Data-driven low-dimensional representation and reconstruction of turbulent flows
ORAL
Abstract
We present an approach that combines physical consistency with data-driven flexibility to encode and decode turbulent flows, addressing the low-dimensional representation and reconstruction of their complex dynamics.
The turbulent velocity fields are encoded as multi-scale vortex filaments, forming intricate vortex entanglements that yield a sparse yet accurate representation of the flow structure, effectively capturing its essential features in a compact and efficient form.
Together with the decoding process, our method enables bi-directional mapping between the turbulent flow and its low-dimensional representation.
We validate our method with various examples, including knot topologies, three-dimensional isotropic turbulence, and both simulated and experimental jet data, demonstrating its effectiveness in encoding and decoding turbulent flows.
The turbulent velocity fields are encoded as multi-scale vortex filaments, forming intricate vortex entanglements that yield a sparse yet accurate representation of the flow structure, effectively capturing its essential features in a compact and efficient form.
Together with the decoding process, our method enables bi-directional mapping between the turbulent flow and its low-dimensional representation.
We validate our method with various examples, including knot topologies, three-dimensional isotropic turbulence, and both simulated and experimental jet data, demonstrating its effectiveness in encoding and decoding turbulent flows.
–
Presenters
-
Tao NI
- Zhejiang University