Login
- Title:
High resolution dataset for flow past arbitrary bodies
- Document type:
- Forschungsdaten
- Publication date:
- 18.09.2023
- Responsible:
- Brahmachary, Shuvayan
- Authors:
- Brahmachary, Shuvayan ; Thuerey, Nils
- Author affiliation:
- TUM
- Publisher:
- TUM
- Identifier:
- doi:10.14459/2023mp1720279
- End date of data production:
- 08.02.2023
- Subject area:
- DAT Datenverarbeitung, Informatik; MAS Maschinenbau
- Other subject areas:
- Computational Data Science ; Computational Aerodynamics
- Resource type:
- Simulationen / simulations; Statistik und Referenzdaten / statistics and reference data
- Data type:
- Tabellen / tables; Datenbanken / data bases
- Other data type:
- NPZ
- Description:
- This dataset comprises the high-resolution (768 x 512) flowfields for incompressible flow past arbitrarily shaped bodies at low Reynolds number (100, approx.) obtained using FoamExtend based immersed boundary method. The dataset is divided into 3 parts viz. (a) Training dataset containing 50 experiments having 100 frames each (27.5 GB, approx), (b) testing datset containing 50 additional experiments having 300 frames each (82.5 GB, approx), and (c) total dataset (i.e., 100 experiemnts) having 1500 frames each (825 GB, approx). Part (a) forms the training dataset used to train the differentiable physics-assisted neural network (DPNN) whereas part (b) forms the testing dataset for the trained DPNN based model. Finally, part (c) serves as additional dataset for further validation and benchmarking.
The dataset only includes the stastically stable velocity flowfields (i.e., time 150s onwards spaced at 0.1s) along with shape mask. The 100 experiments comprises of 100 different geometric configurations (i.e, arbitrary bodies placed in an equilateral triangle position), each at a unique spacing ratio. In addition, the coordinates of the underlying bodies are also provided separately. - Method of data assessment:
- The dataset describes the spatio-temporal behaviour of fluid flow obtained on a structured Cartesian grid using open-source flow solver FoamExtend v4.0. The compressed .npz format of the dataset can be easily loaded, visualised and post-processed in Python. The post-processing tools provided in the following Github repository can be used for the same. https://github.com/tum-pbs/DiffPhys-CylinderWakeFlow/
- Links:
The dataset is related to the publication https://arxiv.org/abs/2308.04296- Key words:
- Machine learning ; Computational Fluid Dynamics ; Differentiable Physics
- Technical remarks:
- View and download (881 GB total, 4 Files)
The data server also offers downloads with FTP
The data server also offers downloads with rsync (password m1720279):
rsync rsync://m1720279@dataserv.ub.tum.de/m1720279/ - Language:
- en
- Horizon 2020:
- CoG- 2019-863850
BibTeX