TL;DR
This paper demonstrates the feasibility of using high-fidelity quasi-Direct Numerical Simulation (qDNS) with the Nektar++ solver to capture complex unsteady flow phenomena, like separation bubbles and turbulent transition, on a compressor blade. The study reveals an optimal level of inflow turbulence that minimizes losses and provides valuable insights into the computational cost, suggesting qDNS is becoming a viable tool for industrial turbomachinery design.
What Problem Does This Paper Address?
Traditional turbomachinery design relies heavily on Reynolds-Averaged Navier-Stokes (RANS) solvers. While cost-effective, RANS struggles to accurately capture unsteady phenomena such as flow separation, stall, and tip-gap interactions, which significantly impact component performance and efficiency. These unsteady events are crucial for improving fuel burn and reliability in modern engines. This paper addresses the need for more accurate, high-fidelity simulation methods that can resolve these complex flow physics, bridging the gap between simplified models and the computationally prohibitive full Direct Numerical Simulation (DNS).
Methods
The authors employ the Nektar++ spectral/hp element solver, a high-order method known for its accuracy in resolving fine flow structures. For this study, an incompressible formulation was used, coupled with a velocity-correction scheme and Taylor-Hood discretisation to avoid checkerboarding effects. A fully quadrilateral mesh was generated for the 2D aerofoil, and a Fourier expansion was used for spanwise extrusion to improve computational efficiency. The simulation achieved quasi-DNS resolution, meaning the grid and time-stepping were fine enough to capture the essential turbulent scales without the full cost of DNS. Synthetic turbulence generation was used to model inflow disturbances.
Datasets
The study uses experimental data for comparison, specifically pressure distribution measurements from a linear cascade compressor configuration. The experimental data is used to validate the simulation results and tune parameters like the angle of incidence. No new datasets were introduced.
Key Insights and Ideas
- Double Separation: At clean inflow conditions, Kelvin-Helmholtz instabilities lead to separation bubbles on both the suction and pressure sides of the blade, with the pressure-side separation not being observed in the original experiments.
- Turbulence as a "Cleaner": Incoming turbulence gradually removes these separation bubbles, but at different rates on each side of the blade.
- Optimal Turbulence Intensity: There exists an optimal level of inflow turbulence intensity that minimizes the momentum thickness, a key indicator of losses. This suggests that controlled turbulence can be beneficial for performance.
- Spanwise Extrusion Cost: The spanwise extent of the computational domain (extrusion length) is a significant factor in modeling costs. The study confirms that a 10% chord extrusion is sufficient to capture the essential physics, balancing accuracy with computational expense.
- qDNS Feasibility: The paper provides concrete data on wall-clock time and computational cost, demonstrating that quasi-DNS resolution is becoming economically viable for specific, critical flow conditions in industrial settings.
Key Results
- The simulation successfully reproduced experimental pressure loading data after adjusting the angle of incidence to 53.5°.
- Kelvin-Helmholtz rolls were observed on both blade sides, initiating transition to turbulence.
- The study identified an optimal turbulence intensity (around 4-5%) that minimized suction side momentum thickness, indicating reduced losses.
- A spanwise extrusion of 10% of the true chord was found to be sufficient for accurate simulation of the transition mechanism.
- Computational costs for qDNS simulations were presented, showing runtimes between 1.25 and 2.5 days on a supercomputing service, which is deemed cost-effective compared to experimental testing.
Limitations and Future Work
The authors acknowledge that while qDNS is becoming more feasible, it's still too expensive to model all components and operating conditions of an aircraft engine with this fidelity. They suggest that qDNS and LES can be used to augment RANS results and experimental data to gain a deeper understanding of critical flow physics. Future work could involve extending this methodology to more complex geometries and a wider range of operating conditions. The paper also notes that while Görtler vortices were hinted at by criteria, they were not directly observed in the vorticity analysis.
Summary updated