Neural network analysis of overturning response under near-fault type excitation
Under strong seismic excitation, a rigid block will uplift from its support and undergo rocking oscillations which may lead to (complete) overturning. Numerical and analytical solutions to this highly nonlinear vibration problem are first highlighted in the paper and then utilized to demonstrate how sensitive the overturning behavior is not only to the intensity and frequency content of the base motion, but also to the presence of strong pulses, to their detailed sequence, and even to their asymmetry. Five idealised pulses capable of representing "rupture-directivity" and "fling" affected ground motions near the fault, are utilized to this end: the one-cycle sinus, the one-cycle cosinus, the Ricker wavelet, the truncated (T)-Ricker wavelet, and the rectangular pulse "Overturning-Acceleration Amplification" and "Rotation" spectra are introduced and presented. Artificial neural network modeling is then developed as an alternative numerical solution. The neural network analysis leads to closed-form expressions for predicting the overturning failure or survival of a rigid block,as a function of its geometric properties and the characteristics of the excitation time history. The capability of the developed neural network modeling is validated through comparisons with the numerical solution. The derived analytical expressions could also serve as a tool for assessing the destructiveness of near-fault ground motions, for structures sensitive to rocking with foundation uplift.
作 者： Nikos Gerolymos Marios Apostolou George Gazetas 作者单位： School of Civil Engineering, National Technical University, Athens, Greece 刊 名： 地震工程与工程振动（英文版） SCI 英文刊名： EARTHQUAKE ENGINEERING AND ENGINEERING VIBRATION 年，卷(期)： 2005 4(2) 分类号： P315 关键词： overturning uplifting pulse near-fault motion neural network rocking spectrum Kocaeli records