Computational Fluid Dynamics simulation
Background and challenges
The rise and rapid progress of AI and ROM (Reduced Order Model) applied to computational physics offer the potential of a game changer in the coming years. As a CFD (Computational Fluid Dynamics) simulation provider, ICON wishes to leverage these technologies, which come with their own challenges. Through its online platform ICON aims to provide an environment addressing multi-fidelity simulation capabilities, which in turn require a heterogeneous set of computing facilities. In such a framework, the high-fidelity simulations are provided by traditional distributed memory (MPI) HPC codes requiring high memory bandwidth and high-speed interconnect such as iconCFD, OpenFOAM, and so on. The data collected from these simulations then need to be stored, standardized, and made available for further processing. Finally. low-fidelity simulations shall be achieved by using the ROM and can be provided to the end-user through a micro-service approach in order to be embedded in further design exploration and optimization systems.
Within this project ICON envisages two steps:
1. Evaluation and testing of existing machine learning codes, understanding and documentation of their use within the MORPHEMIC platform in view of being used by ICON’s ROM manager.
2. Demonstration of the use of Morphemic platform to cater for the computing requirements of the 3 aspects of iconCFD Platform on semi-industrial cases relevant to the automotive industry.
1. Currently, it is a cumbersome and completely manual task to select the right resources for running the simulations. On the other hand, MORPHEMIC enables the automatic selection and interconnection of the most suitable resources according to just the application model and requirements.
2. The running of such simulations depends on the availability of respective HPC resources where the access latency can be quite high. MORPHEMIC can enable to select equally-capable cloud resources to achieve the same performance level with minimal access latency.
3. Both data collection, migration and processing are done manually which causes further delay in the overall simulation workflow. MORPHEMIC will enable to define just the whole application/simulation workflow as well as application model in CAMEL and then take care of both the deployment and adaptive execution of this workflow in the cloud.
4. The cost of using HPC resources is quite high to run the simulations. On the other hand. through the adoption of equally-capable but less expensive cost resources, the simulation can be run in a cost-effective manner while delivering a similar performance level as in the case of pure use of HPC resources. Thanks to the possible use of the right hardware accelerators, the simulation’s performance level can become even higher.