A new body force-based modelling approach for centrifugal compressor post-stall aerodynamics

Abstract

A lower-order modelling approach is developed, whereby the compression system is modelled as an empty duct, with body force-fields imparting turning and losses to the flow. The flow-field solution is obtained transiently by solving the axisymmetric Euler equations with metal and aerodynamic blockage in the relative frame of reference. The solver is coupled with a fully custom body force model consisting of a base turning force, correction force and viscous force component. Anew blockage-mixing model is derived providing estimations for aerodynamic blockage as well as the associated mixing losses and change in flow direction. Separation losses are introduced via the turning force coupled with a shape function which allows adjustment of the reattachment position. The blockage-mixing and separation modelling methods are verified against 3D RANS data, providing consistently reliable predictions. The steady-state, reverse flow performance predictions at different speeds are compared against 3D RANS simulations, showing excellent agreement. Surge simulations are subsequently carried out using an 1D plenum model. As the experimental rig’s plenum details are not provided, the plenum volume is varied until the experimentally reported surge cycle frequency is achieved. Unsteady pressure measurements are compared against the experimental ones, showing excellent agreement in the entire range of the surge cycles. Flow-field contours at different time instances are in concert with the experimentally obtained DPIV measurements, accurately capturing the onset of stall in the vaned diffuser. The simulations indicate deep surge behaviour, confirming conjectures from the experimental report which could not eventually be confirmed during testing. The code constitutes the first successful, validated, through-flow approach, capable of predicting the post-stall, steady-state and transient aerodynamic performance in centrifugal compressors, while reducing the computational cost by at least one order of magnitude compared to conventional CFD modelling approaches.