Abstract
Aquatic animals have evolved an astonishing variety of capabilities for efficient navigation and sense of the environment. They achieve these remarkable feats in realtime using limited information and minimal computational power. Our marine and underwater applications would greatly benefit from agents capable of operating in unknown environments using limited information and computation resources for decision making. In this work, we address the issues of object shape detection, flow pattern prediction and motion parameter estimation by developing prediction models using a physics-rich dataset obtained from numerical simulation of fluid flow systems or closed analytical forms. Recent advances in machine learning techniques, availability of physics-rich flow data and access to computation resources provide ideal conditions to design and develop game-changing solutions to these problems. First, we address the inverse problem of identifying an obstacle shape from distant and static measures of the flow field. This capability found in many aquatic animals enables them to navigate murky fluid environments, locate and discriminate obstacles. Here, we present a data-driven model that uses artificial neural networks to process flow data originating from a stationary sensor array located away from an obstacle placed in a flow. The ability of neural networks to estimate complex underlying relationships between parameters, in the absence of any explicit mathematical description is used. The resulting neural network exhibits remarkable effectiveness in predicting unknown obstacle shapes, especially at relatively large distances for which classical linear regression models are entirely ineffectual. These results have far-reaching implications for the design and development of artificial passive hydrodynamic sensing technology. Second, we focus on estimating long-term dynamics of the nonlinear flow systems to make ’reasonable’ flow pattern prediction models. Our work aims to use an efficient hybrid machine learning technique to improve the estimation of long-term statistics while being resilient to short-term anomalies in determining future states of the system. Our hybrid machine learning technique combines a Long Short Term Memory architecture with a novel attention mechanism. We demonstrate the forecasting capability of this framework using time-series data from the well-known chaotic Lorenz system and subsequently apply it to a planar homogeneous turbulence flow field. Using qualitative visual verification and quantitative power spectra analysis, we conclude that our model can learn and predict the long-term dynamics even when short-term forecasting fails owing to the inherent unpredictability of chaotic systems. These results have far-reaching implications for the use of machine learning in fluid mechanics in general. Finally, we use physics-based learning models to estimate the motion parameters of an oscillating hydrofoil accurately. Aquatic animals primarily oscillate their flexible body surfaces to navigate efficiently. Understanding the nature of this aquatic motion is key to the design and development of efficient propulsion in autonomous underwater vehicles related to offshore applications. In this study, we consider heave motion with a superimposed flexible motion of the hydrofoil between the leading edge and the quarter chord. The flexible motion directly affects the dynamics of the leading-edge vortex and the overall propulsive performance of the system by increasing the thrust. The flexible motion of the hydrofoil ensures that the flow remains periodic at larger heave motions. We present a data-driven model that uses artificial neural networks, to address the inverse problem of estimating the heave and flexible motion parameters based on the thrust profile requirements and physics rich flow feature information of the oscillating hydrofoil motion. We solve the forward problem to evaluate flow coefficients and features for a given heave and flexible amplitudes using computationally intensive simulations. The generated datasets are used for training the neural network model and assess its performance. The resulting trained model exhibits remarkable effectiveness in estimating the heave and flexible motion amplitudes. These results highlight the far-reaching implications that such a deep learning approach has for the design and development of efficient bio-inspired propulsion.