基于图像处理的回油管油气两相流型辨识模型

To explore the phase distribution characteristics of the oil-gas two-phase flow in scavenge pipe of aero-engine lubrication system，and achieve accurate flow pattern identification，a flow pattern identification model with image processing was proposed based on the typical flow patterns images obtained from a horizontal pipe under the working condition of aero-engine. Four typical flow patterns emerged in this experiment: slug，stratified，wavy，and annular flow. By image processing technologies such as bilateral filter，binarization，and wavelet decomposition，feature parameters were extracted from the images and used as input to the model. The identification model based on Elman Recurrent Neural Networks was established through training and verification，and it can successfully identify four different flow patterns. The identification accuracy of the model was 93.06%，and the robustness index macro-F1 was 97.60% on the verification set.