Novel SEConv1D framework for real-time hydrodynamics prediction of the unidentified underwater vehicle

Abstract

The accurate and rapid prediction of hydrodynamic characteristics greatly affects the monitoring of the manoeuvring performance of unidentified underwater vehicles. This work proposes a novel SEConv1D framework for the hydrodynamics prediction of the unidentified underwater vehicle. Firstly, a novel framework integrated with one-dimensional convolutional neural networks (Conv1D) and a squeeze-and-excitation network (SENet) is proposed. Secondly, a hydrodynamic dataset based on computational fluid dynamics (CFD) is constructed and verified by experiments. Finally, the proposed framework is applied to predict the total resistance coefficient (Cd) of REMUS UUV and SUBOFF AFF-1. The predicted results agree well with experimental results, and the error of Cd between the experimental data and predicted results for REMUS UUV and SUBOFF AFF-1 is less than 3.39% and 1.97%, respectively, which proved that the proposed framework is effective. Compared with those of the most popular networks (i.e., support vector machine, multilayer perceptron, artificial neural network and Conv1D), the mean error of the Cd and friction resistance coefficient (Cf) between the CFD and predicted results is small at only 0.19%, depicting reductions of 91.5%, 85.9%, 66.7% and 40.6%, respectively. The per inference time of the proposed framework is only 0.164 s form the real-time prediction.