Search

The design of industrial feed line sections and the selection of appropriate inlet devices requires knowledge about the two‐phase flow morphology and phase fractions in order to reduce costly overdesign. Since experimental data for two‐phase pipe flows at industrial scale are hardly available, current prediction methods are subject to high unertainities. In this study, horizontal two‐phase flow morphologies for different industrial feed line sections were studied using wire‐mesh sensors at multiple locations along the pipe. The obtained phase fraction data are compared with correlations. [ABSTRACT FROM AUTHOR]

• Fuzzy clustering was used for developing a classifier for flow pattern identification. • Powerful characteristics were derived from tomographic phase fraction data. • A novel technique for visualizing fuzzy flow regime affiliations is proposed. • Explicit and transitional patterns are identified satisfactorily. • Performance at pseudo-dynamic operation is demonstrated. Online monitoring of two-phase flow patterns is an essential need in various chemical engineering applications, since the reliability of prediction methods is limited. Therefore, the present study aims at developing a practically applicable algorithm for identifying flow patterns in horizontal gas-liquid flows on the basis of wire-mesh sensor data. Experiments were conducted in a 50 mm i. d. pipe over a wide range of superficial velocities of an air-water mixture. Characteristic features involving the influence of gravity and the spatio-temporal behavior of the flows were derived from tomographic phase fraction data and used as input for fuzzy clustering. Three differently determined sets of cluster centers are compared against a reference classification by human specialist through reclassifying the measurements with the aid of defuzzyfication and, alternatively, by means of a novel visualization technique, that retains the fuzziness of the results. With respect to the latter one, best agreement is reached with cluster centers from fuzzy c-means clustering using all recorded measurements. As a special emphasis is put to the identification of transitional flow patterns, the performance of the algorithm at pseudo-dynamic operation is demonstrated, finally. [ABSTRACT FROM AUTHOR]