Turbulence in Focus: Benchmarking Scaling Behavior of 3D Volumetric Super-Resolution with BLASTNet 2.0 Data

Wai Tong Chung,Bassem Akoush,Pushan Sharma,Alex Tamkin,Ki Sung Jung,Jacqueline H. Chen,Jack Guo,Davy Brouzet,Mohsen Talei,Bruno Savard,Alexei Y. Poludnenko,Matthias Ihme

2023-10-28

Abstract:Analysis of compressible turbulent flows is essential for applications related to propulsion, energy generation, and the environment. Here, we present BLASTNet 2.0, a 2.2 TB network-of-datasets containing 744 full-domain samples from 34 high-fidelity direct numerical simulations, which addresses the current limited availability of 3D high-fidelity reacting and non-reacting compressible turbulent flow simulation data. With this data, we benchmark a total of 49 variations of five deep learning approaches for 3D super-resolution - which can be applied for improving scientific imaging, simulations, turbulence models, as well as in computer vision applications. We perform neural scaling analysis on these models to examine the performance of different machine learning (ML) approaches, including two scientific ML techniques. We demonstrate that (i) predictive performance can scale with model size and cost, (ii) architecture matters significantly, especially for smaller models, and (iii) the benefits of physics-based losses can persist with increasing model size. The outcomes of this benchmark study are anticipated to offer insights that can aid the design of 3D super-resolution models, especially for turbulence models, while this data is expected to foster ML methods for a broad range of flow physics applications. This data is publicly available with download links and browsing tools consolidated at <a class="link-external link-https" href="https://blastnet.github.io" rel="external noopener nofollow">this https URL</a>.

Machine Learning,Computer Vision and Pattern Recognition,Computational Physics,Fluid Dynamics

What problem does this paper attempt to address?

The paper aims to address the following issues:

1. **Data Scarcity**: There is currently a lack of high-fidelity direct numerical simulation (DNS) data for compressible turbulent flows (including both reactive and non-reactive flows), especially in terms of 3D data. The paper addresses this issue by constructing a dataset named BLASTNet 2.0, which includes 744 full-domain samples from 34 different configurations, with a total size of 2.2 TB.

2. **Super-Resolution (SR) Model Evaluation**: To improve applications in scientific imaging, simulation, and turbulence modeling, the paper utilizes the BLASTNet 2.0 dataset to benchmark five different deep learning methods for 3D super-resolution tasks. The study focuses on evaluating the impact of different model architectures, scales, and costs on their predictive performance.

3. **Effectiveness of Physical Loss Functions**: The paper demonstrates that physics-based gradient loss terms remain advantageous as the model scale increases and explores their importance for turbulence super-resolution tasks.

Through these studies, the paper aims to provide valuable insights for designing more effective 3D super-resolution models, particularly in turbulence modeling, and to promote the development of machine learning methods in a wide range of fluid dynamics applications.