For me the main questions are:

- Which problems within Fluid Mechanics are not well resolved with 'traditional' methods and can gain by using some ML/AI approaches? I.e., where / how can we use ML/AI to solve problems that we could not solve before?

- How can we combine 'traditional' and ML/AI approaches so that they complement each other?

I see at least two aspects on which a lot of research is taking place that may fit these criteria:

- defining closure models for LES / RANS simulations. See for example:

Deep neural networks for data-driven LES closure models, A Beck, D Flad, CD Munz, Journal of Computational Physics 398, 108910

Subgrid modelling for two-dimensional turbulence using neural networks, R Maulik, O San, A Rasheed, P Vedula, Journal of Fluid Mechanics 858, 122-144

- performing active flow control / control of dynamic systems arising from Fluid Mechanics, see for example:

Artificial Neural Networks trained through Deep Reinforcement Learning discover control strategies for active flow control, J Rabault, M Kuchta, A Jensen, U Reglade, N Cerardi, Journal of Fluid Mechanics

Control of chaotic systems by Deep Reinforcement Learning, MA Bucci, O Semeraro, A Allauzen, G Wisniewski, L Cordier, L Mathelin, Proc. R. Soc. A 475, 20190351

- In addition, people are suggesting to use ML/AI for identification of flow structures, for example:

Prediction of wall-bounded turbulence from wall quantities using convolutional neural networks, L Guastoni, MP Encinar, P Schlatter, H Azizpour, R Vinuesa

Of course these are only a few of the papers published on the topic, I would recommend that you also consult the literature citing them / cited by them, as usual.

Are you asking if (or assuming that) CFD has reached the level of sophistication that physical models (pipes and tanks) and field measurements (pitot tubes, velocimeters, laser-doppler, etc.) are no longer necessary? I can assure you that this is not true from personal experience. In many cases it depends on what you want to know and how accurate you want to be. If I paid a hundred million dollars for a cooling tower, I wouldn't accept your computer model results and I wrote the software that is the industry standard. I'm not suggesting that nothing can be gained or learned from developing and applying machine learning algorithms and computer modeling, just that the world is much more complex than you might think. One of the funniest things I've experienced in a while happened when my wife (in the USA) asked Siri for directions to an address on a street named Coventry Court. Siri started giving *driving* directions to Coventry, England. Take I-40 to Virginia, no... recalculating, take I-95 to Maine, no... recalculating. I was a little disappointed that Siri didn't suggest swimming across the Atlantic.

To be honest, AI, ML and so on are more like a present "fashion" , a trend.... I don't think that is nothing but an "interpolation" reproduced in a sophisticated way. Have a look to this discussion https://www.cfd-online.com/Forums/main/205618-artificial-intelligence-cfd.html

I recommend

https://www.cambridge.org/core/journals/journal-of-fluid-mechanics/article/deep-learning-in-fluid-dynamics/F2EDDAB89563DE5157FC4B8342AD9C70

https://arxiv.org/pdf/1808.04327

https://arxiv.org/pdf/1905.11075

https://web.stanford.edu/group/ctr/ResBriefs99/petros.pdf

https://pdfs.semanticscholar.org/194f/99282e7313db099fe37e2eaaadb9ca3fdfd8.pdf

Yes, the field of ML for fluid dynamics seems to have an explosive interest confirmed e.g. by the recent reviews:

Article Turbulence Modeling in the Age of Data

Article Machine Learning for Fluid Mechanics

For instance, ML techniques can help to enhance RANS models, which are not universal, give poor predictions and have no visible progress during last twenty years. Note that eddy-resolving tools (DNS, LES, hybrid LES-RANS) are still too expensive for flows in natural conditions (atmosphere, ocean) with large ranges of length/time scales, and improved RANS methods will be needed at least during the next few decades.

As others have pointed out, it is not actually correct that the community hasn't done anything in the field (notice some posted references are even from 1999). However, artificial intelligence is a very large field and it means all and nothing as, quoting from wikipedia: AI is a system’s ability to correctly interpret external data, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation.

The question is, what would you expect it to do in fluid dynamics, which is, as well, a very large field (let me say much larger)?

If we agree that, probably, it is the Machine Learning subfield that is more suited to a possible scientific task, then you should note that a large fraction of the techniques used in the field, especially those related to unsupervised learning, were already mastered by fluid dynamics people without attaching a ML label to them.

In my opinion, except for the larger availability of frameworks and computing power, there is little added today that was not available 20 years ago.

Also, with the obvious disticntion between cases where alternatives exist and those where they don't, for the former I see much more scientific trouble than possible gains, the largest one being reproducibility of results.

For me the main questions are:

- Which problems within Fluid Mechanics are not well resolved with 'traditional' methods and can gain by using some ML/AI approaches? I.e., where / how can we use ML/AI to solve problems that we could not solve before?

- How can we combine 'traditional' and ML/AI approaches so that they complement each other?

I see at least two aspects on which a lot of research is taking place that may fit these criteria:

- defining closure models for LES / RANS simulations. See for example:

Deep neural networks for data-driven LES closure models, A Beck, D Flad, CD Munz, Journal of Computational Physics 398, 108910

Subgrid modelling for two-dimensional turbulence using neural networks, R Maulik, O San, A Rasheed, P Vedula, Journal of Fluid Mechanics 858, 122-144

- performing active flow control / control of dynamic systems arising from Fluid Mechanics, see for example:

Artificial Neural Networks trained through Deep Reinforcement Learning discover control strategies for active flow control, J Rabault, M Kuchta, A Jensen, U Reglade, N Cerardi, Journal of Fluid Mechanics

Control of chaotic systems by Deep Reinforcement Learning, MA Bucci, O Semeraro, A Allauzen, G Wisniewski, L Cordier, L Mathelin, Proc. R. Soc. A 475, 20190351

- In addition, people are suggesting to use ML/AI for identification of flow structures, for example:

Prediction of wall-bounded turbulence from wall quantities using convolutional neural networks, L Guastoni, MP Encinar, P Schlatter, H Azizpour, R Vinuesa

Of course these are only a few of the papers published on the topic, I would recommend that you also consult the literature citing them / cited by them, as usual.

Dear Aamir Sultan,

I would like to suggest you some useful links : Article Machine Learning for Fluid Mechanics

https://arxiv.org/pdf/1905.11075.pdf

https://kth.diva-portal.org/smash/get/diva2:1215657/FULLTEXT02.pdf

Article Application of Machine Learning and Artificial Intelligence ...