A two-dimensional hydrodynamics prediction framework for mantle-undulated propulsion robot using multiple proper...

Area of Science:

• Robotics
• Fluid Dynamics
• Artificial Intelligence

Background:

• Predicting hydrodynamic forces is crucial for designing efficient underwater robots.
• Traditional methods often struggle with the complex fluid dynamics of undulating propulsion.

Purpose of the Study:

• To develop a deep learning framework for predicting hydrodynamic forces on mantle-undulated propulsion robots (MUPRo).
• To introduce an efficient multiple proper orthogonal decomposition (MPOD) algorithm for fluid feature identification.

Main Methods:

• Developed a deep learning framework integrating a novel MPOD algorithm.
• Utilized a long short-term memory neural network to predict temporal coefficients of MPOD spatial modes.
• Validated predictions against simulations and experimental data.

Main Results:

• The MPOD algorithm demonstrated near-linear convergence of L2 error to 0.2% near the undulating boundary.
• The developed framework achieved hydrodynamic force predictions one to two orders of magnitude more accurate than classical POD methods.
• The framework proved economical and reliable for MUPRo applications.

Conclusions:

• The proposed MPOD-based deep learning framework offers significant improvements in predicting hydrodynamic forces.
• This model supports offline parameter planning for aquatic-inspired robots.
• The approach provides a reliable tool for advancing MUPRo design and control.