A deep learning-based ground motion truncation method to improve efficiency of structural time history analysis

Time history analysis is considered as one of the most commonly used methods to evaluate the seismic performance of structures. But it is time-consuming for the computation cases such as complex structures and ground motions with long duration. In this study, a deep learning-based ground motion truncation method is proposed to predict the truncation point of ground motions, so that the calculation time is reduced by using the truncated records instead of original ones. A novel multi-input to single-output neural network model is established to predict the truncation point of ground motion. The input parameters of the network model consider ground motion-related information, structure-related information, and information related to both structure and ground motion, while the output is the truncation position of the ground motion. Three types of structures, above-ground and underground structures included, are used to verify the performance of the truncation method. The results show that the proposed deep learning-based ground motion truncation method is simpler and more efficient to determine the truncation points compared with the existing method while maintaining high accuracy on the calculation results of structural peak displacement response.