Gas Volume Fraction Measurement for Gas-Liquid Two-Phase Flow Based on Dual CNN-Transformer Mixture Neural Network

Accurate gas volume fraction (GVF) measurement in gas-liquid two-phase flow remains a key challenge in industrial process monitoring and control. In order to address this, a deep learning-based soft sensing method is proposed, combining an array conductance sensor with a novel dual CNN-transformer mixture neural network (DCTMNN). The model incorporates two core modules: CTM1, designed to extract and fuse both local and global features, and CTM2, which further enhances the representation of extracted features. A long short-term memory (LSTM) network is, furthermore, integrated with an attention mechanism to improve temporal sequence modeling and adaptively focus on critical features. Experimental data collected from a vertical pipeline setup validate the effectiveness of the approach. The proposed DCTMNN achieves a GVF measurement accuracy of 97.57% within an error margin of ±0.5%, surpassing traditional CNN-based and transformer-based models. These results suggest that DCTMNN provides a robust and accurate solution for GVF estimation and holds strong potential for deployment in practical multiphase flow systems.