Using Neural Networks to Predict Turbulent Flow from Heat Flux Measurements
Author Information
Author(s): Guastoni Luca, Balasubramanian Arivazhagan G., Foroozan Firoozeh, Güemes Alejandro, Ianiro Andrea, Discetti Stefano, Schlatter Philipp, Azizpour Hossein, Vinuesa Ricardo
Primary Institution: KTH Royal Institute of Technology
Hypothesis
Can fully convolutional networks effectively predict velocity fields in turbulent boundary layers using wall heat flux data?
Conclusion
The study demonstrates that fully convolutional networks can accurately predict turbulent flow fields from wall heat flux measurements, paving the way for non-intrusive sensing in practical applications.
Supporting Evidence
- Fully convolutional networks were trained on numerical data and fine-tuned with experimental data.
- The model achieved a 50% lower error compared to previous architectures.
- Predictions were validated against experimental data from a water tunnel facility.
- Different Prandtl numbers were tested to assess the model's robustness.
- Transfer learning was employed to enhance prediction accuracy with limited experimental data.
Takeaway
This study shows that we can use smart computer programs to guess how fast the air is moving near a wall just by looking at how hot the wall is.
Methodology
The study used fully convolutional networks trained on numerical simulations and fine-tuned with experimental data to predict velocity fields from wall heat flux measurements.
Limitations
The predictions are less accurate when using experimental data due to limited data availability and discrepancies between simulation and experimental conditions.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website