Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-Driven Approaches for Thrust Prediction in Underwater Flapping Fin Propulsion Systems

Julian Lee, Kamal Viswanath, Alisha Sharma, Jason Geder, Ravi Ramamurti, Marius D. Pruessner

TL;DR

This paper tackles thrust prediction for underwater flapping-fin propulsion under constrained environments by developing data-driven surrogate models conditioned on fin geometry and kinematics. The authors introduce a discrete, dynamic fin-geometry parameterization that tracks 30 centers of mass and rotates with kinematics, and they apply PCA-based reduced-order representations to compress inputs. They compare dense and LSTM neural networks for instantaneous and full-stroke thrust prediction, evaluating multiple parameterizations (30-fin, reduced kinematic, and 4-parameter reduced) using an experimental dataset with multiple geometries and motion settings. The results show that fin-parameterized networks can generalize to unseen geometries, with LSTM models delivering smoother thrust profiles; however, aggressive dimensionality reduction can degrade generalization, highlighting trade-offs for onboard control deployment and design exploration in unmanned underwater vehicles. Overall, the study demonstrates promising pathways for fast, data-driven thrust prediction to accelerate design and control of bio-inspired UUV propulsion systems, supported by a structured generalizability framework and real experimental data.

Abstract

Flapping-fin underwater vehicle propulsion systems provide an alternative to propeller-driven systems in situations that require involve a constrained environment or require high maneuverability. Testing new configurations through experiments or high-fidelity simulations is an expensive process, slowing development of new systems. This is especially true when introducing new fin geometries. In this work, we propose machine learning approaches for thrust prediction given the system's fin geometries and kinematics. We introduce data-efficient fin shape parameterization strategies that enable our network to predict thrust profiles for unseen fin geometries given limited fin shapes in input data. In addition to faster development of systems, generalizable surrogate models offer fast, accurate predictions that could be used on an unmanned underwater vehicle control system.

Data-Driven Approaches for Thrust Prediction in Underwater Flapping Fin Propulsion Systems

TL;DR

This paper tackles thrust prediction for underwater flapping-fin propulsion under constrained environments by developing data-driven surrogate models conditioned on fin geometry and kinematics. The authors introduce a discrete, dynamic fin-geometry parameterization that tracks 30 centers of mass and rotates with kinematics, and they apply PCA-based reduced-order representations to compress inputs. They compare dense and LSTM neural networks for instantaneous and full-stroke thrust prediction, evaluating multiple parameterizations (30-fin, reduced kinematic, and 4-parameter reduced) using an experimental dataset with multiple geometries and motion settings. The results show that fin-parameterized networks can generalize to unseen geometries, with LSTM models delivering smoother thrust profiles; however, aggressive dimensionality reduction can degrade generalization, highlighting trade-offs for onboard control deployment and design exploration in unmanned underwater vehicles. Overall, the study demonstrates promising pathways for fast, data-driven thrust prediction to accelerate design and control of bio-inspired UUV propulsion systems, supported by a structured generalizability framework and real experimental data.

Abstract

Flapping-fin underwater vehicle propulsion systems provide an alternative to propeller-driven systems in situations that require involve a constrained environment or require high maneuverability. Testing new configurations through experiments or high-fidelity simulations is an expensive process, slowing development of new systems. This is especially true when introducing new fin geometries. In this work, we propose machine learning approaches for thrust prediction given the system's fin geometries and kinematics. We introduce data-efficient fin shape parameterization strategies that enable our network to predict thrust profiles for unseen fin geometries given limited fin shapes in input data. In addition to faster development of systems, generalizable surrogate models offer fast, accurate predictions that could be used on an unmanned underwater vehicle control system.
Paper Structure (18 sections, 11 figures, 7 tables)

This paper contains 18 sections, 11 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Neural Network Parameters
  4. Representing Fin Geometries
  5. Reduced-Order Fin Parameterization
  6. Neural Network Architectures
  7. Experimental Setup
  8. The Dataset
  9. Data Pre-Processing
  10. Data Noise Evaluation
  11. Generalizability Testing
  12. Neural Network Training
  13. Results
  14. Baseline Neural Networks
  15. Fin Parameterization Neural Networks
  16. ...and 3 more sections

Figures (11)

  • Figure 1: Example simple flapping fin propulsion system composed of paired pectoral (side) fins.
  • Figure 2: Example division of flat fin into 10 equal-area regions. Resulting center of masses shown in gray.
  • Figure 3: Experimental data fin geometries for the rectangular fin (left), bio fin (middle), and pt4 fin (right)
  • Figure 4: Location of regional center of masses for flat fins
  • Figure 5: Sample experimental data for a kinematic-shape setting. A sample interval is shown in between the blue bars.
  • ...and 6 more figures