View article

Structural failures caused by a strong earthquake may induce a large number of casualties and huge economic losses. To enhance life safety and disaster-resilience of communities, the current approach aims to design a structure that can withstand a design level earthquake event. To achieve the critical design objective, it is necessary to estimate the nonlinear structural responses under strong earthquake ground motions. Although nonlinear time history analysis is the only possible way to precisely estimate the structural responses in many situations, high computational cost and modeling time may hamper the adoption of the approach in routine engineering practice. Thus, in modern seismic design codes, various simplified regression equations are introduced to replace the onerous and time-consuming nonlinear time history analysis, but the accuracy of the estimated responses is limited. Moreover, the existing methods cannot quantify the uncertain errors in the response estimation, mainly caused by the loss of information in representing input data by selected features, especially regarding ground motion characteristics. To effectively predict the seismic responses without performing dynamic analysis, this study introduces a deep neural network as a regression function and a structural system is considered as a single degree of freedom (SDOF) system. First, a deep neural network (DNN) model that can predict seismic responses of structural systems is developed using a neural network architecture is motivated by the earthquake excitation mechanism. In the DNN model, a convolutional neural network (CNN) is introduced to extract the …