fanyuchang/awesome-machine-learning-fluid-mechanics

Curated list for ML in FM

Awesome Machine Learning for Fluid Mechanics

A curated list of machine learning papers, codes, libraries, and databases applied to fluid mechanics. This list in no way a comprehensive, therefore, if you're the author of any relevant content then please feel free to add it here.

Table of Contents

• Awesome Machine Learning for Fluid Mechanics
  • Frameworks
  • Research articles
    • Editorials
    • Review papers
    • Quantum Machine Learning
    • Interpreted Machine Learning
    • Physics-informed ML
    • Reduced-order modeling aided ML
    • Pattern identification and experimental applications
    • Others
  • ML-focused events
  • Available datasets
  • Online resources
  • Blogs, discussions and news articles
  • Ongoing researchs, projects and labs
  • Opensource code and examples
  • Tutorials
  • Companies focusing on ML
  • Opensource CFD codes

Frameworks

• TensorFlow is a well-known machine learning library developed by Google.

• PyTorch is another framework for machine learning developed at Facebook.

• Scikit-learn is all-purpose machine learning library. It also provides the implementation of several other data analysis algorithm.

• easyesn is a very good implementation of echo state network (reservoir computing). ESN often finds its application in dynamical systems.

• EchoTorch is another good implementation for ESN based upon PyTorch.

• flowTorch is a Python library for analysis and reduced order modeling of fluid flows.

• PySINDy is a package with several implementations for the Sparse Identification of Nonlinear Dynamical systems (SINDy). It is also well suited for a dynamical system.

• PyDMD is a python package for dynamic mode decomposition which is often used for reduced order modelling now.

• PYPARSVD is an implementation for singular value decomposition (SVD) which is distributed and parallelized which makes it efficient for large data.

Research articles

Editorials

1. Editorial: Machine Learning and Physical Review Fluids: An Editorial Perspective, 2021.

Review papers

1. Application of machine learning algorithms to flow modeling and optimization, 1999. (Paper)

2. Turbulence modeling in the age of data, 2019. (Paper)

3. A perspective on machine learning in turbulent flows, 2020. (Paper)

4. Machine learning for fluid mechanics, 2020. (Paper)

5. A Perspective on machine learning methods in turbulence modelling, 2020. (Paper)

6. Machine learning accelerated computational fluid dynamics, 2021. (Paper)

7. Deep learning to replace, improve, or aid CFD analysis in built environment applications: A review, 2021. (Paper)

Quantum Machine Learning

1. Machine learning and quantum computing for reactive turbulence modeling and simulation, 2021. (Paper)

2. Quantum reservoir computing of thermal convection flow, 2022. (Paper)

Interpreted Machine Learning

1. An interpretable framework of data-driven turbulence modeling using deep neural networks, 2021. (Paper)

2. Interpreted machine learning in fluid dynamics: explaining relaminarisation events in wall-bounded shear flows, 2022, (Paper | Data)

Physics-informed ML

1. Reynolds averaged turbulence modeling using deep neural networks with embedded invariance, 2016. (Paper)

2. From deep to physics-informed learning of turbulence: Diagnostics, 2018. (Paper)

3. Subgrid modelling for two-dimensional turbulence using neural networks, 2018. (Paper | Code)

4. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, 2019. (Paper)

5. Neural network models for the anisotropic Reynolds stress tensor in turbulent channel flow, 2019. (Paper)

6. Data-driven fractional subgrid-scale modeling for scalar turbulence: A nonlocal LES approach, 2020. (Paper)

7. A machine learning framework for LES closure terms, 2020. (Paper)

8. A neural network based shock detection and localization approach for discontinuous Galerkin methods, 2020. (Paper)

9. Stable a posteriori LES of 2D turbulence using convolutional neural networks: Backscattering analysis and generalization to higher Re via transfer learning, 2021. (Paper)

10. Data-driven algebraic models of the turbulent Prandtl number for buoyancy-affected flow near a vertical surface, 2021. (Paper)

11. Convolutional Neural Network Models and Interpretability for the Anisotropic Reynolds Stress Tensor in Turbulent One-dimensional Flows, 2021. (arXiv)

12. Physics-aware deep neural networks for surrogate modeling of turbulent natural convection (arXiv)

13. Learned Turbulence Modelling with Differentiable Fluid Solvers (arXiv] )

14. Physics-informed data based neural networks for two-dimensional turbulence (arXiv] | Paper)

Reduced-order modeling aided ML

1. Reservoir computing model of two-dimensional turbulent convection, 2020. (Paper)

2. Predictions of turbulent shear flows using deep neural networks, 2019. (Paper | Code)

3. Reduced-order modeling of advection-dominated systems with recurrent neural networks and convolutional autoencoders, 2020. (Paper | Code)

4. Time-series learning of latent-space dynamics for reduced-order model closure, 2020. (Paper | Code)

5. A deep learning enabler for nonintrusive reduced order modeling of fluid flows, 2019. (Paper)

6. Echo state network for two-dimensional turbulent moist Rayleigh-Bénard convection, 2020. (Paper)

7. DeepCFD: Efficient steady-state laminar flow approximation with deep convolutional neural networks, 2020. (Paper | Code)

8. From coarse wall measurements to turbulent velocity fields with deep learning, 2021. (Paper)

9. Convolutional neural network and long short-term memory based reduced order surrogate for minimal turbulent channel flow, 2021. (Paper, | Data: Contact authors)

10. Direct data-driven forecast of local turbulent heat flux in Rayleigh–Bénard convection, 2022. (arXiv | Paper | Data: Contact authors)

