A temporal super-resolution model for turbulence prediction based on optical flow and artificial neural network

This paper introduces a temporal super-resolution model based on optical flow and artificial neural network to predict the short-term evolution of turbulence in fluid mechanics. The model employs the Farnebäck algorithm to compute optical flow and generate a basic predicted flow field. However, some regions in the flow field do not fully satisfy the usage conditions of the optical flow algorithm, causing significant local errors. Therefore, a multilayer perceptron (MLP) is integrated into the model to learn the relationship between local gradients and prediction errors, providing correction values to improve the prediction accuracy of the optical flow method. To evaluate the model's generalization ability, it is applied to a range of compressible wall turbulence cases with Mach numbers spanning from 2.25 to 10, encompassing typical flow structures such as turbulent boundary layers, shock waves, and separation bubbles. The results demonstrate that the model can accurately reconstruct flow fields between two moments, with the MLP providing significant improvements in the prediction of temperature and density fields using the optical flow method. The computational efficiency of the present model is significantly higher than that of traditional computational fluid dynamics methods.