based on my expericence, I do not think RANS give a correct answ, especially near stall. Stall is very complex, try LES, but the simulation accuracy still needs to be validated by experiments. All in all, rotating stall is very hard to get a accurate simulation result.
Dhinakaran Veeman - thank you for the paper. It sheds some light on the problem I am working on.
Filippo Maria Denaro - Could you please explain in more details why do you doubt that RANS can give an adequate prediction of pre-stall event? I have been planning to deploy metrics, like blockage intensity of blockage factor to determine how close is the compressor to stall.
The are some methods proposed, like in the paper by Xie et al (2018):
Article Effect of RANS Method on the Stall Onset Prediction by an Ei...
on how to deploy RANS to evaluate the pre-stall behaviour correctly. I was thinking of less computation intensive method, that could be implemented in the post processing stage.
Renfang Huang - I do agree that rotating stall is a complex phenomenon. But simply saying it 'can't be done' is not a satisfactory answer. Are you aware of any research paper comparing say the experiments with LES and RANS simulations, pointing out the pros and cons of each of the approaches?
despite the large use in industrial applications, RANS is well known to have poor results in flow problem with massive separation and recirculation and in the prediction of quantities such as skin friction and heat flux at the walls.
I second to the comments of Prof Denaro, Dr Huang et al., the accurate onset of the stalls cells and further development to surge is difficult with RANS and high-fidelity formulations must be used. This being said, RANS can give some direction for industrial cases by means of pressure/mass flow fluctuations.
Other than these, acoustic spectra and temperature increase (in the inducer due to recirculation) is also correlated with the onset of stall.
If you are looking in the blockage as a means to ascertain stall, I would take a look at the thesis of Khalid S A (Article The effects of tip clearance on axial compressor pressure rise /
- Chapter 3) and his contemporaries in the group.
Also, please note that my comment is based on single-stage compressors.
From the review of Prof. Day (DOI: 10.1115/1.4031473), we now know there are mainly three types of flow instability in a compressor, namely rotating instability, rotating stall and surge in ascending order of spatial and temporal scales. It is often the development of small-scale rotating instability (i.e., sub-passage vortical structures) triggers the mid-scale rotating stall (i.e., rotation of sub-annulus separation bubbles/stall cells), and finally leads to the large-scale surge (i.e., fluctuation of flow quantities in the whole piping system).
Based on the physical understanding behind the compressor 'stall', it is now easier to understand why RANS may not be predictive for - you have to resolve the small-scales (i.e., turbulence) first before getting the right prediction of mid/large scales. That's why the choice of RANS turbulence model may yield a 10%-20% uncertainty of numerical stall mass flow (i.e., last converged point with the smallest mass flow) for single/multi-stage axial compressors. I am not familiar with the eigen-based method, but since the RANS flow field is served as an input for this method, I suppose there is a similar level of uncertainty for the choice of RANS turbulence models. Currently, no consensus on the best RANS turbulence model for compressor has been reached. In terms of popularity, SA (usually in-house codes and NUMECA) and SST (usually ANSYS CFX) are among the top list. It should be noted that SST did better than SA in 2D separation flows, but there is no research showing SST is consistently superior in compressor flows featured by 3D separations, and in practice, it did differ case-by-case. Either RANS model can be used, but you need to compare your simulations against the experiment in terms of flow fields especially near the endwalls before going further. Such validation will give you an idea about the capabilities and the limits of your numerical model.
On the other hand, LES/DNS is not within reach for compressor 'stall' now. Although you can resolve the small scales by using LES/DNS with a refined mesh, you need a large flow domain (i.e., inlet duct, compressor stages, outlet pipe, plenum, and nozzle) to capture the large scales. That means the mesh size for your simulation will be very large. Just to have an idea about the mesh size - according to the work of Dr. Hah (DOI: 10.1115/1.4035521), a full-annulus 1.5 stage low-speed (M