Dataset of a parameterized U-bend flow for deep learning applications.

Area of Science:

• Computational Fluid Dynamics (CFD)
• Design Optimization
• Machine Learning

Background:

• Fluid flow and heat transfer in U-bend geometries are critical in many engineering applications.
• Developing efficient design optimization methods requires extensive simulation data.
• Existing datasets may lack the diverse representations needed for advanced machine learning techniques.

Purpose of the Study:

• To introduce a comprehensive dataset of 10,000 fluid flow and heat transfer simulations in U-bend shapes.
• To provide a benchmark for evaluating design optimization algorithms and deep learning approaches.
• To facilitate research by offering multiple data representations, including mesh-based data.

Main Methods:

• Utilized Computational Fluid Dynamics (CFD) to generate 10,000 simulations.
• Characterized each simulation by 28 design parameters.
• Developed three distinct data representations: parameter-objective combinations, 2D image resolutions, and numerical mesh cell values.

Main Results:

• A dataset of 10,000 U-bend fluid flow and heat transfer simulations is now available.
• The dataset includes diverse data types suitable for various machine learning paradigms (supervised, semi-supervised, unsupervised).
• Source code and data generation containers are published, ensuring reproducibility and accessibility.

Conclusions:

• This dataset serves as a valuable resource for advancing design optimization and machine learning in fluid dynamics.
• The inclusion of mesh-based data representation opens new avenues for applying deep learning to simulation data.
• The open-source nature of the data generation tools promotes transparency and further research in the field.