Analysis of Turbulent Flow Data Based on a Spectral Basis Representation
Through the increase of computing power, Large Eddy Simulations (LES) have become an invaluable design tool for industrial applications. However, the increasing amount of data that is generated in each simulation imposes new requirements on data analysis and compression. Two shortcomings of classical approaches are addressed in this study. First, due to storage limitations, only a limited set of quantities can be extracted from the simulation, but needs to be selected in advance. This requires a-prior knowledge about the flow structure and an experienced user. Second, classical post-processing starts after the simulation is completed. Nevertheless, many effects occur at an early stage of the simulation and can give useful insights into trends. Data I/O and computing time can be reduced, if they are monitored carefully. To address both issues, we apply a spectral basis approach for nonlinear dimensionality reduction to turbulent flow data. The spectral basis is derived from the eigenvectors of the Laplace-Beltrami(LB) operator which are optimal for representing smooth function son a surface. The solution is projected to the LB basis to achieve a compact representation of the flow with a spectral separation. We demonstrate our approach on the LES of the turbulent flow through a HVAC duct that was studied intensively in the past, e.g. by Jäger, A. etal., 2008. Results are compared to a PCA analysis based on an extensive mesh study regarding their compactness and ability to identify large-scale and dominant flow features at an earlier state during the simulation.