Boosting Accuracy and Solution Time in Turbomachinery with Fidelity LES

For computational fluid dynamic (CFD) simulation of turbomachinery design, the most widely used numerical tool remains computationally efficient low-fidelity methods based on the Reynolds-averaged Navier-Stokes (RANS) equation. While high-fidelity methods like direct numerical simulation (DNS) and large eddy simulations (LES) provide more accurate results, their use is often constrained by high computational demands and resource intensity.

Nevertheless, there is a growing need for unsteady simulations in turbomachinery applications. Effectively capturing the three-dimensional unsteady flow characteristics is crucial, as periodic domains do not sufficiently represent the interactions and relative movements among the turbomachinery components. Consequently, there is a significant demand for a cost-efficient and high-fidelity LES tool tailored for turbomachinery simulations. This is where Cadence Fidelity LES Solver, formerly Cascade CharLES, emerges as a transformative solution.

Fidelity LES for Turbomachinery Simulations

The Fidelity LES Solver is the first of its kind in the industry, offering high-fidelity CFD capabilities that broaden the practical application of LES across various engineering fields. It addresses the most demanding challenges of fluid dynamics and provides precise solutions for complex CFD problems in areas such as aeroacoustics, aerodynamics, combustion, heat transfer, and multiphase flows.

The three numerical and computational developments that facilitate high-fidelity turbomachinery simulations with Fidelity LES are

1. Advances in wall-modeled LES (WMLES) techniques
2. A highly efficient moving mesh solver that is based on Voronoi diagrams
3. High throughput computing using GPUs

Voronoi Diagram-Based Moving Mesh Solver

For turbomachinery simulations, meshes for stationary and rotating parts are generated separately and remain unchanged during the calculation. An innovative technology is developed for part interface treatment. This method cuts the interface faces based on 2D Voronoi diagrams at every time step and allows information exchange between the stationary and rotating parts through numerical flux. This approach avoids volume-based recutting of the Voronoi diagram and any tree-based search on GPU and thus significantly improves computational efficiency.  Based on this approach, the numerical scheme for the moving solver is conservative, compact, and fast to compute, offering significant performance and accuracy advantages over traditional approaches.

This technique enables a flexible division of the computation domain and is showcased through two examples below: the first employing a surface of revolution through the tip gap and the second utilizing planar interfaces that slice through both the hub and shroud. The adoption of planar interfaces overcomes challenges linked to the tip gap, permitting the initiation of simulations from a coarse mesh without the limitations on mesh or time step size.

Two types of stationary-moving part interface (indicated by the yellow dashed lines) for turbomachinery flow simulations: (a) a surface of revolution that goes through the tip gap and (b) planes that cut the shroud  (Wang, Bose, & Ivey, 2024)

Case Study: NASA 67 Fan Stage

The NASA 67 is a transonic fan stage with 22 rotor blades and 34 stator blades, operating at 16,000 RPM. In this study, full-wheel wall-modeled large-eddy simulations are performed for both the rotor-only and the entire stage configurations. The rotor-only configuration is also simulated using a single blade in a moving reference frame, and the result is used to verify the moving mesh approach. The WMLES results are validated against NASA’s experimental measurements.

The figure below shows the computational domain and its division into stationary and rotating parts. Planar interfaces are employed in this case, and the moving part is highlighted in red.

NASA 67 Fan Stage (Wang, Bose, & Ivey, 2024)

Two types of computational mesh are employed in this study: a purely Hexagonal Close Packed (HCP) mesh with uniform refinement and a hybrid HCP mesh with anisotropic boundary layers (called “strand” mesh).  The details of the computational meshes are listed in the table below

Computational Meshes (Wang, Bose, & Ivey, 2024)

Analysis of meshes for the stage simulations (M4, M5, and M6) revealed that mesh M4, made entirely of HCP with 58 million elements, closely matches experimental data. Mesh M5, blending HCP and strand meshes, further improved the predictions due to finer surface resolutions. Mesh M6 showed minor enhancements, suggesting further refinement in the blade passage has little impact on the predictions when the surface resolution is sufficient. Temperature ratio and adiabatic efficiency closely align with experimental results, with minor discrepancies approaching stall conditions.

Total temperature ratio curves for the rotor-only configuration with 0.5 mm tip clearance simulated on different meshes (Wang, Bose, & Ivey, 2024)

Performance benchmarking on CPUs vs GPUs reveals the efficiency improvements of GPUs over CPUs, significantly reducing simulation time from weeks to within a day. The study underscores GPUs' role in accelerating turbomachinery flow simulations, employing HCP and anisotropic boundary meshes to achieve results closely aligned with experimental data, proving GPU-accelerated flow solvers as a time-efficient solution for turbomachinery applications.

600 CPU core equivalence is observed for the GPU solver (Wang, Bose, & Ivey, 2024)

The adoption of high-fidelity methods is driving significant advances in CFD for turbomachinery design. Fidelity LES addresses these industry demands, combining wall-modeled LES, Voronoi diagram-based meshing, and moving mesh techniques crucial for achieving more accurate and efficient simulations. The NASA 67 Fan Stage case study highlights how these technologies can optimize computational processes and align closely with real-world data. Fidelity LES, leveraging modern computing capabilities, is revolutionizing CFD applications, promising greater precision and efficiency in turbomachinery design.

Reference
Wang, K., Bose, S., & Ivey, C. (2024). GPU-Accelerated Full-Wheel Large-Eddy Simulations of a Transonic Fan Stage. Proceedings of ASME Turbo Expo 2024. London, UK

For more information on the use of Fidelity LES for the NASA 67 Fan Stage simulations, watch the Cadence TECHTALK delivered by Kan Wang, senior principal software engineer, Cadence Fidelity R&D team.