Fish-lateral-inspired pressure sensing neural networks for underwater object identification

In the current project, we numerically study underwater object identification using a hydrofoil with pressure sensors. Using LilyPad, a boundary data immersion method (BDIM) viscous 2D solver, we simulate and generate a large dataset that describes pressure evolution over time around a constant speed moving foil passing by various objects at different distances (ellipses and rectangles of different geometries and orientations). A multivariate convolutional neural network is then constructed and trained on the dataset after the application of the POD dimensionality reduction, mapping the distinctive pressure information of the near body flow feature’s distinctive pressure information to the objects close to the passing foil.