Development of an intelligent dual-beam gamma densitometer for real-time recognition of two-phase flow regime in horizontal pipe

Abstract

This work includes the hardware and software development of a high sampling rate dual-beam gamma densitometer for real-time two-phase flow monitoring. The designed densitometer consists of two fast SiPM-based gamma detectors and a custom embedded system for digital signal processing and running flow regime recognition neural network. An exclusive LSTM deep neural network has been developed to classify flow regimes into six classes including stratified, annular, bubbly, wavy, slug, and plug. Therefore, the proposed design is superior to the traditional approaches by (i), unifying signal processing and flow identification units into a single cost-effective embedded system; (ii), reducing the need for time-consuming signal preprocessing steps from conventional to feed-forward neural networks, and (iii), real-time identification of the flow due to the high refresh rate of the data processing chain. The optimal neural network model has been obtained through comprehensive sensitivity analyses to have the best interpolation and generalization. The experimental evaluations showed that the optimized model recognizes the flow regime based on 6 s of gamma data sequence with an accuracy of ∼92%–97% depending on the flow regime in the pipeline. Furthermore, the real-time experiment proved the successful performance of the densitometer to trace the flow regime across the two-phase flow map.