The prediction of slamming forces acting on two-dimensional (2D) curved bodies is enhanced through the integration of the force decomposition approach and advanced deep neural network (DNN) techniques. The force decomposition approach focuses on deriving decomposition-based coefficients, while the DNN is employed to capture the implicit relationships between these coefficients and their primary influencing factors. Separate DNN models are developed for each coefficient, trained and tested on databases featured diverse 2D geometries defined by cubic B-Spline curves. High-fidelity Computational Fluid Dynamics (CFD) simulations are used to provide accurate decomposition-based coefficients corresponding to each geometry, ensuring data reliability. An analysis of the DNN hidden layer architecture is also conducted to optimize efficiency and prevent overfitting. Based on the well-trained models, the slamming forces acting on 2D bodies resembling the transverse sections of DTMB 5415 are predicted within the force decomposition framework. Comparisons with CFD results demonstrate that the proposed method achieves a good level of effectiveness and accuracy. In addition, further quantitative research reveals the relationship between body dimensions and individual force components. • Predictions of the 2D slamming force are facilitated using a method combining the force decomposition approach and deep neural networks. • Deep neural networks are used to find the implicit relationship between different decomposition-based coefficients and their influencing factors. • The relationships between body dimensions and different force components are investigated by quantitative research.
Sui et al. (Tue,) studied this question.