Damage Assessment in Historical Masonry Minarets with Supervised Machine Learning and Deep Learning Methods

OpenAlex İngilizce

In this study, damage scenarios in a historical masonry minaret were evaluated using ML and DL algorithms. The method consists of three main steps. In the first step, the natural frequencies of the minaret were obtained by the OMA method. In the second step, the FE model of the structure was created, and the frequencies of the mode shapes were optimized to make the dynamic behavior of the model consistent with the real structure. The optimization process was carried out with the Response Surface method using ANSYS Workbench software, ensuring that the dynamic properties of the FE model were accurately represented. In the last step, the scenarios obtained during the optimization of the numerical model were divided into groups and used as data sets for supervised machine learning. The datasets consist of damage indices based on frequencies representing damage rates and mode shapes based on changes in the mechanical properties of the structure. A total of six ML and DL algorithms were used for damage detection, and data sets of two different modes were used in each analysis. The results show that with accuracy rates of up to 96%, the algorithms effectively detect damage using only damage indices derived from the two mode shapes.