Numerical Study and Parameter Prediction in Hydrodynamic Performance of Self-Propelled Wiggling Hydrofoils

Fish employ their bodies and caudal fins to generate a counter-propagating traveling wave, which, in conjunction with the surrounding fluid dynamics, results in a propulsion pattern that is both rapid and efficient. From a mathematical perspective, these waves can be represented as traveling waves. This study focuses on simulating the kinematic state of a NACA65-010 self-propelled hydrofoil, operating under low Reynolds number uniform incoming flow, utilizing the immersed boundary method. Additionally, we investigate the alterations in the fish’s swimming performance when it actively oscillates by observing the hydrodynamic performance of the model under various oscillation parameters. To streamline the numerical simulation process, we employ the long short-term memory network (LSTM), time-sequential convolutional network (TCN), and transformer model to predict the lift and thrust coefficients of the self-propelled hydrofoil. The outcomes demonstrate that both the trailing edge amplitude and vibration frequency significantly influence the model’s propulsion performance, enabling the hydrofoil to generate countercurrent thrust at low Reynolds numbers. Moreover, the average lift and thrust of the self-propelled hydrofoil exhibit gradual increases over time during the forward motion. The LSTM prediction model exhibits superior accuracy and goodness of fit in forecasting the hydrodynamic parameters, with an average absolute error of lift and thrust coefficients below 0.19 and a goodness-of-fit exceeding 0.968. Furthermore, the implementation of this model reduces the overall computational burden of the experimental process by 25%. Through the exploration of oscillation parameters’ impact on hydrodynamic performance, this research sheds light on the underlying mechanisms of fish’s active swimming. Furthermore, the utilization of deep learning techniques alleviates the computational costs and memory requirements associated with traditional computational fluid dynamics (CFD) methods