Turbomachinery flow simulations with SU2: a numerical study

Abstract

Analysing and developing turbomachinery at off-design conditions requires the usage of robust and efficient Computational Fluid Dynamics (CFD) solvers. With the introduction of the computer, many advancements have been made in the field of numerical methods involving these flow problems. These numerical methods are used in order to solve these complex flow structures that are formed due to the interaction between the rotating machinery and the fluid. With the emergence of new design paradigms using computer-aided optimisation, novel solver methods are required which are able to deal with the increase in computational cost and the convergence difficulties following off-design conditions. One of the CFD solvers that is used for turbomachinery design is the open-source software SU2. SU2 is currently developed partially at the Delft University of Technology, where an increase in solver performance with respect to turbomachinery aerodynamics is greatly desired. The current work provides a numerical study of SU2's current performance with respect to turbomachinery analysis, as this is currently unknown.

The current performance of SU2 is to be analysed using steady RANS turbomachinery simulations. The research conducted by Xu et al. \cite{Xu2020} will be used as reference data in order to validate SU2's performance together with data obtained from the CFD solvers CFX and Numeca. The research conducted by Xu et al. resulted in the development of their NUTSCFD solver, which showed strong performance with respect to turbomachinery analysis. Four test cases are set up and used in order to analyse SU2. The test cases that are considered for the numerical study include: the NACA 0012 airfoil, the LS89 turbine cascade, the MTU centrifugal compressor and the 1.5 stage ETH turbine. The first three test cases are validated using the results obtained by the NUTSCFD solver, where the ETH turbine is validated using CFX and Numeca. Following these test cases, SU2's performance is analysed using the residual behaviour. Xu et al. dedicate their performance increase to be the result of the Newton-Krylov method that is implemented in their NUTSCFD solver. SU2 also includes a Newton-Krylov solver, but its performance with respect to turbomachinery is also unknown. A performance assessment with respect to SU2's Newton-Krylov method involving turbomachinery analysis is therefore conducted as well.

The results obtained using the NACA 0012 and LS89 test cases show a discrepancy in solver performance involving SU2. With respect to SU2's Newton-Krylov solver, the NACA 0012 test case shows a reduction in non-linear iterations where this is not found for the LS89 test case. Both results showed however a large difference in performance when compared to the NUTSCFD solver, where SU2's standard solver was also unable to match NUTSCFD. The results obtained following these test cases have led to the development of the MTU test case, where the MTU test case was to be used in order to provide a more accurate comparison between SU2 and NUTSCFD. Instead, SU2 was unable to run the MTU test case, where the solver showed stalling behaviour...