Assessment of uncertainties in damping reduction factors using ANN for acceleration velocity and displacement spectra
Künye
Abdelhamid, A., Benahmed, B., & Palanci, M. (2023). Assessment of uncertainties in damping reduction factors using ANN for acceleration, velocity and displacement spectra. Electronic Journal of Structural Engineering, 23(4), 8-13.Özet
Elastic analysis is performed during the design process, and earthquake forces are computed according to standard damped spectral accelerations, which are assumed to be 5% at most. However, buildings are expected to behave nonlinearly instead of linearly due to moderate to destructive earthquakes. Accordingly, the damping factor between the design and actual behaviour of buildings during earthquake excitation differs. This situation increases the uncertainty of the design process for structures exposed to seismic loads and the variation in the reliable estimation of the structures' seismic response. This study is focused on the investigation of the structural damping uncertainties effect on the structure's response spectra through the assessment of uncertainties in the damping reduction factors (DRF) derived from the acceleration, velocity, and displacement spectra. For this purpose, the Monte Carlo method, which relies on repeated random sampling to obtain numerical results, is used for the estimation of the stochastic DRF. The obtained results indicate that the difference between the deterministic and stochastic DRF is around 21% for displacement and velocity and 28.7% for acceleration spectra. Consequently, the DRF derived from the acceleration spectra is more sensitive to the uncertainties inherent in damping than the DRF obtained from displacement and velocity. Therefore, it is important to take this conclusion into account when using these factors. To estimate the calculated DRF values, an artificial neural network (ANN) was developed for the stochastic DRF calculation. The ANN constitutes a simple and efficient method to predict the stochastic DRF since the error obtained is always less than 6%. According to the developed model, practice-oriented results are evaluated for the future evolution of seismic codes.