Towards Petascale Simulation of Urban Earthquake Impacts

该奖项是 NSF 07-559 计划征集“通过千万亿次模拟和分析加速科学与工程发现”竞赛的结果。 牵头机构是卡内基梅隆大学，并授予加州大学伯克利分校、戴维斯分校和圣地亚哥分校的子奖项。该奖项将为整个高地震活动城市地区、大洛杉矶盆地 (GLAB) 开发基于物理的高保真千万亿次模拟的方法、能力和软件，以评估大地震对建筑物、交通系统组件和地下土木基础设施的工程影响。 地震是美国面临的最严重的自然灾害之一。 模拟它们对建筑环境的影响是千万亿级高性能计算 (HPC) 系统的独特科学驱动力，因为具有不同物理特性的多个长度尺度、大数据量以及对高度可扩展的并行可视化和数据查询的需求。目前的三维地震模拟仅限于线性土壤行为和孤立的个体结构。该项目将超越当前的方法，开发模拟能力，包括土壤、地基和大量结构之间的相互作用；密集建筑区域中结构之间的相互作用；以及结构对自由场运动的影响；以及非线性土壤行为和饱和土壤中孔隙水压力导致液化的影响。 当前的 HPC 能力不足以满足这些复杂的模拟。该项目将使用新兴的千万亿级系统来执行和集成（1）大型沉积盆地的地面运动模拟，（2）土壤非线性行为的模拟，（3）建筑物、桥梁和其他基础设施系统的大型库存模拟，（4）计算数据库，以及（5）可扩展的可视化技术。 该项目将在千万亿级模拟的分层、多尺度方法方面取得新进展。从大型区域、宽带地震模拟到多个次区域，将允许以高度可扩展的方式进行详细建模，以准确捕获整个清单的场地响应和结构响应。域缩减方法将首次使用区域模拟中的地面运动作为多个人口稠密分区的输入。 这些子区域包括高度非线性土壤模型以及数百座建筑物和桥梁以及嵌入式基础的详细模型。将开发一种新的可扩展的隐式-显式时间积分方法，为子区域的复杂地震模拟提供最佳的计算性能。 将开发数据分析和可视化功能，以便在与模拟相同的并行处理器上运行，从而大大减少移动数据的需要。使用这种方法进行数据密集型超级计算、现场可视化和新的计算数据库系统将提供前所未有的理解地震影响的能力。新功能将提供信息，通过深入了解感兴趣的细节，同时保持全球背景，促进理解和决策。选择 GLAB 是因为地震风险高、公共政策需求重要以及它为验证和确认提供的丰富信息。 GLAB 测试台的年度里程碑经过校准，可以尽早利用 HPC 资源，并随着千万亿级系统上线而进行模拟扩展。 模拟的方法、应用和结果将有助于灾害规划和管理，因为改善备灾和减灾需要详细了解城市系统在大地震中的表现。 从事该项目的研究生将有机会通过土木工程、计算工程和计算机科学的多学科研究创造新知识。

This award is an outcome of the NSF 07-559 program solicitation, "accelerating Discovery in Science and Engineering through Petascale Simulations and Analysis" competition. The lead institution is Carnegie Mellon University, with subawards to the Universities of California at Berkeley, Davis, and San Diego. This award will develop methodologies, capability, and software for high-fidelity, physics-based petascale simulations of an entire high seismicity urban region, the Greater Los Angeles Basin (GLAB), to assess the engineering impacts of large magnitude earthquakes on buildings, transportation system components, and the underground civil infrastructure. Earthquakes are one of the most severe natural hazards facing the United States. Simulating their effects on the built environment is a unique science driver for petascale high-performance computing (HPC) systems because of the multiple length scales with different physics, large data volumes, and need for highly scalable parallel visualization and data querying. Current three-dimensional earthquake simulations have been limited to linear soil behavior and isolated, individual structures. Advancing beyond current approaches, this project will develop simulation capability that includes the interaction between the soil, foundations, and large inventories of structures; the interaction between structures in densely built areas; and the effect of the structures on the free-field motion; as well as the nonlinear soil behavior and the effect of the pore water pressure in saturated soils leading to liquefaction. Current HPC capability is inadequate for these complex simulations. This project will use emerging petascale systems to perform and integrate (1) ground motion simulation of large sedimentary basins, (2) simulation of the nonlinear behavior of soil, (3) simulation of large inventories of buildings, bridges, and other infrastructure systems, (4) computational databases, and (5) scalable visualization techniques. The project will make new advances in a hierarchical, multi-scale methodology for petascale simulation. Going from a large regional, broadband earthquake simulation to multiple subregions will allow detailed modeling in a highly scalable manner to capture site response and structural response accurately for entire inventories. The Domain Reduction Method will be applied for the first time using ground motion from a regional simulation as input for multiple highly populated subregions. These subregions include models of highly nonlinear soil and detailed models of hundreds of buildings and bridges along with embedded foundations. A new scalable implicit-explicit time integration method will be developed to provide optimal computational performance for the complex earthquake simulations for a subregion. Data analysis and visualization capability will be developed to run on the same parallel processors as the simulation, drastically reducing the need to move data. Using this approach for data-intensive supercomputing, in-situ visualization and a new computational database system will allow unprecedented ability to understand earthquake impacts. The new capability will present information that facilitates understanding and decision-making by drilling down to the detail of interest while maintaining the global context. The GLAB was selected because of the high seismic risk, important public policy need, and the wealth of information that it provides for verification and validation. The annual milestones for the GLAB test bed are calibrated to take early advantage of HPC resources, with scaling of the simulations to take place as petascale systems are brought online. The methodology, applications, and the results of the simulations will be useful for disaster planning and management, since it is the detailed knowledge of how an urban system performs in a large earthquake that is needed for improving disaster preparedness and mitigation. Graduate students working on the project will have the opportunity to create new knowledge through multidisciplinary research in civil engineering, computational engineering, and computer science.