Neural network model for seismic response of braced buildings

Abstract

Non-ductile, concentrically braced frames are a common lateral force-resisting system used in low-to-moderate seismic regions in the USA. However, their dynamic responses to earthquake ground motions have not been well examined. Structural engineers usually design them conservatively as brittle structures with a small response-modification factor, while building codes restrict their use to low-rise buildings. In this paper, seismic responses of two typical non-ductile concentrically braced frames, one of three storeys and one of nine storeys, were predicted through a neural network model. Twelve input parameters, covering non-linear features from structural components and the uncertain nature of earthquake ground motions, were used in the modelling. Numerical results extracted from thousands of non-linear time-history analyses under one set of moderate ground motions were used to develop the model. Sensitivity analyses were conducted to evaluate the impacts of input parameters on the peak inter-storey drift ratio, designed as an output parameter in the model. The results are shown to be promising considering the uncertainties in both ground motions and the characteristics of structures.