No Travel? No Problem.

Remote Participation
Lightning Talk: In Situ Anomaly Detection and Reduced Order Surrogate Models for DNS of Turbulent Combustion
Event Type
Online Only
Big Data
Data Analytics
Data Management
Registration Categories
TimeSunday, 14 November 20212pm - 2:30pm CST
DescriptionExascale computing will provide a unique opportunity to approach device-scale first principles direct numerical simulation (DNS) and enable access to physics regimes previously unattainable. With the advantages of access to “bigger, more complex” problems come challenges of data management and requirements for new tools for data discovery. There is an inherent need to carry out data discovery and efficiently manage computational requirements in situ while simulations are performed on DOE leadership class machines. Moreover, with increasing computational resources, scientists will want to increase the scale and complexity of problems they are interested to investigate. Machine learning (ML) has emerged as an integral part of advancing state-of-the-art DNS and extending this valuable simulation approach to the exascale and beyond. On the one hand, strategies are being designed to enable the implementation of predictive in situ reduced order models (ROMs) that are computationally more efficient than conventional DNS through in situ model reduction and data compression. On the other hand, strategies are designed to detect anomalous physics behavior used to computationally steer downstream analysis. These strategies will be described in the context of combustion simulations at extreme scale.
Back To Top Button