Black-box real-time identification of sub-regime of gas-liquid flow using Ultrasound Doppler Velocimetry with deep learning

Abstract

Gas-liquid flow has a strong relationship with energy transfer, production, and transportation. A black-box method based on velocity information and deep learning was proposed to identify the sub-regimes of gas-liquid flow in real time, including two parts—data acquisition and establishment of the identification model. The Ultrasound Doppler Velocimetry (UDV) method was employed to acquire the velocity information of gas-liquid flow in a horizontal pipe non-intrusively. Seven different sub-regimes are defined to perform the identification task. Poor identification accuracy was obtained with raw data; therefore, denoising was added to improve the identification accuracy. The results show that the proposed method is feasible and effective, and the Current model can achieve high accuracy similar to existing models while having the fastest identification speed. In addition, the data from untrained flow conditions were tested, and all convolutional neural network (CNN) models achieved identification accuracy higher than 91.5%. The proposed method helps in identifying two-phase flow, and its accurate and straightforward characteristics indicate broad application potential.