<p>In recent years, deep learning has made great progress in many fields such as image recognition, natural language processing, and convolutional neural networks have even enabled accurate image and video super-resolution in real-time. Super-resolution is an innovative technique that upscales the resolution of an image or a video. This method enables us to reconstruct high-fidelity images from low-resolution data. With the rapid development of deep learning techniques, various super-resolution models built upon different architectures ranging from the classic Convolutional Neural Networks(CNN) to more recent and promising Generative Adversarial Networks(GAN) have been explored and developed which often deliver state of the art results. However, recent attempts to benefit from temporal correlations in super-resolution, especially those concerning turbulent fluid data have been limited to naive or inefficient architectures. This study performs super-resolution analysis on turbulent flow fields spatially and temporally using various state-of-the-art machine learning techniques to reconstruct high-resolution flow fields from low-resolution flow field data, especially keeping in mind the need for low resource consumption and rapid results production/verification. The dataset used for this study is obtained from a Direct Numerical Simulation of forced isotropic turbulence on a 1024^3 periodic grid, solved using a pseudo-spectral parallel code with a Reynolds number varying around 433. Time integral of viscous terms is performed using integral coefficients. The dataset comprises 3 velocity components and the pressure profile of the steady-state.&nbsp;Multiple deep learning models have been implemented on forced isotropic turbulent flow fields to transform them into highly resolved outputs. Apart from the utilization of different standalone models, we have also used transfer learning. This technique utilizes pre-trained models and allows us to mix them too, so as to minimize the computational resources and the time required for the implementation of the super-resolution models. The advantages presented by this method far exceed the&nbsp;expectations and the outcomes of regular single structure models. The ability to derive different features and layers from different models to construct a new one that encompasses the best of various different models highlights the innovativeness of transfer learning. The results obtained through these models are then compared using MSE, SSIM, PSNR metrics among others in order to evaluate the upscaled results, find the balance between computational power and output quality, and then identify the most accurate and efficient technique for spatial and temporal super-resolution. Finally, we discuss the exhaustive range of applications of super-resolution-based computation for numerical and experimental studies in fluid dynamics and how these could be translated to real-world applications.</p>

A Comparative Study of Various Deep Learning Techniques for Spatio-Temporal Super-Resolution Reconstruction of Forced Isotropic Turbulent Flows

Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2021/volume/85543

ASME 2021 International Mechanical Engineering Congress and Exposition (IMECE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.