FLAME 2023: Diving into the Future of Fluid Dynamics and Machine Learning

2023-10-11 · Source: Blog of the TransferLab — appliedAI Institute · Field: Technology & Digital — Artificial Intelligence & Machine Learning, AI for Scientific Computing · Depth: Advanced, medium

Summary

The Stanford FLAME AI Workshop 2023 explored the integration of machine learning with fluid dynamics, turbulence, and combustion, featuring expert lectures and an ML challenge. Speakers included George Karniadakis on Physics-informed Neural Networks (PINNs), Anima Anandkumar on Neural Operators, Stephan Hoyer on Deep Learning with Differentiable Physics, and Steve Brunton on Machine Learning for Scientific Discovery. A central component was a Kaggle ML challenge focused on super-resolution in turbulent flows, where participants aimed to upscale 16x16 pixel images to 128x128 pixels using a dataset of over 1000 training samples. The author achieved 10th place using a simple upscaling method with convolutional layers, while leading approaches often utilized EDSR modifications or a super-resolution neural operator.

Key takeaway

For AI Scientists and Research Scientists working on fluid dynamics or complex simulations, consider integrating machine learning techniques like physics-informed neural networks or neural operators to enhance model accuracy and accelerate computational processes. Your team could explore super-resolution methods for turbulent flow simulations, as demonstrated by the FLAME AI challenge, to improve efficiency and potentially achieve higher fidelity results with smaller datasets.

Key insights

AI techniques are transforming simulation by integrating physics-based modeling with data-driven machine learning.

Principles

• Physics-informed neural networks enhance traditional simulations.
• Neural operators learn mappings between function spaces.
• Super-resolution can accelerate turbulent flow simulations.

Method

The FLAME AI challenge involved reconstructing high-resolution (128x128) turbulent flow images from low-resolution (16x16) inputs using a dataset of over 1000 samples.

In practice

• Apply PINNs for physics-constrained ML tasks.
• Explore neural operators for complex system modeling.
• Utilize super-resolution for accelerating fluid dynamics simulations.

Topics

• Neural Operators
• Physics-informed Neural Networks
• Fluid Dynamics Simulation
• Image Super-resolution
• Turbulent Flows

Code references

• samuelburbulla/FLAME-AI
• google-deepmind/graphcast

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, AI Researcher, Machine Learning Engineer, AI Student

Related on AIssential

• AI-powered design comes to high-speed racing — AI surrogate models, like GIST, can drastically accelerate complex physics simulations while maintaining high accuracy.
• A Validated LBM Dataset and Pipeline for Surrogate Modeling of Turbulent 3D Obstructed Channel Flows — A validated LBM dataset and pipeline enable standardized evaluation of neural operators for 3D turbulent flows.
• Climate Physicists Face the Ghosts in Their Machines: Clouds — Cloud modeling is a critical bottleneck in climate prediction, driving new AI-physics hybrid approaches.
• Advances in Scientific Machine Learning for Coupled Fluid Flow and Transport — SciML, combining reduced-order models and neural networks with HPC, efficiently approximates complex coupled fluid flow and transport, significantly cutting computational costs.
• Learning Physical Operators using Neural Operators — A physics-informed framework improves neural operator generalization and temporal continuity for PDE solving via operator splitting.
• AI Models Trained on Physics Are Changing Engineering — Large physics models drastically accelerate engineering design by replacing or augmenting traditional physics simulations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Blog of the TransferLab — appliedAI Institute.