Flow Pattern Identification of Two-Phase Flow Using Neural Network and Empirical Mode Decomposition

Flow pattern identification of two-phase flow involves two key processes: feature extraction and classification. We attempt to combine Empirical Mode Decomposition (EMD) and backpropagation (BP) neural network to solve this identification problem. Differential pressure signal of two-phase flow, which is representatively non-stationary and multi-component signal, contains much information about flow. EMD is applied to differential pressure signal to obtain frequency components with different scales. The normalized energy of frequency components is extracted as features. Five flow patterns such as bubble flow, plug flow, stratified flow, slug flow and annular flow are investigated using BP neural network. The experimental results of oil-gas two-phase flow indicate that the proposed method is effective for classifying flow pattern.