Two-phase flow regime characterization via evolutionary features of interfacial structure in rod bundles

Accurate identification of flow regimes is essential for selecting appropriate constitutive relations, improving transfer and reaction performance, and ensuring safe operational limits in engineering systems. Previous studies have investigated through three main routes: geometry-based criteria, signal-feature analysis, and mechanism-based models. Although machine learning has shown potential, its effectiveness is closely tied to the choice of input variables. Existing studies commonly used inputs such as flow images, pressure fluctuations, and void fraction measurements. In two-phase flow systems, since the mass, momentum, and energy transfer are fundamentally controlled by the interface, parameters that characterize interfacial structure can serve as highly informative signatures of regime change. We thus establish an alternative identification strategy that derives representative interfacial structure metrics with the aid of the drift-flux model, characterizes their trends with local regression, and delineates regime boundaries using Otsu's thresholding methods. The identified transition boundaries under different rod bundle geometries and pressure conditions are subsequently analyzed and compared with results from four alternative methods: the Liu–Hibiki model, the TRACE model, the wavelet analysis approach, and the neural network approach. © 2026 Elsevier Ltd.