For early recognition of flow regime in pipeline-riser system, use of signals from accelerometers are suggested instead of a commonly used pressure gauge signal. A main factor that discriminates between stable flow and severe slugging is drastically decreased vibration in liquid accumulation of the severe slugging. In this study, a support vector machine is employed for binary classification to identify between stable flow and severe slugging. For multi-class classification, a neural network is applied to recognize four classes of stable flow, two types of severe slugging, and irregular transition. The performance is also analyzed based on the signal length employed.


Pipeline-riser structures are generally used in offshore oil and gas field. Transported mixtures of liquids, gases, and solid components through horizontal (or inclined) pipeline are raised up to topside via the vertical riser. In the pipeline-riser system, a typical severe slugging can be produced with undesired blowout. It can cause liquid overflow, low production, high pressure in separators, corrosion increase, overload on gas compressors, and extra fatigue induced by cyclic impact (Hill and Wood, 1994; Kang, 1996; Pedersen, 2017; Schmidt, 1985; Sun and Jepson, 1992; Yocum, 1973). The severe slugging should be detected and controlled as quickly as possible to ensure flow assurance for safety and economic benefits.

Many studies have classified multiphase flow regimes using machine learning algorithms (Goudinakis, 2004; Mi, 1998; Rosa, 2010; Trafalis, 2005; Wu, 2001; Ye and Guo, 2013; Zou, 2017). For the pipeline-riser system, Goudinakis (2004) investigated an experiment in a S-shaped riser system and measured differential pressures. An input of a neural network (NN) was the normalized differential pressures in 100 seconds. Classes of bubble, oscillation, slug, and severe slugging 1 were identified. Ye and Guo (2013) also performed an experiment in the S-shaped riser to acquire differential pressures. The length of sample was 20 minutes. Least square support vector machines (LS-SVM) were trained with statistical features including mean, variance, skewness, and power spectral densities. Five classes of severe slugging 1, severe slugging 2, severe slugging transition, oscillation, and stable flow were recognized. Zou (2017) obtained differential pressure in the pipeline riser system, calculated feature vectors using mean and range of differential pressures, and adopted the LS-SVM. Samples of various length were analyzed, especially the shortest samples of 6.8 seconds were tested based on the LS-SVM trained by samples of 8.99 seconds. Severe slugging, oscillation, stable like flow, and stable flow were classified. The commonly used pressure gauge has advantages in low cost and application at wide range of temperatures and pressures. However, a pressure tap can be blocked; the pressure can be affected not only by flow regime but also by the phase's velocity (Rosa, 2010); and it is difficult to move into another place after its first installation.

This content is only available via PDF.
You can access this article if you purchase or spend a download.