If you are asking about RANS then use k-w SST. It works well in augmented tubes because it can handle near wall situations better than the other classical models.
If you are not concerned about the computational time then k-w SST is the best choice however, k-epsilon realizable also works reasonably well as it is more robust then k-w SST but obviously less accurate. It is always a trade off between the accuracy and computational resource. I have attached a paper for your reference. Go through section 3.2 Turbulence model selection. Hope that helps!
Hi, if you are asking about RANS then use the k-w SST or RNG K-e turbulence models. But If you are asking about LES then use the Model Wall-Adapting Local Eddy. Good luck Nazaruddin Sinaga