Hi all! I try to model a stage of a water pump. I use 'stage' and 'frozen rotor' interface. Results for optimal regime are quite good (well coincidence with experiment). But when I solve the problem for small (non optima)l flow rates (left part of head vs. flow rate curve) results can very differ from experiment (100% or more). What is the reason? What can you recommend to improve the results? Thank you for your responses!

Your pump is probably in the stall regime. This means the flow becomes very unsteady, and you may need to run your case transient (with a transient rotor stator interface) and you may need a finer mesh.

in the off design condition,the flow is more complex then the optimal regime,stall(as mention in Georg), separation, secondary flow,rotor-stator or rotor-volute interaction even the distribution of the flow rate in the passage are all difficult to calculate. i think finer mesh will give a different result of integral value such as head.but may not be more precise to the experiment result ,you could try a grid sensitve test,to make sure it is problem of the quantity of the grid.and take care of the near wall grid,make sure that is work well with solver.but the great number of the grid element may cause the problem of convergency.try different turbulence model.i think for the first one is standard k-epsilon model for it is easy to get convergent, then is the SST,which is a pretty good engineering model for foresee the separation flow.but all of the eddy viscosity model in CFX can not handle the anisotropy turbulence in the rotating frame of refernce for the Coriolis forces,so even if you can get well performent prediction in design point,you can not get the secondary flow in the passage, velocity distribution and turbulence intensity may quiet different from experiment result.these need to be interpretated with theory of turbulence. try transient calculation or even high level model (RSM,LES..)with the RANS result.the rotating effect is considerated in any of the available Reynolds stress transport model in ANSYS CFX.i do not think LES is suiable for the engineer application as i have tried some,for it waste of time and computation source .at the end of LES case ,you may get more than 100G data file to post,well, it is hard to post also.�

Thanks guys!