PREDICTION OF SLUG FLOW USING

VARIOUS MACHINE LEARNING

The successful and economic exploitation and transportation of hydrocarbon is a paramount

requirement in the oil and gas industry. With conventional oil and gas reserves continually

being depleted, oil and gas activities in many oil producing nations have shifted to deep

offshore. To ensure that these oil and gas wells are economically viable, most are tied back to

the existing platform in order to reduce capital expenditure (CAPEX). Thus, marginal fields

or deep offshore fields once deemed uneconomical to explore due to being too small to

accommodate standalone offshore processing facility or operating in a plateau or decline

phase production can now be economically produced through a multiphase flow pipeline.

This has created a very popular trend of tying longer multiphase transport pipelines from the

well clusters and wellhead platforms into the production platforms. Invariably, the

transportation of these hydrocarbon by this method faces several flow assurance challenges.

One of the major challenges commonly faced is a phenomenon called Slugging.

Severe slugging may be described as a cyclic flow regime that causes pressure, flow and

temperature oscillations which leads to intermittent delivery of liquid (oil and water) and gas

to processing facilities during hydrocarbon extraction and transportation. It manifests in

pressure and flow fluctuations capable of causing upset in topside process facilities and

structural integrity issues in the pipeline riser system. These fluctuations can cause separator

flooding, production reduction, platform trips and plant shutdown. The large and rapid

variation in flow reduces the average flow output which could be as large as 50%. This

relative inefficiency results in substantial profit losses which may put a production company

out of business (Di Meglio et al., 2010). Therefore, there is a need to handle severe slugs in a

more efficient way.

In offshore oil and gas installations, severe slug is an undesired flow regime in the well-

pipeline-riser systems. Severe slugging causes several oil and gas production problems as it

has proved to have negative impact on the daily production (Havre et al., 2000; Pedersen et

al., 2015; Pedersen et al., 2017). Some of these severe slug related issues include an overload

on gas compressors, fatigue in the transportation pipelines, increased corrosion (Sun et al.,

1992; Zhou and Jepson, 1994; Kang et al., 1996), reductions in production (Isaac et al.,

2011), production slop and high pressure and liquid overflow in the downstream gravity

separators (Yang et al., 2010).

The prevention, limitation or control of severe slug flow is extremely important to the oil and

gas industry as evidently it enhances production, provides significant economic benefit and

improves safety. As a result of this, several studies have been made, and methods and

systems designed and proposed to provide an efficient and reliable solution. Traditionally, the

most popular approach for controlling slugging flow is limiting the topside choke valve

opening, this method however increases back-pressure in producing wells and a reduction in

production rate.

While slug flow can be controlled by choking the riser production, an understanding of the

slugging root causes may present opportunities for minimizing production losses by

employing proactive predictive and prevention methods for slug flow prevention. Operators

of offshore production systems collect and store real-time production data from hundreds or

thousands of sensors. This data is used to monitor and optimize the production. There is the

need to utilize this large amount of data to create a more accurate predictive model based on

machine learning techniques. This techniques promise to make slug flow prevention a more

precise operation, optimizing production, saving cost and ensuring enhanced safety.
1.2 AIM AND OBJECTIVES

This work is aimed at developing a new approach based on a predictive machine learning

models to predict slug flow conditions in a riser system.

To achieve this aim, the research objectives were:

1. To collect and analyse flow data from a pipeline-riser system.

2. To develop a machine learning based predictive model to predict slug flow conditions

3. To investigate the efficiency of the model

4. To compare the model to other existing slug flow prevention and remediation

methods.

1.3 SIGNIFICANCE OF STUDY

This study would provide insight and vital information on the application of AI and machine

learning analysis for real-time well surveillance and its application in preventing severe

slugging

This study also hopes to create and implement a predictive model based on machine learning

system capable of analysing the big data, detecting anomalies and predicting the emergence

of slug flow in pipeline riser systems between it gets the chance to pose a threat to life and

property.

1.4 SCOPE OF STUDY

The scope of this research covers the modelling of slug flow parameters from a typical deep

water offshore pipeline riser scenario to develop a slug flow predictive model capable of

identifying severe slug flow conditions before they pose a threat.

2.1 MULTIPHASE FLOW

As it regards fluid mechanics, multiphase flow can be described as the simultaneous flow of

two or more immiscible phases of matter (gas, liquid, or solid). This flow may consist of two

major liquid components (two-phase flow) or three major liquid components (three-phase

flow). However, although the major components of multiphase flow are gas and liquid, in oil

and gas production other possible components such as sand and dissolved salt can also be

found in most flow streams.

The interactions between these phases for a given pipe configuration (horizontal, inclined or

vertical), subject to the flow rates of the constituent phases give rise to what is usually called

flow regime. Figure 2-2 for example, shows the various flow patterns observed by Shaha

(2000) for a horizontal pipe.

