採用數據驅動人工智慧與機率式耐震性能評估法之區域韌性評估架構開發

The development of sustainable and resilient cities is an important issue aroundthe world due to extreme natural hazards. Because of the limitation of naturalresources, it is much important to enhance the urban resilience for Taiwan. Mosturban regions are crowded with buildings and people, which plays a key role inthe evaluation of seismic resilience. Current methodology widely used forassessing seismic performance only describes the responses (i.e., inter-storydrift) of a building subjected to different levels of earthquakes, so it cannot beused to bridge the gap between earthquake intensity and decision variables suchas economic loss, and repair time. Hence, this project aims to develop aframework using probabilistic seismic performance assessment methodology,namely, FEMA P-58, to quantify regional seismic resilience. To accomplish aregional assessment, data-driven artificial intelligence including deep learningand machine learning are adopted in the proposed framework to harvest big datafor numerical modeling of thousands of buildings. Big data collected in this studyincludes building geometric information harvested from street-view and aerialimages using deep neural networks, and parameters predicted by machinelearning models such as random forest for nonlinear finite-element modeling.Reinforced concrete moment frames with masonry-infilled walls are selected astarget buildings in a region to be evaluated using localized FEMA P-58methodology. The proposed framework will be developed using highperformance computers to train deep-learning models and to conduct millions ofnonlinear time-history analyses for regional assessment.With the proposed data-driven models, several databases will be also publishedin this project including building geometric information, experimental data of RCframes with masonry-infilled walls, and dynamic responses of buildings. Thesedatabases can be further utilized to develop machine-learning and deep-learningsurrogate models to address traditionally laborious works in the field of civilengineering including structural health monitoring and regional damageassessment. In addition, resilience data obtained using the proposed frameworkcould be used to assist in natural hazard mitigation and sustainable development.