Fundamental frequency prediction of historic masonry towers based on artificial neural networks

Abstract

Historic masonry towers, with their brittle materials, slenderness, and distinctive shapes, are highly susceptible to lateral excitations. The urgency of preserving surviving ones in earthquake-prone regions has become apparent. There is a prioritization of identifying and reinforcing the most vulnerable masonry towers. Predictions are based on the earthquake spectrum specific to each region, effectively alerting to the seismic vulnerabilities of towers constructed within those areas. Rather than formulating relationships based on known geometrical parameters, this study relies on an Artificial Neural Network (ANN)-based model to promptly estimate the fundamental frequency of masonry towers. Measurements taken from 19 actual masonry towers are utilized for training the networks. Various distinct tower parameters as well as their combinations are considered. Only geometrical information is taken into account while material properties are not considered. The performances of ANNs are directly compared to some empirical equations. The ANN-based techniques are evaluated by testing with 20 different towers that are not considered in the training process. The proposed ANN tool demonstrates practicality and robustness when estimating the lowest frequency of masonry towers based only on geometrical information. The slenderness ratio that is rarely considered in existing equations remarkably enhances the accuracy of fundamental frequency anticipation.