| Popis: |
System simulations are a useful tool when analyzing complex systems. For example, simulations can be cheaper than physical testing of the actual system. Models canalso be created for systems that do not exist yet, that would be impossible to analyze experimentally. The models used in these system simulations require appropriate input data in order to give results with the required accuracy. Since each systemcan be made up of several different engineering domains, it can be difficult to inte-grate results from many different sources. For this thesis, the aim was to develop a methodology that allows CFD results to be integrated into a simulator using system identification and open-source standards. The system to be analyzed was a pipe connected to a heat exchanger and con-sumer of cooling power. The pipe model was to be able to calculate the pressure drop inside the pipe, based on the mass flow and specific enthalpy of the fluid. The fluid domain for the CFD simulations was divided into 5 smaller pipe sections in order to reduce the computational cost. The sections were meshed and verified, and a total of 100 steady-state flow cases were selected to give input and validation data for the system identification. The cases were run using the FEM-based Altair AcuSolve CFD solver. The extracted data from these cases was then imported to Altair romAI in order to create reduced-order models using neural networks. A model based on lookup-tables was also created. These models were then simulated as part of the cooling system and compared to the current model. The main result of the thesis is a proposed methodology for creating standardized ROMs based on CFD data. In addition, several important factors to consider before using the proposed methodology are presented. These include the intended use of the ROM, knowing the flow inside the system, what resources are available,and any potential licensing issues. The results also showed that steady-state data can be successfully used for system identification of steady-state models. The ROMs created using the method for this thesis gave results that were generally quite close to the currently used handbook equations. At lower Reynolds numbers, the CFD results were more unreliable, giving possible errors in the pressure drop calculatedby the ROMs at those flow cases. |
