Hydrodynamic Force Estimation with Inception Neural Networks using Distributed Pressure Measurement

Marine animals achieve efficient and exceptionally agile locomotion through their unique propulsion methods. However, due to the lack of effective computational methods, the mechanism behind their propulsion has remained a mystery. To address this, we propose a hydrodynamic estimation approach based on distributed pressure using the Inception method, which can quickly and accurately predict hydrodynamic forces without the need for complex fluid-structure interaction calculations. We designed four Inception blocks to construct a convolutional neural network and established relationships between sensor pressure measurements and generated lift force. Simulation results validate the effectiveness of our method, indicating that even with a small number of sensors, lift force can be estimated with high accuracy. The four Inception methods all show good performance in accuracy and can track the lift curve globally and perform well in the peak, valley, and abrupt portions. But V3 and V4 exist different degrees of overfitting. The factorized convolution strategy helps improve computational efficiency.