Akademik Veri Yönetim Sistemi
Journal of Building Engineering, cilt.112, 2025 (SCI-Expanded, Scopus)
Seismic assessment of buildings is crucial for mitigating seismic risk in earthquake-prone urban areas. Traditional methods range from simple, subjective visual street surveys to complex, time-consuming nonlinear analyses. This paper introduces a novel multi-level artificial neural network (ANN) methodology for estimating the fundamental structural period (T) and maximum interstory drift ratio (MIDR) of reinforced concrete (RC) frame buildings, key indicators of seismic performance. Using a comprehensive dataset of 4417 RC building designs constructed between 1960 and 2020 from Türkiye, a synthetic dataset is generated to enhance the ANN training. This combined dataset allowed ANN models to predict T and MIDR with high accuracy under various input scenarios. The best-performing ANN model achieved mean absolute errors of 2.96 % and 15.68 % in T and MIDR prediction, respectively, demonstrating exceptional predictive accuracy. By employing explainable artificial intelligence techniques, the influence of each input parameter on the predictions is revealed, ensuring transparency and interpretability. The method is validated by comparing the seismic demand estimations based on predicted MIDR values against observed damage from the recent 2023 earthquakes in Türkiye. Additionally, the ANN model is applied to a simulation dataset to generate a national seismic demand map of Türkiye based on predicted MIDR values to demonstrate practical applicability. The proposed ANN model offers a scalable and efficient tool for policymakers and engineers to assess and mitigate seismic risks, laying the groundwork for future advancements in large-scale seismic vulnerability assessments.