Leveraging Machine Learning for Structural Response Characterization of Brick Masonry

Abstract

Advanced techniques like data-driven approaches and machine learning are crucial for understanding and designing resilient masonry buildings to seismic and other hazards. The study highlights the potential of machine learning for low-cost, fast structural assessment of buildings, which will significantly improve the existing vulnerability assessment procedures and increase the reliability of the results at lower investments. This study presents the utilization of machine-learning models to estimate the building’s structural response, which is helpful for the vulnerability characterization of the building. Considering the natural period of vibration of the building as an essential structural response parameter, it is predicted using the standard building features collected in post-disaster surveys. This study further analyzes the importance of building features, such as different geometric configurations and material properties, for a building’s time-period response. Seven different machine-learning models were trained and evaluated for prediction accuracy using model evaluation metrics such as MAE, MSE, RMSE, and R2, of which seven models, which gave an R2 value of more than 0.5, were considered for detailed study. Among various models, the model with the CatBoost algorithm was the best and had the highest model accuracy.