Runtime configurable systems for computational fluid dynamics simulations

Abstract

The development of a new framework for the construction of modular simulation software is presented as one of the possible ways to increase usability and speed of development of numerical simulation tools. Over the last few decades many authors have indicated that modularity and adaptability have become important requirements for modern numerical simulation tools. The type of architecture proposed by several authors is the plug-andplay simulation tool which allows users to link simulation components together like Lego bricks. Through a literature review it was discovered which requirements should be placed on a plug-and-play simulation tool. Furthermore several frameworks and applications which have plug-in capabilities were located but no evidence was found that any of these simulation tools have full run-time configuration capabilities. This research has addressed this deficiency through the development of a framework for the development of numerical simulation tools. One of the main differences between the new framework and existing systems is that the new system is capable of performing all operations at run-time. Modules are detected, loaded and used at run-time. The framework needs no prior knowledge of the type and availability of components, thus providing true plug-and-play capabilities. Initially a prototype framework was implemented to verify that a plug-and-play simulation tool could be developed. A simple set of components was implemented to verify the usefulness of the framework. The components, when coupled, formed the basis for a numerical simulation [234], With the lessons which were learned during the implementation and verification of the prototype a second version was developed which removed some unnecessary complexity from the prototype and added additional capabilities which improved modularisation and flexibility. This made the framework more powerful due to the increase in flexibility and modularity. The cost for this improvement is an increase in the complexity of the internals of the framework which required several innovative solutions in order to allow the framework to deal with component detection and loading, variable handling and simulation scheduling. To verify that the capabilities of the second version of the framework satisfied the requirements, two sets of components were implemented. The first component set wrapped the FreeALE flow solver thereby allowing the framework to be used in fluid flow simulations. The second set contained a surrogate model, global optimiser and the necessary utilities to initialise both the surrogate model and the flow solver. Implementation verifications for the flow solver and optimiser were performed and returned satisfactory results. Finally both component sets were combined at run-time to perform an optimisation based on the results from fluid flow simulations. The results obtained during these calculations showed that the framework had the capability to develop new multi-physics simulations using existing CFD tools in a run-time environment thereby achieving the objective of a plug-and-play development system.