Key points are not available for this paper at this time.
Rotor–stator interaction is an inherently unsteady phenomenon in turbomachinery that significantly influences the performance of turbomachinery. Accurate prediction of the unsteady turbomachinery rotor–stator interaction flow remains a great challenge considering computational cost. In the Reynolds-averaged Navier–Stokes framework, the harmonic balance (HB) method emerges as a potential reduced-order modeling technique, offering significant computational savings over traditional unsteady methods, and revealing unsteady flow characteristics that are elusive to the steady mixing-plane method. In this study, a graphical processing unit (GPU)-based solver utilizing the finite volume method is developed to accelerate the computation of the HB method compared to the traditional central processing unit (CPU)-based solver. An implicit data-parallel block-Jacobi lower-upper relaxation (DP-BJ-LUR) method is first proposed to better fit the distinct parallel architecture of GPU. The HB method with different harmonics, as well as unsteady time marching method, is conducted to evaluate the accuracy and acceleration for convergence of the proposed method by a quasi-three-dimensional radial slice case and a full three-dimensional case for National Aeronautics and Space Administration (NASA) Stage 35 compressor. Acceleration performance of GPU-based solver, impact of relaxation steps on the DP-BJ-LUR method, and numerical accuracy are compared in detail. A maximum speedup of 102 times with 1 harmonic and 90 times with 12 harmonics is achieved by the GPU-based solver on a single NVIDIA Ray Tracing Texel eXtreme 3080Ti GPU compared with the CPU-based solver on a single CPU core of Intel® Xeon® Platinum 9242.
Yong et al. (Sun,) studied this question.