for structural analysis. The comprehensive validating cases confirmed the powerful capability and significant superiority of proposed GmaPSFA in realizing rapid ground-motion-adaptative probabilistic seismic fragility assessment catering for modern PBEE.