As CFD users increasingly exploit ultrascale HPC capability and approach the realm of exascale, they will run studies consisting of many related simulations at very high fidelity. This will give engineers and researchers both the ability and need to assess the quality of their simulations. When simulation quality has been vetted, engineering judgments and investment decisions can be made with confidence.
Uncertainty Quantification (UQ) provides a means to support this need and Intelligent Light has been working on integrating UQ tools including Sandia National Lab's Dakota in an extract-based (XDB) workflow where processing takes place while solutions are in memory and data storage/transfer needs are dramatically reduced.
Dr. Earl P. N. Duque will present "Uncertainty Quantification with Dakota-OVERFLOW within an in situ based FieldView XDB Workflow" at the Overset Grids Meeting. Meet with Dr. Duque and Joe Oliver, Global Sales Manager, to learn how FieldView users are getting more information from their data. Put the latest in high-productivity post-processing, in situ processing and CFD data management techniques to work for you.
Intelligent Light is proud to sponsor this year's symposium which will be held at the Future of Flight Aviation Center in Mukilteo, Washington on October 17-20, 2016.
Attending? Request a meeting.
Uncertainty Quantification with Dakota-OVERFLOW Within an In Situ Based FieldView XDB Workflow
Earl P.N. Duque, Intelligent Light
Dan Heipler, Intelligent Light
NASA has called out the integration of VV/UQ tools into engineering workflows as a vital component of its "CFD Vision 2030 Study". In that document, the authors assert that as exascale computing becomes available, the capability to perform many highly resolved CFD simulations will become ubiquitous, enabling the ability, and need, to perform UQ and sensitivity analysis. To address this need, this presentation summarizes current work whereby the Dakota code drives an ensemble of OVERFLOW2 jobs to explore a nested epistemic-aleatory parameter space. Concurrently, OVERFLOW2 instrumented with VisIt/Libsim creates FieldView XDB surfaces via in situ processing to reduce the data for post-hoc visualization and analysis. The presentation presents UQ concepts in general and their implementation within the current framework.
This work is sponsored by a DOE Office of Science grant (DE-SC0015162)