Learning to classify wakes from local sensory information

Aquatic organisms exhibit remarkable abilities to sense local flow signals contained in their fluid environment and to surmise the origins~of these flows. For example, fish can discern the~information contained in various flow structures and utilize this~information for obstacle avoidance~and prey tracking. Flow structures created by flapping and swimming bodies are well~characterized in~the fluid dynamics literature; however, such characterization relies on classical methods that use an~external observer to~reconstruct global flow fields. The reconstructed flows, or wakes, are then classified according to the unsteady vortex patterns. Here,~we propose a new approach for wake identification: we~classify the wakes resulting from a flapping~airfoil by applying machine learning algorithms to local~flow information. In particular,~we simulate the wakes of an oscillating airfoil in an~incoming flow, extract the~downstream vorticity information, and train a classifier to learn the different flow structures and classify~new ones. This data-driven approach provides a promising framework for underwater navigation and detection~in application to autonomous bio-inspired vehicles.