Three-dimensional super-resolution reconstruction of turbulent spray flow fields

Super-resolution (SR) with deep neural networks has emerged as a promising technique for enhancing the resolution of turbulent flow fields in experiments and numerical simulations. Substantial research has been conducted on single-phase turbulent flows, but little effort has been made for turbulent spray combustion. This paper presents the first application of three-dimensional SR to two-way coupled particle-laden flows with an SR factor of eight. Appropriate data treatment is investigated. An SR model is developed that is capable of learning and reconstructing the momentum interaction between turbulence and spray droplets. The results demonstrate that particle information is an essential input for SR models to correctly learn and reconstruct the subgrid modulation of turbulence by particles. A data padding mode consistent with the boundary conditions of the flow fields and sample-wise data normalization are recommended based on the test outcomes. Moreover, the applicability of a recently proposed spectral loss to three-dimensional turbulence SR is verified.