Analysis and identification of gas–liquid two-phase flow pattern based on variational mode decomposition and mutual information complex network

Abstract The identification of gas–liquid two-phase flow patterns is crucial to fluid dynamics research and industrial applications. This study proposes a novel flow pattern identification framework that integrates variational mode decomposition (VMD), complex network analysis, and deep learning. First, the four-channel conductance signals are decomposed into intrinsic mode functions (IMFs) using VMD to capture subtle flow characteristics. Then, mutual information (MI) is used to quantify the nonlinear correlation between each pair of IMFs. Subsequently, a weighted complex network is constructed by assigning each IMF as a node and quantifying the MI between IMFs pairs as edge weights, thereby uncovering the underlying nonlinear interactions within the gas–liquid two-phase flow. In addition, multiple topological indicators, including the total length of the maximum spanning tree and the maximum participation coefficient, are extracted and analyzed to investigate the structural characteristics of different flow regimes. Finally, the topology of each complex network is represented as an adjacency matrix, which serves as input to deep learning frameworks for gas–liquid flow pattern identification. Experimental results on a self-constructed dataset demonstrate that the proposed method can effectively identify different gas–liquid two-phase flow patterns, achieving an accuracy of 95.56 %. This research establishes a complex network based on VMD and MI, effectively eliminates noise interference in the signals, extracts key structural features, and improves the accuracy and interpretability of gas–liquid two-phase flow pattern identification.