Research on flow pattern identification model of oil-gas two-phase flow in scavenge pipe

To understand the variations in pressure drop and heat transfer characteristics within the scavenge pipe of aero-engines, studying and attempting to discriminate the flow patterns of two-phase flow inside the scavenge pipe is of great significance. To achieve this, this paper establishes a flow pattern identification model. High-speed photography was utilized to capture images of four distinct flow patterns inside the scavenge pipe under typical operating conditions. Through image preprocessing, feature extraction, and Relief-F feature selection, the primary texture and shape features are obtained as inputs for the identification model. Four machine learning methods, namely unsupervised learning K-means, supervised learning backpropagation neural network (BP), radial basis function neural network (RBF), and support vector machine (SVM), are selected for flow pattern identification. For the optimization of hyperparameters in supervised learning methods, this paper utilizes the particle swarm optimization (PSO) algorithm. Consequently, PSO-BP, PSO-RBF, and PSO-SVM models are further established. After inputting the two types of features, texture and shape, into the mentioned models, a comparison of the classification accuracy and generalization ability of the four models is conducted. The results indicate that, for the flow pattern identification problem of oil-air two-phase flow inside the scavenge pipe studied in this paper, the most suitable identification model is the PSO-SVM model.