A swift neural network-based algorithm for demand estimation in concrete moment-resisting buildings

Abstract

Rapid evaluation of demand parameters of different types of buildings is crucial for social restoration after damaging earthquakes. Previous studies proposed numerous methodologies to measure the performance of buildings for assessing the potential risk under the seismic hazard. However, time-consuming Nonlinear Response History Analysis (NRHA) barricaded implementing a prompt loss estimation for emergency confronting actions. The present study proposes a swift framework for demand estimation in concrete moment-resisting buildings using artificial neural networks. For this purpose, a simplified model is developed based on the HAZUS method. To eliminate the required time-consuming NRHA from the post-earthquake actions, Artificial Neural Networks (ANNs) are used. Before the event, ANNs are studied to estimate the demand parameters using a set of time-history analyses. This study applies to a suite of 111 earthquake events, originally developed in the SAC project and uniformly scaled from 0.1 g to 1.5 g , to achieve a generalized prediction model. Bayesian Optimization (BO) algorithm is carried out to tune the architecture of the NNs. Results reveal that the presented approach is reliable for predicting the structural response, and is cost-effective compared to the conventional NRHA. This framework can be implemented in the body of a risk assessment platform to expedite the postearthquake actions required for crisis management.