A Time-Parallel Monte Carlo Approach for Unsteady Systems
Presenter
DescriptionSeveral challenges arise when considering the design and simulation of large-scale civil structures under the action of wind. First, the dimensions and complexity of these structures require the use of large meshes that involve the resolution of enormous systems of equations. Second, the randomness associated to the wind conditions adds a considerable complexity in the analysis of the flow, which requires the use of uncertainty quantification methods. Third, the unsteady and chaotic behaviour of the wind requires the analysis of long periods of time of the flow conditions to ensure convergence of the statistics. State of the art parallel computing techniques can be exploited to tackle the first two challenges. Large systems can be solved by splitting the spatial domain in partitions that are run in parallel in independent processors. Similarly, the uncertainty of the wind can be tackled by non-intrusive methods such as Monte Carlo algorithms, which allow to generate independent realizations of the wind that can be run in parallel. Nonetheless, the parallelization of long-time analyses represents an open research topic, the key issue being that the time dimension cannot be parallelized. Current work presents some results about the use of ensemble average approaches in minimizing the time to solution of the simulation.
TimeThursday, 8 July 202111:00 - 11:30 CEST
LocationErnesto Bertarelli
Session Chair
Event Type
Minisymposium
CS and Math
Climate and Weather
Engineering