Successful turbulence modeling requires engineering judgment of:

a) Flow physics

b) Computer resources available

c) Project requirements

i) Accuracy

ii) Turnaround time

d) Choice of Near-wall treatment

General modeling procedure to be followed:

1) Calculate characteristic Reynolds number and determine whether flow is turbulent.

2) Estimate wall-adjacent cell centroid y+ before generating the mesh.

3) Prepare the mesh to use wall functions except for low-Re flows and/or flows with complex near-wall physics (non-equilibrium boundary layers).

4) Begin with Standard k-ε and then change to RKE (realizable k-ε), S-A (Spalart-Allmaras), RNG k-ɛ, SKW (Standard k-w), SST k-w if needed.

5) Use RSM for highly swirling, 3-D, rotating flows.

6) Validate the results with experimental data for choosing the right turbulence model for further calculations.

7) Remember that there is no single, superior turbulence model for all flows.

Since your flow domain is S shaped, there would be severe adverse pressure gradients in the flow and re circulation in the flow may occur. Hence, a suitable and low computational cost model in Ansys Fluent is "Realizable k–ε model".

Which likely to provide superior performance for flows involving rotation, vortex shedding behind bluff bodies, boundary layers under strong adverse pressure gradients, separation, and re-circulations.

For better accuracy, you can go towards Reynolds Stress models which are computationally expensive, need more space and time.

But before choosing any turbulence model, it is a general practice to use several suitable turbulence models and validate with experimental data.

Thank You

k–ε model should be a good choice.

try k-epsilon and k-omega models