High fidelity simulations of three compressor cascades and a fundamental trailing edge flow are used to interrogate, and improve upon, the errors present in conventional RANS simulations. Using the optimum turbulence viscosity, calculated from the high fidelity simulations, it is found that error in moderate Reynolds number (Re_c>300,000) compressor cascades is dominated by errors due to the turbulence model, as opposed to errors inherent to the use of a turbulence viscosity or the RANS method itself. Significant errors due to using turbulence viscosity based RANS to model turbulence are found in a compressor cascade operated at a chord-wise Reynolds number of 148,000.

Focusing on a compressor cascade's suction surface boundary layer, it is found that much of the error when using the k-ω SST turbulence model is due to incorrect non-equilibrium boundary layer modelling. Using a Design of Experiments based approach, it is found that by reducing β_1, a turbulence model constant affecting near wall modelling of ω, significant improvements in non-equilibrium modelling can be achieved. These improvements are found to be robust with respect to blade loading style and Mach number (within the subsonic regime), and also lead to improvements in blade performance parameter prediction. These results are used to investigate the impact of non-equilibrium boundary layer effects on compressor cascade performance.

It is found that vortex shedding trailing edge flows cannot be satisfactorily modelled using conventional steady RANS simulations. Using the Selective Frequency Damping technique, two reasons for this are found: errors arising from the Boussinesq approximation, and the fact that the correct time-averaged flow is an unstable solution of the RANS equations, leading to convergence issues. It is shown that, despite not being able to stabilise the RANS equations on the correct mean flow, Reynolds stresses calculated from the high fidelity simulation are the optimum RANS equation source terms. By modelling the vortex shedding process while simulating turbulence, Unsteady RANS simulations are shown to qualitatively predict all the main features of the fundamental trailing edge flow.