Comparative CFD Simulations of a Soft Robotic Fish for Undulatory Swimming Behaviors

Area of Science:

• Robotics
• Fluid Dynamics
• Machine Learning

Background:

• Autonomous underwater vehicles (AUVs) are increasingly studied, with robotic fish offering superior maneuverability.
• Understanding the hydrodynamic performance of robotic fish is crucial for their development.

Purpose of the Study:

• To propose a prediction strategy for the hydrodynamic performance of robotic fish.
• To analyze undulatory swimming behaviors using computational fluid dynamics (CFD) and deep learning.

Main Methods:

• A 2D robotic fish model was created for CFD simulations.
• A dynamic network method was used to orient the network based on wavy motion.
• Deep learning models (LSTM, CNN, GRU) were employed for force prediction.

Main Results:

• Kinematic parameters like flapping frequency and speed were analyzed for their effects on swimming efficiency and drag.
• The CNN-GRU model achieved the highest prediction performance with a root mean square error of 0.0228.
• The proposed method showed superior performance compared to CNN and LSTM models.

Conclusions:

• The study successfully proposed a deep learning-based prediction strategy for robotic fish hydrodynamic performance.
• The findings provide insights into optimizing robotic fish swimming efficiency and maneuverability.
• The CNN-GRU model demonstrated the best accuracy in predicting forces, highlighting the potential of deep learning in AUV research.