Regional Seismic Loss Assessment by Deep-Learning-based Prediction of Structural Responses

As urban systems become more complex and sophisticated, the vulnerability of densely populated areas under earthquake hazard also increases. To establish risk-based strategies for hazard mitigation and recovery at the urban community level, many research efforts have been made for probabilistic seismic risk assessment (PSRA). When performing PSRA, structural responses are usually estimated by fragility functions or nonlinear static procedures. It is, however, noted that developing fragilities of each of numerous structures in a large area may require huge computational cost whereas nonlinear static procedures may not incorporate variabilities of the structural responses given a seismic intensity. Recently, the authors developed a deep-learning-based approach for probabilistic evaluation of the structural responses for a wide class of hysteretic behavior and ground motions. To reduce the computational cost of a regional seismic loss estimation and improve its accuracy, this paper proposes a new PSRA using the deep-learning-based method. To demonstrate the applicability of the proposed method and its merits, a hypothetical example of PSRA is investigated. In addition, this paper proposes a procedure to determine the optimal number of sensors, in which the deep-learning-based method is used to evaluate the seismic loss. Furthermore, the trained deep neural network model is employed as a surrogate model for a real-time PSRA. The deep-learning-based PSRA and the procedure to determine the sensors for installation are expected to improve PSRA at community level in terms of efficiency and applicability, and provide new insights into the seismic risk assessment and management of urban systems.The first and the second author are supported by the project Development of Life-cycle Engineering Technique and Construction Method for Global Competitiveness Upgrade of Cable Bridges of the Ministry of Land, Infrastructure and Transport (MOLIT) of the Korean Government (Grant No. 16SCIP-B119960-01), and the third author is supported by the Korean Federation of Science and Technology Societies (KOFST) grant funded by the Korean government (MSIP: Ministry of Science, ICT and Future Planning). Finally, this research was enabled in part by support provided by WestGrid (www.westgrid.ca) and Compute Canada (www.computecanada.ca)