A deep-learning approach for reconstructing 3D turbulent flows from 2D observation data.

Area of Science:

• Fluid dynamics
• Computational science
• Machine learning

Background:

• Turbulent flows exhibit complex, chaotic behavior across multiple scales, challenging accurate prediction.
• High-fidelity flow data is often inaccessible for full-scale applications due to cost and complexity.
• Deep learning offers a potential solution for reconstructing flow fields from limited data.

Purpose of the Study:

• To develop a generative adversarial network (GAN)-based model for reconstructing 3D velocity fields.
• To reconstruct full flow fields from sparse, unpaired 2D velocity observations.
• To reduce the cost and complexity of obtaining full-scale flow data.

Main Methods:

• Utilized a generative adversarial network (GAN) architecture.
• Trained the model on cross-plane 2D velocity observations.
• Reconstructed 3D velocity fields from the 2D input data.

Main Results:

• The GAN model successfully reconstructed 3D flow fields.
• Accurate flow structures, statistics, and spectra were preserved in the reconstructions.
• The model demonstrated effectiveness in reconstructing 3D flows from 2D experimental measurements.

Conclusions:

• The developed GAN-based model enables accurate 3D flow reconstruction from 2D data.
• This approach significantly reduces experimental setup complexity and data storage requirements.
• The method holds promise for applications requiring cost-effective, high-fidelity flow analysis.