A rapid reconstruction method for pressure loads in nonlinear liquid sloshing

Existing studies rarely address rapid reconstruction of distributed sloshing pressure loads, despite their importance for tank structural safety. To address this issue, this study proposes an end-to-end method integrating reduced-order modeling and deep learning for full-field pressure reconstruction from wall pressure signals. First, training samples are constructed based on Latin Hypercube Sampling (LHS), and a sloshing state classification model is developed using the Froude number and numerical simulation results to automatically distinguish between linear and nonlinear sloshing. Then, a common modal basis is constructed, and a Weighted Proper Orthogonal Decomposition (WPOD) method is introduced to improve the reconstruction accuracy in impact-dominated regions under nonlinear sloshing. Finally, a bidirectional long short-term memory (Bi-LSTM) network is employed to establish the mapping relationship between pressure responses and modal coefficients, enabling distributed liquid sloshing pressure field reconstruction. The results show that the proposed method achieves high reconstruction accuracy under both linear and nonlinear sloshing conditions. For strongly nonlinear sloshing (Froude number approximately 0.23), the minimum coefficient of determination R2 reaches 0.946, while for linear sloshing, R2 exceeds 0.967. Meanwhile, the computational efficiency is significantly improved. This study provides a new approach for efficient reconstruction of liquid sloshing pressure loads.