Yes, the concept of Y+ and wall modeling are both based on the assumption of wall bounded turbulent flow and this assumption would collapse for separated flow. For the case of separated flow, the Y+ concept and wall function would become inadequate for a universal modeling.

I agree with Guangzhong Gao 高广中 . Research in this area is still ongoing—particularly regarding non-equilibrium wall functions, which are specifically designed to handle adverse pressure gradients, unsteady flows, and separation more effectively than equilibrium models, which assume a balance between production and dissipation. Of course, such models are not yet perfect.

y+ is a general law, not a value... it simply address the non-dimensional distance from a wall in terms of the stress velocity.

y+ = (utau*L/ni)(y/L)=(Retau)*y*

In a turbulent BL the flow is 3D and separated but we can assume that is laminar-like (viscous sub-layer) at y+ =O(1). That is no separation for the mean flow appears.

The "separation" concept is therefore to be applied to the mean flow, for example a separation due to the geometry, like the flow behind an obstacle.

Here the definition of boundary layer and distance from the wall has no longer sense in terms of inner variable.

In other words, we have to limit the role of the boundary layer in the range of attached mean flow.

Hi! You ask multiple interesting questions here.

"Is y+ ≈ 1 Always Optimal in LES/DES?"

Well to me it really depends on the complexity of the case you want to simulation. y+ ≈ 1 should be used in what is called Wall-Resolved LES (WRLES) which is maybe the most accurate approach after DNS. Yet WRLES often are very expensive and can rarely be applied in complex cases. If the Reynolds number is too large, Wall-Modeled LES (WMLES) should be prefered and Detached Eddy Simulation (DES) is one way to do it by using RANS turbulence model at the wall.

"Does the Traditional y+ Concept Collapse Under An Considerable Flow Separation?"

As mentioned in an earlier answer, it is not y+ per se that collapses but the relationship between u+ and y+ that follows a universal law known as the Von Karman law-of-the-wall derived a hundred years ago. For this law to be applicable, the boundary layer has to remain attached of course. And so even though you have the right to consider y+ regardless of the type of flow, the usual u+/y+ relationship will not verified in this case.

"Should We Shift from Universal Approaches to Wall Modeling on A Case-by-Case Basis?"

This is a very interesting question as it is core in today's work for theoricists in LES and has strong implications. Today's consensus is that probably there is no single mathematical function that can describe u+=f(y+) and that would be valid in all cases. Current research trends comprise the coupling of 3D RANS models (taking into account pressure gradient and convective terms) at the wall with LES away from the wall and/or ML-based wall models to account for these out-of-equilibrium phenomena

Cheers!