SURFNet: Super-Resolution of Turbulent Flows with Transfer Learning using Small Datasets

Obiols-Sales, Octavi; Vishnu, Abhinav; Malaya, Nicholas; Chandramowlishwaran, Aparna

doi:10.1109/pact52795.2021.00031

Cited by 15 publications

(8 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Following this study, supervised CNN-based super-resolution analysis has been actively studied for a range of flows. Obiols-Sales et al [57] proposed a CNN-based super-resolution model called SURFNet and tested its performance for wakes around various NACA-type airfoils, ellipses, and cylinders. SURFNet includes a transfer learning-based augmentation [92].…”

Section: Supervised Learningmentioning

confidence: 99%

See 1 more Smart Citation

Super-resolution analysis via machine learning: a survey for fluid flows

Fukami

Fukagata

Taira

2023

Theor. Comput. Fluid Dyn.

View full text Add to dashboard Cite

This paper surveys machine-learning-based super-resolution reconstruction for vortical flows. Super resolution aims to find the high-resolution flow fields from low-resolution data and is generally an approach used in image reconstruction. In addition to surveying a variety of recent super-resolution applications, we provide case studies of super-resolution analysis for an example of two-dimensional decaying isotropic turbulence. We demonstrate that physics-inspired model designs enable successful reconstruction of vortical flows from spatially limited measurements. We also discuss the challenges and outlooks of machine-learning-based super-resolution analysis for fluid flow applications. The insights gained from this study can be leveraged for super-resolution analysis of numerical and experimental flow data. Graphical abstract

show abstract

Section: Supervised Learningmentioning

confidence: 99%

“…Wurster et al [154] proposed a hierarchical GAN to perform super resolution of fluid flows. Analogous to SURFNet [57], a hierarchical GAN is first trained with low-resolution data sets. The model weights are then transferred to training with higher-resolution flow fields.…”

Section: Semisupervised and Unsupervised Learningmentioning

confidence: 99%

Super-resolution analysis via machine learning: a survey for fluid flows

Fukami

Fukagata

Taira

2023

Theor. Comput. Fluid Dyn.

View full text Add to dashboard Cite

show abstract

“…Fukami et al [31] presented a standard CNN and an improved hybrid Downsampled Skip-connection Multi-scale (DSC/MS) model to reconstruct super-resolution of laminar and turbulent flows. Obiolssales et al [32] developed Super-resolution Flow Network (SURFNet), a novel approach to reconstruct fine-scale flow physics from coarse grid data that can be used to accelerate high resolution turbulent Computational Fluid Dynamics (CFD) simulations. Liu et al [33] proposed two models, namely Static CNN (SCNN) and Multiple Temporal Paths CNN (MTPC), to address the super-resolution reconstruction of turbulent flows from low resolution coarse flow field data.…”

Section: Deep Learning-based Super-resolution Generation For Fluid Flowsmentioning

confidence: 99%

“…Obiols‐sales et al. [32] developed Super‐resolution Flow Network (SURFNet), a novel approach to reconstruct fine‐scale flow physics from coarse grid data that can be used to accelerate high resolution turbulent Computational Fluid Dynamics (CFD) simulations. Liu et al.…”

Section: Related Workmentioning

confidence: 99%

Pix2PixSSR: Spatial super‐resolution synthesis and visualization for time‐varying volumetric data

Ma,

Chen

2023

IET Image Processing

View full text Add to dashboard Cite

Upscaling of the time‐varying volume data is significant, since it can be used in in situ visualization to help scientists fast analyse complex simulations relevant to time‐varying volume data. A novel deep learning method called Pix2Pix spatial super‐resolution (Pix2PixSSR), which can be used to generate spatial super‐resolution of the time‐varying volume data is proposed here. It consists of two main components: One is a variant UNet‐like generator that takes the low resolution volume sequence as input and generates the high resolution counterparts; one is a PatchGAN discriminator that takes both low and high resolution volume sequences as input and predicts their realness. To validate its advantages, we qualitatively and quantitatively compare it with the state‐of‐the‐art upscaling techniques. More specifically, two experiments are performed. The first experiment uses the same variable of a time‐varying volume dataset for training and inference, while the second experiment uses different variables for training and inference. The experimental results show that for most cases, the Pix2PixSSR can generate the most similar super‐resolution to the ground truth, compared to the state‐of‐the‐art techniques.

show abstract

“…The idea is that the knowledge or important features of one problem gained by training the former neural nets can be transferred to other problems. Transfer learning has been widely used in image recognition (Yin et al 2019;Jin, Cruz, and Gonc ¸alves 2020), natural language processing (Ruder et al 2019) and recently in PINNs (Goswami et al 2019;Obiols-Sales et al 2021;Song and Tartakovsky 2022;Desai et al 2021). To the best of our knowledge, the present work is the first work to employ transfer learning for learning solution operators of evolutionary PDEs.…”

Section: Introductionmentioning

confidence: 97%

Transfer Learning Enhanced DeepONet for Long-Time Prediction of Evolution Equations

Wang

2023

AAAI

View full text Add to dashboard Cite

Deep operator network (DeepONet) has demonstrated great success in various learning tasks, including learning solution operators of partial differential equations. In particular, it provides an efficient approach to predicting the evolution equations in a finite time horizon. Nevertheless, the vanilla DeepONet suffers from the issue of stability degradation in the long- time prediction. This paper proposes a transfer-learning aided DeepONet to enhance the stability. Our idea is to use transfer learning to sequentially update the DeepONets as the surro- gates for propagators learned in different time frames. The evolving DeepONets can better track the varying complexities of the evolution equations, while only need to be updated by efficient training of a tiny fraction of the operator networks. Through systematic experiments, we show that the proposed method not only improves the long-time accuracy of Deep- ONet while maintaining similar computational cost but also substantially reduces the sample size of the training set.

show abstract

SURFNet: Super-Resolution of Turbulent Flows with Transfer Learning using Small Datasets

Cited by 15 publications

References 23 publications

Super-resolution analysis via machine learning: a survey for fluid flows

Super-resolution analysis via machine learning: a survey for fluid flows

Pix2PixSSR: Spatial super‐resolution synthesis and visualization for time‐varying volumetric data

Transfer Learning Enhanced DeepONet for Long-Time Prediction of Evolution Equations

Contact Info

Product

Resources

About