This paper presents a comprehensive validation and verification study of turbomachinery Reynolds-averaged Navier-Stokes flow solvers on the transonic axial compressor TUDa-GLR-OpenStage. Two commercial solvers namely Ansys CFX and Numeca FineTurbo are adopted to provide the benchmark solutions, which can be used for verification of other RANS solvers in the future. Based on these solvers, five sets of grids, two advection schemes (i.e., central difference and second-order upwind), four turbulence models (i.e., SA, SA-RC, SST and EARSM) and two rotor-stator interface models (i.e., mixing plane and sliding plane) are investigated to quantify their effects on predicting the performance and the flow field of the compressor stage. Results show that the choices of grid density and turbulence model are most sensitive to the prediction, leading to 5% and 7% variation in compressor performance characteristics, respectively. Regarding the choice of grid density, a method to estimate the grid discretization error is demonstrated, which is transferrable to other cases. Regarding the choice of turbulence model, the EARSM model is found overall most accurate among the investigated models, and the limitations and deficiencies of the rest models are discussed in detail based on the analysis of the mean flow fields and the eddy viscosity fields. The grids and the major CFD results presented in this work are open-accessed to the community for further research. The results and discussions presented in this paper provide a useful reference for future practices of RANS simulations for compressors.
The influence of the moving endwall on the behavior of the tip leakage flow in compressors holds substantial significance. This work investigates the unsteadiness of the compressor's tip leakage flow under the impact of the moving endwall, employing both experimental methods and unsteady Reynolds-averaged Navier–Stokes (URANS) simulations. First, a numerical decomposition technique is proposed to isolate the tip leakage flow within a controlled diffusion airfoil (CDA) cascade. Subsequently, URANS simulations based on the CDA cascade are conducted at various incidence angles (−6° to +7°), tip clearances (clearance to chord: 0.83% to 3.33%), and flow coefficients (0.4 to 0.6), and additional URANS simulations based on the Virginia Tech cascade are also performed. The results indicate that the moving endwall induces a greater inclination of the tip leakage flow in the pitchwise direction, allowing it to impinge upon the pressure side of the adjacent blade. The impingement, if sufficiently intense, leads to the periodic generation of unsteady tip leakage flow by altering the blade loading. Moreover, the degree of unsteadiness diminishes significantly as the endwall speed and clearance height decrease. Additionally, it is observed that the breakdown of the tip leakage vortex does not serve as the root cause of the periodic unsteadiness. These findings underscore the critical role played by the moving endwall in the unsteadiness of the compressor's tip leakage flow.
The moving endwall has a great influence on the development and stability of axial compressor leakage flow. This paper presents a novel experimental facility with a high-speed moving endwall for studying axial compressor leakage flow. The uniqueness of the design concept is that using a large disk simulates the high-speed moving endwall. When R/Cx = 16, theoretical analysis shows that the maximum linear velocity difference is about 2.5% while the maximum axial velocity difference of the mid-three passages is less than 5%. Single-passage simulations show that the disk radius of R/Cx = 16 can achieve an acceptable accuracy in terms of static pressure, total pressure, and density flow. Seven-passage simulations confirm that the mid-three passages have small errors from the axial velocity difference. Subsequently, preliminary experimental results obtained from the experimental facility are presented. The results reveal that the moving endwall significantly changes the distributions of the total pressure loss and static pressure coefficient. The relative difference in the averaged total pressure loss between the experiment and CFD is 11.33% and 7.69% for the static and moving endwall, respectively. It is expected that the experimental facility will make more useful contributions to the understanding of axial compressor leakage flow in the future.
Deep transfer learning is frequently employed to address the challenges arising from limited or hard-to-obtain training data in the target domain, but its application in axial compressors has been scarcely explored thus far. In this paper, a multi-objective optimization framework of a transonic rotor is established using deep transfer learning. This framework first pre-trains deep neural networks based on the peak efficiency condition of 100% design speed and then fine-tunes the networks to predict the performance of off-design conditions based on the small training dataset. Finally, the design optimization of the transonic rotor is carried out through non-dominated sorting genetic algorithm II. Compared to neural networks that are trained directly, transfer learning models can achieve higher prediction accuracy, particularly in scenarios with small training datasets. This is because the pre-trained weights can offer a better initial state for transfer learning models. Moreover, transfer learning models can use fewer samples to obtain an approximate Pareto front, making the optimized rotor increase the isentropic efficiency at both peak efficiency and high loading conditions. The efficiency improvement of the optimized rotor is attributed to the reduction of the loss associated with the tip leakage flow by adjusting the tip loading distribution. Overall, this study fully demonstrates the effectiveness of transfer learning in predicting compressor performance, which provides a promising approach to solving high-cost compressor design problems.