Damage Detection in Historical Masonry Minarets using Machine Learning and Optimization-Based Structural Assessment

This study investigates the effectiveness of supervised machine learning techniques in detecting damage in a historical masonry minaret characterized by variable wall thicknesses and diverse geometric features along its height. The minaret is systematically partitioned into 10 distinct regions, each sharing similar physical properties, allowing for a localized evaluation of structural integrity. By leveraging the inherent mechanical properties of these regions, the study conducts an optimization procedure to align experimentally measured frequencies with their numerical counterparts, thereby ensuring consistency and accuracy in the data. Throughout the optimization, damage ratios and indices are derived, capturing the variations in mechanical properties and mode shapes, which serve as critical indicators of structural degradation. Advanced machine learning algorithms are employed to classify and predict damage states based on these indices. The results demonstrate that the integration of damage indices — extracted from diverse mode shapes — with supervised learning models can reliably detect and quantify damage, even in complex structures. Overall, this research highlights the potential of combining experimental dynamic data with numerical models and machine learning to enhance the damage detection process in historical masonry structures, thereby contributing to more effective strategies for preservation and maintenance.