Convergence problem in RANS simulation of unsteady flow

I am simulating an air flow through a two dimensional rectangular channel. The channel splits into two channels, one of them is followed by a 90 degree bend downwards and leads into a cylindrical volume through which the air flows out of the geometry.
Here is a picture of the geometry:

For the simulation I use the RANS approach with the k-omega SST model. However, the solution doesn’t really converge. I located the problem in the area of the transition from the channel into the cylinder where the jet bends towards one side but doesn’t completely stabilize there. Instead, it keeps flapping around which can be seen in an animation of a velocity contour over a few hundred of iterations:

I assume that such a flow situation is indeed pretty unstable not only numerically but physically too. My question is the following: Given the transient nature of the flow shouldn’t the RANS model just average out these fluctuations? Can a transient flow lead to instabilities in a steady RANS simulation? Do you know papers about this topic?

I would really appreciate your help!