Inferring Structural Motion and Properties in Fluid Through Wake Flow Sensing

Abstract Understanding the dynamics and properties of structures moving in fluid flows is crucial for applications in underwater robotics, offshore systems, and marine sensing. As an object moves in the fluid with a Reynolds number above 10, periodic vortex shedding can form, creating unique wake flow signatures. This work explores a hybrid framework for inferring structural motion and properties by analyzing wake flow signatures generated by structures in the fluid. By leveraging machine learning techniques that are rotation-invariant, the proposed approach decouples the orientation of the structure from other latent features that influence the wake flow signatures. These latent features are then used to infer the structural properties via a neural network model. To evaluate the effectiveness of the approach, we generate diverse wake flow signatures from bluff bodies using computational fluid dynamics models. Despite the high-dimensional variability of wake patterns, the machine learning model identifies a few key nonlinear features that describe the variations. These features are found to be distinctive for different structural motions and properties in the fluid. We demonstrate the success of this hybrid framework and highlight its advantages over traditional methods. The developed inference model for wake flow sensing can add a new modality to monitor the states of underwater systems.