Flow Regime Identification in Boiling Two-Phase Flow in a Vertical Annulus

Abstract

This work describes the application of an artificial neural network to process the signals measured by local conductivity probes and classify them into their corresponding global flow regimes. Experiments were performed in boiling upward two-phase flow in a vertical annulus. The inner and outer diameters of the annulus were 19.1 mm and 38.1 mm, respectively. The hydraulic diameter of the flow channel, DH, was 19.0 mm and the total length is 4.477 m. The test section was composed of an injection port and five instrumentation ports, the first three were in the heated section (z/DH = 52, 108 and 149 where z represents the axial position) and the upper ones in the unheated sections (z/DH = 189 and 230). Conductivity measurements were performed in nine radial positions for each of the five ports in order to measure the bubble chord length distribution for each flow condition. The measured experiment matrix comprised test cases at different inlet pressure, ranging from 200 kPa up to 950 kPa. A total number of 42 different flow conditions with superficial liquid velocities from 0.23 m/s to 2.5 m/s and superficial gas velocities from 0.002 m/s to 1.7 m/s and heat flux from 55 kW/m2 to 247 kW/m2 were measured in the five axial ports. The flow regime indicator has been chosen to be statistical parameters from the cumulative probability distribution function of the bubble chord length signals from the conductivity probes. Self-organized neural networks (SONN) have been used as the mapping system. The flow regime has been classified into three categories: bubbly, cap-slug and churn. A SONN has been first developed to map the local flow regime (LFR) of each radial position. The obtained LFR information, conveniently weighted with their corresponding significant area, was used to provide the global flow regime (GFR) classification. These final GFR classifications were then compared with different flow regime transition models.