Nonparametric seismic fragility curves for structural systems using ordinal regression

초록

This study presents a nonparametric ordinal regression framework for constructing seismic fragility curves for structures. This approach preserves the ordered structure of the damage states without relying on predefined distributional assumptions, as in traditional methods. The proposed methodology employs a nonparametric maximum likelihood estimation coupled with kernel density estimation to estimate a flexible link function, thereby avoiding assumptions regarding a specific error distribution. Three numerical case studies were considered: a four-story reinforced concrete (RC) moment-resisting frame building, a three-span concrete box-girder bridge, and a four-story steel special moment-resisting frame building. Each structural system was analyzed using nonlinear time-history analyses with 320 ground motions, and the damage states were defined based on the simulation results. The proposed nonparametric ordinal model (NPOM) achieved test accuracies of 82 %, 83 %, and 81 % for the RC frame, bridge column, and steel frame, respectively, outperforming the ordinal probit model (OPM), which achieved test accuracies of 78 %, 78 %, and 76 %, respectively. Furthermore, seismic fragility curves for the three structures were developed and compared using two ordinal models (NPOM and OPM) and the traditional linear regression method (LRM) under the assumption of a lognormal distribution. For all three structures, the LRM-based fragility curves exhibited higher median fragility values for higher limit states than the fragility curves derived from the ordinal models. For the RC and steel frames, OPM yielded median fragility values that differed from those of NPOM by up to approximately 12 %, whereas for the bridge column the OPM and NPOM fragility curves were comparable.

키워드