Three-dimensional ESRGAN for super-resolution reconstruction of turbulent flows with tricubic interpolation-based transfer learning

Abstract

Turbulence is a complicated phenomenon because of its chaotic behavior with multiple spatio-temporal scales. Turbulence also has irregularity and diffusivity, making predicting and reconstructing turbulence more challenging. This study proposes a deep-learning approach to reconstruct three-dimensional (3D) high-resolution turbulent flows from spatially-limited data using a 3D enhanced super-resolution generative adversarial networks (3D-ESRGAN). In addition, a novel transfer-learning method based on tricubic interpolation is employed. Turbulent channel flow data at friction Reynolds numbers Reτ = 180 and Reτ = 500 were generated by direct numerical simulation (DNS) and used to estimate the performance of the deep-learning model as well as that of the tricubic interpolation-based transfer learning. The results, including instantaneous velocity fields and turbulence statistics, show that the reconstructed high-resolution data agree well with the reference DNS data. The findings also indicate that the proposed 3D-ESRGAN can reconstruct 3D high-resolution turbulent flows even with limited training data.

A transformer-based synthetic-inflow generator for spatially developing turbulent boundary layers

“…2021, 2022 b ; Yu et al. 2022). In a GAN model that is used for image generation, two adversarial neural networks called the generator () and the discriminator () compete with each other.…”

“…(Goodfellow et al 2014) have shown great success in image transformation and super-resolution problems (Mirza & Osindero 2014;Ledig et al 2017;Zhu et al 2017;Wang et al 2018). Generative adversarial network-based models have also shown promising results in reconstructing HR turbulent flow fields from coarse data (Fukami et al 2019a;Fukami, Fukagata & Taira 2021;Güemes et al 2021;Kim et al 2021;Yousif et al , 2022bYu et al 2022). In a GAN model that is used for image generation, two adversarial neural networks called the generator (G) and the discriminator (D) compete with each other.…”

“…Deep learning is a subset of machine learning, where deep neural networks are used for classification, prediction and feature extraction (LeCun, Bengio & Hinton 2015). Recently, several models have shown great potential in solving different problems in the field of turbulence, such as turbulence modelling (Wang, Wu & Xiao 2017;Duraisamy, Iaccarino & Xiao 2019), turbulent flow prediction (Lee & You 2019;Srinivasan et al 2019), reduced-order modelling (Nakamura et al 2021, flow control (Rabault et al 2019;Fan et al 2020;Park & Choi 2020;, non-intrusive sensing (Guastoni et al 2021;Güemes et al 2021) and turbulent flow reconstruction (Deng et al 2019;Fukami, Fukagata & Taira 2019a;Kim et al 2021;Yousif, Yu & Lim 2021, 2022bEivazi et al 2022;Yu et al 2022) .…”

Explaining wall-bounded turbulence through deep learning

“…A convolutional neural network (CNN) is used for predicting the velocity field, as discussed by Schmekel et al [30]. Note that CNNs and other computer-vision architectures have been successfully used in the context of turbulent-flow predictions [28,[58][59][60][61]. The convolution operation is described by Equation ( 4), where f is the input three-dimensional (3D) tensor, h the filter, G the output, and m, n and p the indices of the output tensor:…”

Physics-constrained deep reinforcement learning for flow field denoising

“…2021; Yu et al. 2022; Yousif et al. 2023 a ), where deep learning is a subset of machine learning, in which neural networks with multiple layers are used in the model (LeCun, Bengio & Hinton 2015).…”