With the insane growth of data science, I notice that there's hardly any data-driven fluid dynamics research out there. What could explain this phenomenon?
I have heard that fluid dynamicists generally want to do their research based on first principles, i.e. use physics, and not use data-driven methods that might not care for domain knowledge.
Having said that, I'm still somewhat surprised that not more fluid dynamicists have done data-driven work.
Well, for people doing physics research, especially mathematical analysis, data-driven approaches are basically useless. A deep NN can't really guarantee that its output is realistic nor can it guarantee that it will generalize to situations never seen in the training data. If one wants to understand the underlying mechanisms, then ML can't really help usually.
I know there are also people in physics-based animation that don't use it either, since for some applications deep NNs are too slow and/or hard to train. Maybe they are less interested in it as well. Either way, this could likely change soon.
On the other hand, there are plenty of data-driven fluid dynamics work! For example:
Kutz, Deep learning in Fluid Dynamics, 2017
Zhang & Duraisamy, Machine Learning Methods for Data-Driven Turbulence Modeling, 2015
Beck et al, Deep Neural Networks for Data-Driven Turbulence Models
Wiewel et al, Latent-space Physics: Towards Learning the Temporal Evolution of Fluid Flow, 2018