Parameter Optimization of the Bio-Inspired Robot Propulsion Through the Deep Learning Based Reduced Order Fluid-Structure Interaction Model

Zixiang Ying,Linxiang Wang,Roderick Melnik
DOI: https://doi.org/10.1016/j.oceaneng.2022.111436
IF: 5
2022-01-01
Ocean Engineering
Abstract:In this paper, an effective model (POD-NIROM) is proposed, which makes full use of Long short-term memory Neural Network (LSTM NN) and proper orthogonal decomposition (POD) to predict the fluid dynamics around the moving boundary, as well as the soft robot locomotion. This is the first time that the proposed model has been used to optimize the propulsion parameters of the robotic fish, in which the body stiffness of the robot has been taken into consideration. To discuss the reliability of the presented model, the prediction and simulation of robot swimming performance are compared with experimental measurements, the results show a consistent trend. Finally, the trained model is used to optimize the robot's propulsion parameters. The results show that the robot has the best propulsion speed and propulsion force when the dimensionless wavenumber k˜ is about 0.7. Compared with the high-fidelity model, the average relative standard deviation of the present model is 2.60%, but the calculation cost is reduced by 99.4%. Reasonable prediction and efficient calculation make the proposed POD-NIROM has great potential in the process of bio-inspired robot swimming prediction or bio-inspired robot propulsion parameter optimization.
What problem does this paper attempt to address?