Machine learning-based seismic fragility and seismic vulnerability assessment of reinforced concrete structures

Abstract

Many studies have been performed to put quantifying uncertainties into the seismic risk assessment of reinforced concrete (RC) buildings. This paper provides a risk-assessment support tool for purpose of retrofitting and potential design strategies of RC buildings. Machine Learning (ML) algorithms were developed in Python software by innovative methods of hyperparameter optimization, such as halving search, grid search, random search, fine-tuning method, and the k-fold cross-validation, to derive the seismic fragility curve for accelerating seismic risk assessment. Proposed ML methods significantly reduced the computational efforts compared to conventional procedure of seismic fragility assessment. The prediction results can be combined with considered hazard curves for the purpose of seismic risk assessment of RC buildings. To prepare the training dataset, Incremental Dynamic Analyses (IDAs) were performed on 165 RC frames to achieve 1121184 data points. Performance indicators showed that the algorithms of Artificial Neural Networks (ANNs), Extra-Trees Regressor (ETR), Extremely Randomized Tree Regressor (ERTR), Bagging Regressor (BR), Extreme Gradient Boosting (XGBoost), and Histogram-based Gradient Boosting Regression (HGBR) had higher performance, which achieved acceptable accuracy and fitted to actual curves. In addition, Graphical User Interface (GUI) was introduced as a practical tool yet reliable for seismic risk assessment of RC buildings.