International Society for Industrial Process Tomography

Flow Regime Identification with Single Plane ECT
Using Deep Learning

Ru Yan¹, Saba Mylvaganma^1*

¹ Department of Electrical Engineering, IT and Cybernetics, Faculty of Technology, Natural Sciences and Maritime Sciences, University of South-Eastern Norway, 3901 Porsgrunn, Norway

^*Email: saba.mylvaganam@usn.no

ABSTRACT

In multiphase flow studies, different flow regimes occur and there has been strong interest in non-intrusive real time sensing methods for identifying these. In this paper, a single plane 12-electrode electrical capacitance tomography system delivering capacitance data at 500 frames per second, with 66 independent capacitances per frame is used in a multiphase flow loop using different combinations of air and water mass flow rates. Various flow regimes were generated in a measurement campaign leading to 84 different time series of measurement data with the corresponding high-speed camera images. For the analysis of data for identification of flow regimes, data from the 12-electrode electrical capacitance tomography system was logged in for usage in neural network based data analytics.When these capacitance values from 100 frames captured in 200ms are used for training a neural network, there will be 6600 inputs, making the system unwieldy for real time operation. We resort to some statistical pre-processing of the measurement data to obtain representative values of the 66 measurements per frame, thus reducing the size of inputs. Measurement data are then fed to a Deep Belief Network consisting of two Restricted Boltzmann Machines, delivering 5 outputs consisting of the identified flow regimes, viz. plug, slug, annular, stratified and wavy flows.

Keywords electrical capacitance tomometry (ECTm), flow regime, deep learning, Deep Belief Network (DBN), Restricted Boltzmann Machines (RBN), Model Free Adaptive Control

Industrial ApplicationOil & Gas, Multiphase flow studies, CFD