Figure 2-2 Gas-liquid flow regimes in horizontal pipes

2.2.1 Flow regimes in horizontal pipes

A major feature of multiphase flow which makes it an extremely challenging problem to

understand is the fact that multiphase flow tends to take varying forms. In the case of gas-

liquid flow which is the most common multiphase flow type encountered in oil and gas

production several scenarios may occur. The gas may only be represented as small amounts

of tiny bubbles in the liquid. That kind of flow occurs when there is relatively little gas

compared to liquid, at the same time as the liquid flows fast enough to create sufficient

turbulence to mix the gas into the liquid faster than the gas can rise to the top of the pipe.

Example of steady-state flow regime map for a horizontal pipe.

Based on the flow conditions, the flow regimes are usually organised in a graphical form

usually referred to as a flow regime map. To develop a flow regime map, several mapping

parameters are required, and have been used by other authors, such as the phase superficial

velocities Froude number and variation of the phase velocities (Paglianti et al., 1996; Ouyang
and Aziz, 1999; Ouyang and Aziz, 2002). Figure 2.1 shows an example of steady-state flow

regime map for a horizontal pipe.

2.2.2 Flow regimes in vertical pipes

The flow regimes identified in vertical pipelines are often different from that of the horizontal

pipeline (Weisman and Kang, 1969). The challenge of the lack of a universal flow regime

map for interpreting two-phase flow in the vertical pipes still exist. This is due to the

significant effect of phase properties and the pipe diameters on multiphase flow regimes

(Spedding et al., 1998). Despite these limitations, the main types of flow regimes, which are

identified in the vertical pipeline include the bubbly flow, the slug flow, the churn flow and

the annular flow (McQuillan and Whalley, 1985). Figure 2.5 shows typical flow patterns and

the flow regime map for a vertical pipeline.

Figure 2. Gas-liquid flow regimes in vertical pipes.

Most of the published measurements have been carried out on horizontal and vertical pipes,

which is also what has been represented in this report. Pipelines generally follow the terrain

and most often have other inclinations, so the complexity is often larger than illustrated here.
 Figure 2.5: Flow pattern for two-phase gas-liquid vertical flow,

2.3 MULTIPHASE SLUG FLOW TYPES

The transportation of hydrocarbon in a multiphase flow in horizontal and vertical pipelines is

one of the most important aspects of oil and gas production. However, this practice has

consequences, one of which is the rise of slug flow. Slug flow is the intermittent flow of

liquid and gas with inherent unsteady behaviour that manifests in pressure and flow

behaviour capable of causing upset in topside process facilities and structural integrity issues

in the pipeline-riser system. Slug flow is one of the most undesired multiphase flow regimes,

due to the associated instability which imposes a major challenge to flow assurance in the oil

and gas industry. Due to the fact that slug flow may be encountered in both horizontal and

vertical pipelines, there are several types of slug flow. They include three slug types based on

the formation mechanism (Sharma et al., 2002, Godhavn et al., 2005): Hydrodynamic

Slugging, Operationally Induced Slugging and Terrain /Severe Slugging

2.4 SLUG FLOW MODELS

Due to the negative impact of severe slug flow, slug elimination or slug control is of extreme

important in oil and gas production. With deep-water exploration depth reaching all-time

highs, pipeline risers would continue to be relevant and required. As a result, companies must

find ways to control severe slugging in a cheap, safe and sustainable way. Several researches

have explored several severe slugging control techniques and have proposed several models,

techniques and modifications based on experimental, theoretical and field studies.

This section reviews these severe slugging control techniques and their objectives based on

the underlying technologies. The current control techniques can be classified into two, based

on the underlying scientific and/or technological principles employed. The two classifications

are:

2.5.1 Passive Slug Control Techniques

Passive Slug attenuation techniques involves the control of slug flow by creating a change in

the process mainly through design modifications in the facility. These techniques and

modifications include reducing the flowline diameter, dual or multiple risers, riser-base

mixers (Yocum, 1973; Ogazi et al., 2010), slug catcher (Miyoshi et al., 1988), the use of flow

conditioner in the pipeline (Ying et al., 2013a; Xing et al., 2013b; Makogun et al., 2011;

Adedigba et al., 2006), the venturi device of Almeida and Goncalves (1999), self-gas lifting

method of Sarica and Tungesdal (2000) and the bubble breaker of Schrama and Fernandes

(2005).

