Akademik Veri Yönetim Sistemi
International Journal of Architectural Heritage, 2026 (SCI-Expanded, AHCI, Scopus)
Historic masonry towers, due to their brittle materials, high slenderness, and unique geometry, are highly vulnerable to lateral excitations. Their preservation requires reliable methods to identify vulnerable structures, particularly through accurate estimation of predominant frequencies for seismic vulnerability assessment, as the dynamic response of slender towers is governed mainly by the fundamental mode and influenced by geometry and boundary conditions. This study proposes a data-driven framework that leverages Deep Neural Networks (DNNs) to predict the fundamental frequencies of masonry towers using both geometric and material parameters. To further enhance predictive capability, a novel Planet Optimization Algorithm-DNN (POA-DNN) model is developed by integrating a metaheuristic optimization strategy, with deep learning to address the nonlinear and multi-input nature of the problem. A comprehensive dataset of 52 real historic towers is assembled, including total height, effective height, minimum width at the base, slenderness ratio, façade opening ratio, and Young’s modulus. The performance of POA-DNN models trained with two input combinations (geometry only; geometry + materials) is systematically evaluated. The proposed POA-DNN framework, incorporating additive geometrical and material parameters, significantly improves the frequency prediction reliability, limiting the maximum error to about 30%, which is markedly lower than that of classical DNNs and empirical models.