Surrogate modeling in structural vibration problems with dynamic mode decomposition

Frank Blubaugh

DOI: https://doi.org/10.1121/10.0015794

2022-10-01

The Journal of the Acoustical Society of America

Abstract:Solving large-scale vibration problems presents a difficult challenge for complicated geometries that is only addressable through the use of large-scale compute resources. These problems not only require extensive compute time and cost, but are also expensive in engineering hours for modeling and analysis. While optimization techniques can help limit engineering time in the loop, the computational requirements of these models have made applying traditional optimization techniques to this class of problem untenable. Dynamic Mode Decomposition (DMD) is an approach that can help bridge this gap by building high-fidelity surrogate models allowing for the inline development of a surrogate model to dramatically reduce the computational time and complexity of a problem. This approach enables fast frequency sweeping while capturing the dominant dynamics of the model. DMD has historically been applied to other high-data volume models such as computational fluid dynamics, medical imaging and controls, as well as less structured problems including sociology and market behavior. This paper will cover the fundamentals of Dynamic Mode Decomposition, the extension of the theory to apply the technique to vibration problems, the demonstration of a potential workflow, and show how this technique performs on simple test cases illustrating performance speed ups for classic problems from existing literature.

acoustics,audiology & speech-language pathology