Rapid post-earthquake safety assessment of low-rise reinforced concrete structures

Abstract

Many countries throughout the world have experienced large earthquakes, which cause building damage or collapse. After such earthquakes, structures must be inspected rapidly to judge whether they are safe to reoccupy. To facilitate the inspection process, the authors previously developed a rapid building safety assessment system using sparse acceleration measurements for steel framed buildings. The proposed system modeled nonlinearity in the measurement data using a calibrated simplified lumped-mass model and convolutional neural networks (CNNs), based on which the building-level damage index was estimated rapidly after earthquakes. The proposed system was validated for a nonlinear 3D numerical model of a five-story steel building, and later for a large-scale specimen of an 18-story building in Japan tested on the E-Defense shaking table. However, the applicability of the safety assessment system for reinforced concrete (RC) structures with complex hysteretic material nonlinearity has yet to be explored; the previous approach based on a simplified lumped-mass model with a Bouc-Wen hysteretic model does not accurately represent the inherent nonlinear behavior and resulting damage states of RC structures. This study extends the rapid building safety assessment system to low-rise RC moment resisting frame structures representing typical residential apartments in Japan. First, a safety classification for RC structures based on a damage index consistent with the current state of practice is defined. Then, a 3D nonlinear numerical model of a two-story moment frame structure is created. A simplified lumped-mass nonlinear model is developed and calibrated using the 3D model, incorporating the Takeda degradation model for the RC material nonlinearity. This model is used to simulate the seismic response and associated damage sensitive features (DSF) for random ground motion. The resulting database of responses is used to train a convolutional neural network (CNN) that performs rapid safety assessment. The developed system is validated using the 3D nonlinear analysis model subjected to historical earthquakes. The results indicate the applicability of the proposed system for RC structures following seismic events.