Mamba-augmented residual network for rapid wake field prediction of underwater vehicles

Establishing an end-to-end mapping from design parameters to wake flow fields of underwater vehicles holds critical importance for enabling rapid design optimization. However, the wake field exhibits complex and nonlinear flow characteristics, making it difficult to accurately predict for conventional methods. This study presents a Mamba-enhanced residual network for efficient and precise reconstruction of the underwater vehicle wake field with variable rudder configurations. Specifically, the network integrates Mamba modules with residual channel attention blocks (RCAB) to synergistically capture local features and global spatial dependencies. Furthermore, a self-attention mechanism enhances the model's sensitivity to the rudder shape design variables. Extensive validation experiments on the standard DARPA SUBOFF model, benchmarked against several state-of-the-art methods, consistently demonstrate the superior capability of the proposed network in predicting the detailed wake field structures for a wide range of rudder design configurations.