Session: 1.2 - Machine Learning, Reduced Order Modeling in CFD and Design Optimization

Paper Number: 158198

158198 - An Ai-Powered Seal-Whisker-Inspired Hydrodynamic Sensing System 

Abstract: 

Seal whiskers, or vibrissae, are highly specialized sensory organs that enable these animals to navigate turbulent waters and detect subtle changes in flow patterns. Drawing inspiration from this remarkable natural capability, this paper presents the development of a novel AI-powered sensor array designed to decode complex flow environments and provide real-time insights into fluid dynamics. By mimicking the sensory mechanics of seal whiskers, we propose an array of biomimetic flow sensors capable of identifying the shape and position of upstream objects moving through diverse fluid media.

We have successfully fabricated functional prototypes of miniature sensors that demonstrate high sensitivity, rapid response, durability, repeatability, and multidirectional detection. The sensor design features a stiff whisker insert embedded within a soft polydimethylsiloxane (PDMS) base, integrated with foil-type strain gauges that convert base deformations into measurable electrical signals. Sensor performance has been rigorously validated through a series of tests, including bending, fatigue analysis, dipole flow sensing, and wake detection experiments.

Building on this foundation, we are designing an optimized configuration for a sensor array capable of capturing intricate flow dynamics with high spatial and temporal resolution. To facilitate controlled testing, we have constructed a custom water tank measuring 16 ft × 4 ft × 3.5 ft. A dual-axis towing carriage system is under development to precisely control the movement and positioning of test objects, ensuring consistent and reproducible water trails. A second towing carriage will enable precise maneuvering and data collection from the sensor arrays during experimentation.

The sensor arrays are integrated with advanced AI algorithms designed to predict the shape and position of upstream objects based on signals generated by the whisker array. To train and validate these models, we use two datasets: a simulated dataset with randomly positioned whiskers and varying object shapes within a 3D space, and an experimental dataset collected from water tank trials where whisker arrays detect objects of different shapes and positions. The training and testing scenarios include single-body and multi-body configurations with varying shapes.

Given the time-series nature of the data, we use LSTM, GRU, and RNN architectures to capture temporal dependencies. Embeddings are used for classification and regression tasks, with performance evaluated using metrics such as classification accuracy, F1 score for classification tasks, and normalized mean residual error (MRE), L1, and L2 distances for regression tasks. These models also offer interpretability, providing valuable insights into the sensing mechanisms of the whisker arrays.

In the future, this AI-driven sensor system can be adapted to power autonomous underwater agents for a range of applications, including trail tracking, environment mapping, and other underwater exploration tasks.

This research highlights the potential of combining biomimetic designs with advanced AI to create intelligent sensing systems for complex fluid environments, paving the way for significant advancements in underwater robotics and hydrodynamic research.

potential of a novel AI-powered sensor array designed to decode complex flow environments

and provide real-time insights into fluid dynamics. By mimicking the sensory capabilities of seal

whiskers, we propose an array of biomimetic flow sensors capable of decoding both the

upstream body shape and position of objects moving through various fluid mediums

Presenting Author: Biao Geng Rochester Institute of Technology

Presenting Author Biography: