Enhancement of experimental and numerical data

Recording and restoring high-speed frames of reactive flows in both DNS simulation and laser diagnostics is highly required to better understand the turbulent combustion, but in most times, limited by the hardware. In previous study has been developed a deeplearning model to interpolate low-speed frames of reactive flows to obtain high-speed reactive flows. The model was built based on the concept of optical flow and a U-net neural network architecture. The model trained on DNS simulation datasets is effective in providing reasonable estimation of the intermediate frames between two consecutive laser diagnostic frames. The proposed data-driven closure model is based on a Generative Adversarial Network previously designed for small-scale turbulence reconstruction. The objective is to obtain a statial and temporal enhancement of the experimental data through the GAN trained on numerical data with similar physical parameters.