Incremental dynamic analysis via hierarchical Bayesian modelling for efficient seismic fragility analysis and uncertainty quantification
提出分层贝叶斯增量动力分析框架，通过少量非线性时程分析预测完整响应，将计算成本降至传统方法的26%且误差小于4%，生成更真实的地震易损性曲线。
• Hierarchical Bayesian model (HBM) captures inter- and intra-ground motion variability in Incremental Dynamic Analysis (IDA). • Updated HBM robustly estimates full nonlinear time-history analysis (NLTHA) outputs even with limited NLTHAs. • HB-IDA limits errors in final seismic fragility curves to within 4% while requiring only about 26% of the computational cost of conventional IDA in the example. • By combining measured NLTHA results with Bayesian-predicted responses, HB-IDA generates fragility curves with credible uncertainty bands, yielding more realistic seismic performance estimates than conventional IDA. • A highly efficient intensity measure does not necessarily yield more reliable fragility estimates, as it may underestimate uncertainty in unobserved cases. Broadening the variety of NLTHA evidence yields greater improvements in uncertainty quantification than merely increasing the number of simulations. Seismic fragility analysis of structures is crucial for performance-based earthquake engineering. Incremental Dynamic Analysis (IDA) demands extensive nonlinear time‐history analyses (NLTHA) across multiple ground motion (GM) records and intensity levels, making it time‐consuming. To mitigate costs and quantify uncertainty in IDA, this study introduces a Hierarchical Bayesian IDA framework (HB-IDA). HB-IDA treats each ground motion as a basic unit and uses a two-level Bayesian model to capture both inter- and intra-GM variability. Exploiting similarity among GMs, HB-IDA fully scales only a small subset of motions (those with the shortest duration). The remaining motions undergo a single NLTHA run, and their missing responses are predicted by the Bayesian model. By combining NLTHA outputs with Bayesian predictions, HB-IDA produces fragility curves with clear uncertainty bands that better reflect realistic structural behavior, achieving a balance between accuracy and robustness. Moreover, the optimal intensity measure that fits the model best may not necessarily improve realism, and that widening the range of evidence yields greater benefits than simply running more NLTHA cases. Using underground tunnels within a representative liquefiable soil profile as case studies, this study highlights the potential of integrating surrogate models and hierarchical Bayes to provide efficient and robust fragility analysis.