Citation

Zhang Y, Burton HV. Struct. Saf. 2019; 78: 12-22.

Copyright

(Copyright © 2019, Elsevier Publishing)

DOI

10.1016/j.strusafe.2018.12.004

PMID

unavailable

Abstract

A pattern recognition approach is proposed to quantitatively assess the residual structural capacity of earthquake-damaged tall buildings. Sequential nonlinear response history analyses using as-recorded mainshock-aftershock ground motions are conducted to generate distinct feature patterns comprised of spatially distributed global and local engineering demand parameters (EDP) within the tall building. Residual structural capacity is assessed based on the median spectral intensity corresponding to the collapse prevention performance level. Dispersion-based filtering and feature selection using Least Absolute Shrinkage and Selector Operator (LASSO) are performed to effectively reduce the high dimensional feature space while selecting the most informative ones. The features that survive the filtering but excluded by LASSO are reserved and grouped based on their correlations with those that are selected. These reserved features can be utilized if the selected ones are unavailable. Predictive models using Support Vector Machine are constructed to map the EDP-based features to the residual structural capacity of the tall building, where satisfactory performance is observed as measured by the root mean square errors in the testing dataset. In addition to guiding post-earthquake inspections and residual structural capacity assessments, the proposed framework can inform optimal sensor placement as well as provide time-dependent limit state evaluation in aftershock environments.

Language: en

Keywords

Feature selection; LASSO regression; Pattern recognition; Residual structural capacity; Support vector machine; Tall buildings