Classifying vortex wakes using neural networks

TL;DR

This paper demonstrates that neural networks can classify vortex wake patterns from local vorticity data, enabling flow analysis without full flow field knowledge, which is useful for aquatic sensing and flow control.

Contribution

The study introduces a neural network approach to classify vortex wakes using local measurements, providing a new method for flow pattern recognition.

Findings

Neural networks accurately classify three vortex wake types.

The network identifies key features distinguishing wake patterns.

Local vorticity data suffices for flow pattern classification.

Abstract

Unsteady flows contain information about the objects creating them. Aquatic organisms offer intriguing paradigms for extracting flow information using local sensory measurements. In contrast, classical methods for flow analysis require global knowledge of the flow field. Here, we train neural networks to classify flow patterns using local vorticity measurements. Specifically, we consider vortex wakes behind an oscillating airfoil and we evaluate the accuracy of the network in distinguishing between three wake types, 2S, 2P+2S and 2P+4S. The network uncovers the salient features of each wake type.