Training bioinspired sensors to classify flows.

We consider the inverse problem of classifying flow patterns from local sensory measurements. This problem is inspired by the ability of various aquatic organisms to respond to ambient flow signals, and is relevant for translating these abilities to underwater robotic vehicles. In Colvert, Alsalman and Kanso, B&B (2018), we trained neural networks to classify vortical flows by relying on a single flow sensor that measures a 'time history' of the local vorticity. Here, we systematically investigate the effects of distinct types of sensors on the accuracy of flow classification. We consider four types of sensors-vorticity, flow velocities parallel and transverse to the direction of flow propagation, and flow speed-and show that the networks trained using transverse velocity outperform other networks, even when subjected to aggressive data corruption. We then train the network to classify flow patterns instantaneously, using a spatially-distributed array of sensors and a single 'one time' sensory measurement. The network, based on a handful of spatially-distributed sensors, exhibits remarkable accuracy in flow classification. These results lay the groundwork for developing learning algorithms for the dynamic deployment of sensory arrays in unsteady flows.