Optimisation of SA-PGω Turbulence Model for Multi-Stage Compressor at Off-Design Conditions

Abstract The Spalart-Allmaras (SA) turbulence model is commonly used for compressor simulations but often overestimates blockage size, leading to an underprediction of the stall margin. To address this shortcoming, the SA-PGω model was proposed. This model utilities the dimensionless vortical pressure gradient to identify blockage regions characterized by 3D swirling, adverse pressure gradients driven, and low-momentum flows. It mitigates blockage by increasing eddy viscosity in these regions. The model includes three adjustable coefficients, initially calibrated using rotor exit radial profile data from NASA Rotor 67 under peak-efficiency and near-stall conditions, establishing a suitable coefficients range. This paper presents a data-driven approach to fine-tune the SA-PGω model coefficients for industrial multi-stage compressors. A surrogate model based on an artificial neural network (ANN) is first trained to predict the aerodynamic performance of a 4-stage compressor. Based on the surrogate model, a multi-objective optimisation is conducted to determine the optimal coefficients by minimizing errors relative to experimental data. The quantities of interest (QoIs) used for optimisation are the surge margin and overall pressure ratio at two mass flow rates. The trained model is initially validated by predicting stage performance and radial profiles of the same compressor. It is then applied to predict the performance of a 5-stage compressor that has not been seen by the optimisation. Overall, the CFD results show good prediction accuracy compared with measurements, highlighting the versatility of the SA-PGω model for multi-stage compressor applications.