Application of data-driven method in flow field reconstruction of unmanned underwater vehicle pump-jet propulsor

To address the difficulty of obtaining high fidelity flow field information for unmanned underwater vehicle pump-jet propulsor, the accuracy of super-resolution flow field reconstruction by using a hybrid down-sampling skip connection/multi-scale reconstruction model of data-driven methods is investigated. This model can use the nonlinear distribution of bias and weight to establish a complex relationship between the low resolution and the super-resolution flow fields. Comparing with the experimental and numerical simulation methods, which has the advantages of high efficiency, low cost, and high accuracy. Meanwhile, the uncertainty distribution of super-resolution reconstruction of pump-jet propulsor was analyzed by using the variational Bayesian theory. The results indicate that the hybrid data-driven model with variational Bayesian theory has higher accuracy in reconstructing the flow field of pump-jet propulsor, and high reconstruction errors are mainly distributed in the hub area and rotor blade rotation area. The present method can increase the low resolution flow field of the unmanned underwater vehicles pump-jet propulsor by 256 times, high uncertainty area is mainly distributed at the peak of the reconstructed curve.