On-the-fly Algorithm of Total Dynamic Mode Decomposition using Incremental Singular Value Decomposition
Please login to view abstract download link
Dynamic Mode Decomposition (DMD) [1, 2] has been known as a useful method for extracting the dominant dynamic flow structure from a complex unsteady flow field. One challenge of DMD for engineering problems is that the conventional algorithms require massive computational resources with respect to memory consumption and data storage. On-the-fly algorithms of DMD [3, 4] have been proposed to address those restrictions. We further enhance the on-the-fly DMD algorithm by using Incremental Singular Value Decomposition (ISVD). An advantage of our on-the-fly algorithm is that it doesn’t require saving the entire flow field in the disk during CFD simulation for both computing DMD modes and for performing Sparsity-Promoting DMD (SPDMD) [5] to identify dominant modes. We verified that our method computes the relevant modes to the conventional Total DMD [6] combined with SPDMD.
