Multiphase flow pattern recognition in pipeline-riser system by statistical feature clustering of pressure fluctuations

In the offshore petroleum industry, it is important to recognize the severe slugging in multiphase flow patterns in oil and gas transportation through a pipeline-riser system. After analyzing the inadaptability of the existing methods, we proposed a simple and practical method of measuring the multiphase flow patterns based on pressure fluctuations. Our recognition was carried out in three steps. First, the outlet pressure signals were selected and processed because of their accessibility to practical applications in oil fields. Second, statistical and principal component analysis were performed on the sampled signals to obtain the clear interrelations between the signals and flow patterns and to extract useful features for forming flow pattern clusters for classification in the feature space. Finally, machine learning was applied to the clusters for constructing classifiers to predict the flow patterns automatically. The experimental results from a small-scale flow loop and an offshore petroleum field show that the proposed method is feasible and effective for recognizing the multiphase flow patterns in the pipeline-riser system.