Seismic Damage Prediction For Buildings In Urban Areas Based On GPU-CPU Cooperative Computing

In the21stcentury, severe earthquakes have occurred throughout the world.Besides improving the seismic resistance capacity of buildings, the seismic damageprediction for existing buildings in urban area is also very effective to reduce thecasualties and economic losses caused by earthquakes. In traditional seismic damageprediction methods based on single-degree-of-freedom models and static analyses,higher-order vibration and full properties of earthquake cannot be considered. In recentyears, seismic damage prediction based on refined models and nonlinear time-historyanalyses has been adopted in more and more studies, but it is difficult to use suchmethods widely because of the huge computing workload.To overcome these problems, a new seismic damage prediction method for urbanareas based on multi-scale models and nonlinear time-history analyses is proposed inthis research, in which GPU/CPU cooperative computing is adopted. The solutions forthree important issues of seismic damage simulation for both normal and importantbuildings are proposed: modeling method, high-performance computing, andpost-process visualization. The major achievements of this research are as follows:(1) For a large quantity of normal buildings in urban areas, a multi-story shearmodel suitable for regional seismic damage simulation is proposed. The effect ofhigher-order vibration of buildings and velocity pulses of ground motion can beconsidered by using this model, and the damage location of building can be obtained.(2) A method to determine the parameter of the proposed multi-story shear modelfor normal buildings is proposed. For a normal building in urban areas, only somemacro properties (e.g. number of stories, structural type, construction period, etc.) arerequired to determine the parameters of its computing model, which means that thecomputing model of the whole urban area can be generated rapidly from the data in GISdatabase.(3) A method for nonlinear time-history analyses of normal buildings based onGPU/CPU cooperative computing is proposed, and the corresponding program iscompleted. The performance-to-price ratio can be up to39times of the similar methodbased on CPU computing only. The fully performance of the program can be obtained ifsingle-precision floating-point is used and the block size is set to32. (4) For the few important buildings in urban areas, the open-source finite-elementsoftware OpenSees is applied for the seismic damage prediction. A program which canconvert the SAP2000model to OpenSees model, and a GPU-accelerated solver forOpenSees are developed. Thus, the modeling of OpenSees can be easier and rapidly,and the performance of the analysis by using the GPU-accelerated solver can be9–15times of that by using the original solver.(5) Post-processing visualization programs for both normal buildings andimportant buildings are completed. These programs have some very useful features forenhancing the readability of prediction results, time-history animation, displacementcontour, and interactive functions.