Seismic vulnerability assessment of slender masonry structures: A machine learning approach with vibration-based assessment and nonlinear dynamic analysis

Background: Historic slender structures like minarets are at risk of earthquakes and pose significant preservation challenges. Objective: This study proposes a Machine-Learning (ML) approach to predict the seismic vulnerability of masonry minarets by integrating dynamic properties and numerical modeling. Method: The natural frequencies and curvature mode shapes were assessed using vibration analysis. Three-dimensional Finite Element (FE) models incorporating nonlinear material behavior and geometric characteristics were created and validated against experimental responses. Nonlinear dynamic analysis using various ground motions was used to assess the seismic response, forming the basis for a vulnerability assessment method. Using these results, ML-based techniques such as Multiple Linear Regression (MLR), Random Forest (RF), K-Nearest Neighbor (KNN), and Support Vector Regression (SVR) were developed to predict seismic vulnerability. Results: The RF model was effective in evaluating seismic vulnerability achieving lowest error value of 0.02 and a correlation coefficient of 0.96. The RF model delivers results matching those of existing seismic vulnerability curves found in the literature. Conclusions: The proposed methodology, which combines vibration-based assessment, numerical modeling, and machine learning, offers a reliable approach for assessing the seismic safety of minarets. This may provide valuable perception for making decisions about the retrofit and preservation of cultural heritage structures.