As a degree of freedom in three-dimensional blade design of axial compressors, the sweep technique significantly affects aerodynamic performance of axial compressors. In this paper, the influence of backward sweep rotor configurations on aerodynamic performance of a 1.5-stage highly loaded axial compressor is studied by numerical simulation. The aim of this work is to improve

Strong three-dimensional effects within low Reynolds number compressor cascades invalidate the assumption of turbulence isotropy, significantly reducing the accuracy of solving the Navier–Stokes equations using Reynolds-averaged Navier-Stokes (RANS) methods. Although compressor experiments can provide more reliable results than RANS under low Reynolds number conditions, they only offer flow field data at limited observation, making it difficult to reconstruct the complete flow field. Large eddy simulation (LES), which do not rely on empirical models, can naturally capture flow separation near the boundary layer under low Reynolds numbers. In this study, representative high-fidelity data points, potentially obtainable from experimental measurements, are sampled from LES flow fields. This paper proposes a Multi- Fidelity Deep Neural Network (MFDNN) architecture based on multi-source data fusion strategy for achieving flow field reconstruction of low Reynolds number compressors. The MFDNN model consists of two components: a low-fidelity model and a high-fidelity model. A dual-modal training strategy is introduced to couple the training of the low-fidelity and high-fidelity models via the modal weighting coefficient. The MFDNN model leverages the reconstructed low-fidelity flow field as a physical constraint for reconstructing the high-fidelity flow field, ensuring that the reconstructed field remains smooth and physically consistent even with very sparse high-fidelity data. The MFDNN model is capable of accurately reconstructing compressor cascade flow fields under various low Reynolds number conditions, with significantly improved accuracy in key entropy-producing regions, such as the flow separation region on the suction surface and the wake region. Even in the absence of sufficient boundary conditions and sampling points, the flow field reconstructed by the MFDNN model maintains reliable accuracy, exhibiting excellent performance in solving inverse problems of turbomachinery.