Frequency-Based Damage Detection with ANN and Seismic Behaviour Assessment of a Historical Masonry Tower

Nowadays, Artificial Neural Networks (ANNs) are widely used in damage detection because of their powerful computational and exceptional ability to recognize patterns in historical structures. This paper presents the seismic behaviour assessment and structural damage detection performance of the ANN trained with natural frequencies to identify the damage on the Hagia Sophia Bell Tower in Trabzon, Türkiye. Reduction in the stiffness of the structures causes changes in the structural dynamics. Natural frequencies, being key indicators in damage detection, were utilized as input parameters to train the ANN in this study. The study’s scope included updating of the initial finite element (FE) model constituted with SAP2000 software according to the experimental dynamic characteristics obtained from the Civil Engineering Vibration System developed by our work team. The seismic behaviour of the tower was also examined through time history analyses under the effect of strong earthquakes. The required data for the ANN analyses was obtained from the FE modal analysis of the tower. The ANNs dataset was generated separately by simulating both single and double damage scenarios. Possible twelve damage zones were determined according to the places where the highest stresses occurred in the earthquake analysis of the tower. The reduction in the elasticity modulus of each region defined on the tower was considered as damaging effect and used as ANN output parameter. It was shown that ANN trained with natural frequencies can effectively identify the location and severity of damage.