Unraveling flow patterns through nonlinear manifold learning.

Area of Science:

• Fluid dynamics
• Data science
• Computational physics

Background:

• Characterizing complex flows requires computationally intensive measurements.
• Handling large flow datasets in real-time hinders progress in flow control and sensing.
• Current methods for flow pattern analysis are often limited by data processing demands.

Purpose of the Study:

• To develop a novel framework for unsupervised characterization of flow patterns.
• To apply nonlinear manifold learning to experimental flow data.
• To demonstrate the relationship between manifold topology and flow regimes.

Main Methods:

• Utilized isometric feature mapping (Isomap), a nonlinear manifold learning technique.
• Applied Isomap to experimental video data of a circular cylinder wake.
• Analyzed flow from steady to turbulent regimes without direct velocity measurements.

Main Results:

• Demonstrated that manifold topology is intrinsically linked to the flow regime.
• Showcased that Isomap global coordinates can effectively reveal salient flow features.
• Validated the framework's ability to characterize complex flows unsupervisedly.

Conclusions:

• Nonlinear manifold learning offers a powerful approach for unsupervised flow pattern characterization.
• Isomap provides a computationally efficient alternative to traditional kinematic and kinetic measurements.
• This framework facilitates advancements in real-time flow analysis, control, and distributed sensing.