[TurBoris]

Hi All,

Currently we have been conducting a high speed centrifugal compressor stage tests that has a pressure ratio more than 5 at its design point.

Our pre-test CFD results are off at some points, thus we have been testing all parameters (turbulence model, mesh, test geometry differences ets) in our post -test CFD models.

I doubt the radial location of the mixing-plane interface would influence the results. We have placed the interface middle of the vaneless space (1.05 x impeller exit radius). Can this 5% offset from impeller trailing edge change the diffuser inlet absolute Mach number ? (my thesis: an increase of abs velocity by letting higher tip speed in velocity triangles) In real operation the flow enters the absolute frame where impeller blades ends. However, in CFD the interface is generally located 5% away from tip radius for meshing issues as you know.

Before I do a simulation with a new interface location I would like to get your comments and the math (if possible) behind this modeling technique.

Thank you.

[ghorrocks]

Have you tried using a frozen rotor interface? You can somewhat account for the frozen rotor's simplification of just modelling a single point in the rotation by sweeping it across a full pitch. Or if you want to go the whole way try a transient rotor-stator simulation.

Both the frozen rotor and transient rotor-stator approaches do not do averaging at the interface, so this will tell you if this averaging is an issue.

[Opaque]

First of all, we need to differentiate modeling results from experimental data. If we do not do that, we do not a solid basis to derive conclusions.

Once you pick a set of equations (let us call it model), the solution is a function of the discretization error and you want to make the solution independent of those errors: mesh independent, and time step independent (assuming transient).

If your solution is discretization error independent, and it does not match your experimental data where do you look next?
1 - Experimental error, i.e. do we know the bounds of our measurements
2 - "Model translation/selection", i.e. is the selected model appropriate for the experimental setup?

In the case of using the mixing plane, what is the benchmark? Theoretically, the mixing plane is a reduced model of the transient rotor-stator interaction between the components. The time-averaged solution is the benchmark.

I am not particularly fond of the frozen-rotor approach anymore. It was ok 20 years ago. You must be careful, the trailing wake is too long to be physically possible since the next pitch interaction will disturb it.