径流、斜流式压缩机百参数智能优化设计理论方法及工程应用

The synergistic growth of high-performance computing capabilities and the maturation of computational fluid dynamics (CFD) technology have fundamentally reshaped the landscape of turbomachinery design. Specifically, in the development of radial and mixed-flow compressors design, surrogate model-based intelligent optimization methods are gaining significant prominence. These data-driven approaches, which minimize reliance on traditional empirical knowledge and designer intuition, offer the compelling advantages of substantially shortening research and development cycles and reducing associated costs. Nevertheless, conventional surrogate-based optimization paradigms are frequently hampered by significant inherent limitations that curtail their effectiveness in complex, real-world scenarios. These limitations typically include: a restricted number of design control parameters, which prevents the detailed capture of intricate 3D geometric features; surrogate models that provide insufficient or localized predictive information, often failing to accurately represent performance across diverse operating conditions; and suboptimal utilization of simulation sample data, leading to inefficient exploration of the vast design space. Consequently, these traditional methods struggle to address advanced engineering challenges characterized by high degrees of design freedom and a complex web of competing design objectives. To bridge this gap and unlock the next level of performance in radial and mixed-flow impellers, our research group has dedicated recent efforts to proposing and systematically developing a novel theoretical and methodological framework, which we term the “hundred-parameter intelligent optimization”. This advanced framework radically expands the dimensionality of the design space from the conventional tens of parameters to over one hundred. This expansion facilitates, for the first time, comprehensive and direct control over both the impeller’s intricate internal passage geometry and its aerodynamic performance across the entire operating range. This achievement represents a pivotal advancement toward creating optimization methods that are more general, flexible, and truly intelligent. This paper provides a comprehensive overview of our group’s recent contributions to this high-dimensional optimization paradigm, focusing on innovations across three foundational pillars: parameterization, surrogate modeling, and sampling. Key methodological breakthroughs include: (1) a full-three-dimensional parametric geometry definition coupled with an automated mesh topology generation method, enabling high-throughput and hands-off CFD analysis; (2) a robust surrogate model, called full range prediction model (FRPM), which is capable of accurately predicting the full operating range performance, from choke to near-stall conditions; (3) a model training framework based on flow information, which leverages the extraction of data features directly from the CFD flow fields, thereby enriching the model with physical insights and enhancing its fidelity; and (4) adaptive batch sampling strategy (ABSS) that intelligently queries the design space to maximize information gain and accelerate convergence. To substantiate the efficacy of this framework, practical application cases and experimental validation results are presented for a hundred-parameter optimization of radial and mixed flow impellers. The findings conclusively demonstrate that our high-dimensional optimization system can effectively and efficiently resolve complex, real-world engineering problems involving multidisciplinary constraints, multi-objective trade-offs, and full-range performance requirements, all within an acceptable computational budget. This work establishes the hundred-parameter approach as a critical and promising direction for the future of intelligent design optimization in the fluid machinery domain.