Identification of Gas-Liquid Flow Regimes From a Space-Frequency Representation by Use of an Impedance Probe and a Neural Network

The identification of two-phase flow patterns has been widely studied, and the diagnostic procedures are traditionally based on statistical or spectral signal analysis, while the spatial information related with the geometrical topology of the phase distribution in the pipe is never taken into account. The aim of this study is to demonstrate how the exploitation of both spectral and spatial information leads to an unambiguous identification of the flow patterns. Experiments are performed on a 30 meters long horizontal air-water loop. By simultaneously analyzing the power spectral density of the signals delivered by a multi-electrode impedance sensor, we obtain a space-frequency representation which exhibits particular features of the different flow regimes. They can be characterized by a set of 3 scalar parameters, quantifying respectively the localization in space, in frequency and the shape of the spectral content. The final demonstration of this space-frequency characterization is provided by the use of a multi-layer neural network, trained on a 80 tests database. This net exhibits a successful identification rate above 80% when used in blind real-time tests.