Incremental Proper Orthogonal Decomposition Based Method for the Analyzation of Large Scale High Fidelity Simulation Results

Qingsong Wang,Zhen Zhang,Xinrong Su,Xin Yuan
DOI: https://doi.org/10.33737/gpps19-bj-026
2019-01-01
Abstract:Proper Orthogonal Decomposition (POD) analysis is widely used in processing large amounts of high-dimensional data in order to obtain low-dimension approximation, in which way we can capture the most typical modals of the phenomena.Based on this, POD is a powerful tool to analyze complex flow fields.However, there are generally only several modes that dominate unsteady factors, so it is wasteful to calculate all of the modes in analyzing large-scale unsteady calculation results.What's more, the number of grid in simulation of gas turbine is generally huge, so a complete POD which needs full storage costs a lot in memory.Based on incremental singular value decomposition (SVD), a fast incremental POD (iPOD) approach is developed.To compute a full SVD by adding vectors, we iterate the update procedure with truncation based on the user-defined rank, in which way, the left and right singular vector matrices as well as the singular value matrix are updated incrementally, resulting in an online algorithm whose memory requirements reduced several levels.In this work, both 40 and 60 modes are adopted in analyzing the large eddy simulation results of shaped hole film cooling to capture the dominated flow structures and further understand the effect on cooling effectiveness.In comparison, the results of iPOD achieve a good agreement with that of exact method at a low computational cost, demonstrating that the iPOD method is appropriate in large scale high fidelity flow fields analysis.
What problem does this paper attempt to address?