A data-driven method for constructing stator cascade datasets based on flow field separation criteria in axial compressors

Data-driven approaches are increasingly popular, where reliable datasets being crucial for accurate flow field modeling. However, most existing experimental design methods rely on mathematical approaches that lack physical interpretability. In compressor aerodynamics, severe flow separation represents the physical limits of the dataset. Flow fields with significant separation not only lack computational accuracy but also have limited engineering applicability. Therefore, filtering such samples through flow field classification is essential. The differentiator of this paper lies in applying a data-driven method to classify flow fields during dataset construction. The dataset, containing 889 numerical samples, was labeled into three categories: two-dimensional separation, three-dimensional open corner separation, and no obvious separation. The support vector classification (SVC) algorithm was applied to analyze characteristics and classify flow field structures in compressor stator cascades. The feature analysis reveals that the axial velocity density ratio (AVDR) plays a key role in flow field classification. Validation results demonstrate that the model can rapidly detect severe separation samples, enabling quick qualification of blade profile geometries and operating conditions. This capability significantly reduces computational resource demands, confirming the model’s suitability for both dataset generation and preliminary blade design phases. Furthermore, input features with clear physical relevance, such as secondary flow intensity, enhance model accuracy.