I've noticed some problems with the turbulence models available in Fluent for predicting pressure loss in axial-flow turbomachinery fan and compressor bladings.

Has anyone done any validation of Fluent for this type of applications? Which turbulence models do you normally use?

The case I ran recently was a fan OGV. This OGV has a high Re number (2 million), a fairly high Ma number (close to Ma 1 on the suction side) and the blade is fairly thin with a quite sharp leading edge.

At first I used our "standard" turbulence modeling in Fluent, that is non-equilibrium wall functions and realizable k-epsilon for design iterations and low-Re two-layer realizable k-epsilon modeling for validation simulations. Both these simulations showed too high losses and I quickly tracked the problem down to an overproduction of turbulent energy in the leading edge stagnation region.

I know that this is a common problem with many k-epsilon models, but the variable Cmu in Fluent's realizable model usually reduces this stagnation problem. For this thin blade it does not seem to be able to handle it very well though. I've used this model in Fluent before with success for more low-speed, lower-Ma and thicker blading without seeing these leading-edge problems. I also know that Shih's non-linear variant of this model works well also for high-speed thin compressor type of blading (I used it in my own code as a PhD student 10 years ago).

I've also tried the Spalart-Almares model in Fluent and this model produces better results. However, I would prefer to use a better two-equation model which is more likely to predict stall etc. better.

Hence, I continued to experiment with the Kato-Launder option which is available with the standard k-epsilon model in Fluent (in the text-interface). However, I found to my dissapointment that the Kato-Launder fix is not available with the low-Re models in Fluent and does not seem to work with enhanced wall-treatment. It only works with wall-functions. This is a pity. Otherwise Kato-Launder is a very effective way of taking away the over-production of turbulent energy in the leading edge region.

I also tried the SST k-omega model but found that also this model suffers from the leading-edge stagnation region overproduction of turbulent energy and that this produces too high losses. This is a bit surpricing. I have also run this model in CFX and there I don't see any sign of this problem. CFX most likely has a different implementation of the SST model using some limiters or something which takes away this problem. Anyone know what the difference is between CFX's and Fluent's SST models?

The conclusion from my tests is that for axial-flow fan and compressor blading the only model in Fluent which produces reasonable loss predictions is the fairly simple one-quations Spallart-Allmares model.(I also ran some tests with RSM and got a bit better results than with k-epsilon, but I don't consider RSM as a robust enough model to use for design simulations).

Anyone have any other ideas or suggestions?

Please check your e-mail