Intelligent recognition of severe slugging in a long-distance pipeline-riser system

In this work, experiments on flow regimes during two-phase flow in a long-distance pipeline-riser system were carried out. The test loop with inner diameter 46 mm consists of a long horizontal pipeline with length 1657 m, followed by a 30 m downward inclined section, and ended at an S-shaped riser with height 11.2 m. A simple method utilizing pressure signals and artificial neural networks is proposed to identify the flow regimes. Firstly, characteristic parameters of pressure signals in time-domain and frequency-domain are extracted, then they are reduced by principal component analysis, and finally flow regimes are recognized by artificial neural networks. The influence of pressure signals at different positions along the pipeline, which includes the horizontal section, the inclined section, the riser section and the separator, on the overall recognition rate is discussed. Higher recognition rates can be achieved by using pressure signals at and near the bottom of the riser rather than pressure signals at other locations along the pipeline. Additionally, a higher recognition rate can be obtained using the pressure difference of the riser instead of the pressure signal. Based on pressure signals at and near the bottom of the riser, when flow regimes are divided into two categories, a satisfactory recognition rate of 93% can be obtained with the signal length of 60 s; When the signal length exceeds 120 s, the recognition rate is about 96%, and it remains unchanged as the signal length increases; When flow regimes are classified into four categories, the recognition rate is about 90% if the signal length exceeds 120 s and remains unchanged as the signal length increases.