I haven't done CFD in a long time so my opinion might have little value.
With that said, it sounds like when it comes to turbulence the issue is predicting if you're likely to have some within a specific region, without doing the calculation itself.
Machine learning could be this predictor I guess. The way I can envision it being implemented, is that maybe there is a criterion that could be found by analyzing the vector flux on the edges of that region. Above some value, you'd likely have turbulence within.
Then you could maybe combine it to some analysis of the structures inside, cracks, etc, based on the RMS, friction coefficients and so on.
But now the issue would be how to define those criterions, and this is definitely outside my area of expertise
image source