Pattern identification and experimental applications

1. Deep learning in turbulent convection networks, 2019. (Paper)

2. Time-resolved turbulent velocity field reconstruction using a long short-term memory (LSTM)-based artificial intelligence framework, 2019. (Paper)

3. Unsupervised deep learning for super-resolution reconstruction of turbulence, 2020. (Paper)

4. Nonlinear mode decomposition with convolutional neural networks for fluid dynamics, 2020. (Paper)

Others

1. Forecasting of spatiotemporal chaotic dynamics with recurrent neural networks: a comparative study of reservoir computing and backpropagation algorithms, 2019. (Paper)

2. Data-assisted reduced-order modeling of extreme events in complex dynamical systems, 2018. (Paper)

3. Nonlinear mode decomposition with convolutional neural networks for fluid dynamics, 2020. (Paper)

4. Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations, 2020. (Paper)

5. Engine Combustion System Optimization Using Computational Fluid Dynamics and Machine Learning: A Methodological Approach, 2021. (Paper)

6. Physics guided machine learning using simplified theories, 2021. (Paper | Code)

ML-focused events

1. International Workshop on Data-driven Modeling and Optimization in Fluid Mechanics, 2019, Karlsruhe, Germany.

2. Symposium on Model-Consistent Data-driven Turbulence Modeling, 2021, Virtual Event.

3. Turbulence Modeling: Roadblocks, and the Potential for Machine Learning, 2022, USA.

4. Mini symposia: Analysis of Real World and Industry Applications: emerging frontiers in CFD computing, machine learning and beyond, 2022, Yokohama, Japan.

Available datasets

1. KTH FLOW: A rich dataset of different turbulent flow generated by DNS, LES and experiments. (Simulation data | Experimental data | Paper-1)

2. Vreman Research: Turbulent channel flow dataset generated from simulation, could be useful in closure modeling. (Data | Paper-1 | Paper-2)

3. Johns Hopkins Turbulence Databases: High quality datasets for different flow problems. (Database | Paper)

4. CTR Stanford: Dataset for turbulent pipe flow and boundary layer generated with DNS. (Database | Paper)

5. sCO2: Spatial data along the tube for heated and cooled pipe under supercritical pressure. It includes around 50 cases, which is a good start for regression based model to replace correlations. (Data | Paper-1 | Paper-2)

Online resources

• A first course on machine learning from Nando di Freitas: Little old, recorded in 2013 but very concise and clear. (YouTube | Slides)

• Steve Brunton has a wonderful channel for a variety of topics ranging from data analysis to machine learning applied to fluid mechanics. (YouTube)

• Nathan Kutz has a super nice channel devoted to applied mathematics for fluid mechanics. (YouTube)

• For beginners, a good resource to learn OpenFOAM from József Nagy. OpenFOAM can be adapted for applying ML model coupled with N-S equations (e.g. RANS/LES closure). (YouTube)

• A course on Machine learning in computational fluid dynamics from TU Braunschweig.

Blogs, discussions and news articles

1. Convolutional Neural Networks for Steady Flow Approximation, 2016. (Autodesk)

2. CFD + Machine learning for super fast simulations, 2017. (Reddit)

3. What is the role of Artificial Intelligence (AI) or Machine Learning in CFD?, 2017. (Quora)

4. Supercomputing simulations and machine learning help improve power plant, 2018.

5. When CAE Meets AI: Deep Learning For CFD Simulations, 2019. (Ubercloud)

6. Machine Learning in Computational Fluid Dynamics, 2020. (TowardsDataScience)

7. Studying the nature of turbulence with Neural Concept's deep learning platform, 2020. (Numeca)

8. A case for machine learning in CFD, 2020. (Medium)

9. Machine Learning for Accelerated Aero-Thermal Design in the Age of Electromobility, 2020. (Engys)

10. A general purpose list for transitioning to data science and ML, 2021.

11. A compiled list of projects from NVIDIA where AI and CFD were used, 2021.

12. AI for CFD, 2021. (Medium)

13. 4 Myths about AI in CFD, 2021. (Siemens)

Ongoing researchs, projects and labs

1. Center for Data-Driven Computational Physics, University of Michigan, USA.

2. VinuesaLab, KTH, Sweden.

3. DeepTurb: Deep Learning in and of Turbulence, TU Ilmenau, Germany.

4. Thuerey Group: Numerical methods for physics simulations with deep learning, TU Munich, Germany.

Opensource code and examples

• Repositiory machine-learning-applied-to-cfd has some excellent examples to begin with CFD and ML.

• Repository Computational-Fluid-Dynamics-Machine-Learning-Examples has an example implementation for predicting drag from the boundary conditions alongside predicting the velocity and pressure field from the boundary conditions.

• Image Based CFD Using Deep Learning

• Deep-Flow-Prediction has the code for data generation, neural network training, and evaluation.

• TensorFlowFoam with few tutorials on TensorFlow and OpenFoam.

Tutorials

• Tutorial on the Proper Orthogonal Decomposition (POD) by Julien Weiss: A step by step tutorial including the data and a Matlab implementation. POD is often used for dimensionality reduction.

Companies focusing on ML

• Neural Concepts is harnessing deep learning for the accelerated simulation and design.

• Flowfusic is a cloud based provider for CFD simulation based upon OpenFOAM. They are exploring some use cases for AI and CFD.

• byteLAKE offers a CFD Suite, which is a collection of AI models to significantly accelerate the execution of CFD simulations.

Opensource CFD codes

Following opensource CFD codes can be adapated for synthetic data generation. Some of them can also be used for RANS/LES closure modeling based upon ML.

1. Nek5000
2. OpenFOAM
3. PyFr
4. Nektar++
5. Flexi
6. SU2