In one of the much earlier studies and application of passive slug control techniques, Yocum

(1973), identified several different solutions for process changes, which still are being used in

practice today to handle the slug. These solutions can be categorized into three groups:
 1. Reducing the incoming line diameter near the riser to establish a stable flow regime;

2. Using dual multiple risers, instead of a single riser

3. Using fluid remix device, which purposely mixes fluids at the riser base to avoid

liquid accumulation, hence to prevent a stratified flow to progress into a severe

slugging.

These three kinds of solutions form the fundamental basis for most passive slug control

methods. However, more studies have been carried out on these proposed methods, possible

variations and limitations. Ogazi et al. (2010) reported that the reduction of pipe diameter is

still subjected to the constraint of varying production rates throughout the life of a field. Also,

while pipeline size reduction may solve the problem of severe slugging, they create an ideal

environment for the formation of hydrodynamic slug. As a result, there is the challenge of

selecting the ideal and most efficient size reduction for slug attenuation for a field. Ogazi et

al. (2010) also raised the question of the possibility and practicality of laying a small size

pipe.

In a study based on the suggestion of the application of dual risers, Kaasa (1990) proposed a

subsea separator (T-splitter) to distribute the liquid and gas into two risers as a means of

severe slugging elimination. The effectiveness of this technique is questionable as possibility

of liquid carry over into the gas riser exists. Same liquid might fall back into the pipeline at

low gas flow rate thereby blocking the entrance into the gas riser. Prickaerts et al. (2013) also

investigated the slug flow behaviour in a pipeline leading to a dual riser. The pipeline was

split into two risers with the aid of a non-symmetric branch T-splitter. The liquid phase was

reported to have preference for the second riser while the gas phase flow through the first

riser. For various conditions investigated, it was reported that the second riser stands a chance

of experiencing a considerable back pressure due to gravity dominated flow while both risers
have a typical riser base pressure which shows the likelihood of producing slug in both risers

as shown in Figure 2-9. Apart from additional cost for a second riser, the issue of appropriate

splitter to achieve optimum separation of the phases into the risers remain unresolved.

Figure 2-9 Slug attenuation using dual risers (Prickaerts et al., 2013)

Another passive slug control technique is the use of flow conditioners. A flow conditioner is

referred to a specific device that is installed in the pipeline with the objective to affect the

original flow regime. A typical example of this is a Wavy Pipe developed by Xing et al.

(2013) at Cranfield University (UK). A 7-bend Wavy Pipe is illustrated in Figure 2. and it is

placed close to the riser base. This was done to artificially introduce a number of small slugs

through the wavy pipe, so that a severe riser slug can be avoided due to the fact that the

movement of the gas in the pipeline to the riser base is accelerated compared with the liquid

accumulation.

In a similar study, Adedigba et al. (2006) and Adedigba (2007) investigated the possibility of

using a novel helical pipe section upstream a riser pipe to mitigate slugging. The setup is as

shown in Figure 2. This method was reported to hinder the formation of stratified flow
upstream the riser pipe, reduce the region of severe slugging and when severe slugging

occurs, its severity was said to be substantially reduced. Though this method shows the

potential for severe slug attenuation, like many other passive techniques, the challenge lies in

the area of operability.

Almeida and Goncalves (1999) developed and patented a venturi-shaped device as one type

of flow conditioner, which consists of a convergent nozzle section followed by a divergent

diffuser section. This device is supposed to be located as part of the horizontal pipeline near

to the riser base. Venturi-shaped devices can give a pressure drop causing a mixing effect and

converting the stratified flow to a non-stratified flow temporarily.

A similar functional flow conditioner was developed by Makogan (2007). The device was

installed near the riser base to hinder the formation of stratified flow and to accelerate the

fluid into the riser. Although the severity of the slugging was reported to be reduced however

there was an increase in pressure which could potentially lead to reduced production. It

should be noticed that the flow conditioner approach is similar as the permanent choking

approach proposed by Jansen et al. (1996), thereby they both may have a payoff with a

reduced production rate (Ogazi et al., 2009; Pedersen et al., 2014).

2.5.2 Active Slug Control Techniques

In active slug control methods, the mitigation is achieved with the help of an external

influencer which could be manual or automated. This approach involves some automatic

feedback control mechanism, which manipulates some actuators, which installed in the

process system, subject to some sensor feedback signals. These signals can be from pressure,

temperature and/or flow transmitters, depending on which specific system is studied

(Pedersen et al., 2015).

The use riser topside choke valves as a means of active slug control has been studied for

many years, and typical work can be found in Havre and Dalsmo (2001); Di-Meglio et al.

(2012a); Storkaas and Skogestad (2008); Jahanshahi et al. (2012).

2.5.2.2 Gas-Lifting

It has been proved that using artificial gas-lifting is also an effective approach in elimination

severe slugs (Asheim (1988); Plucenio et al. (2012)), through a huge amount of gas might be

needed to generate an actual effect on the flow pattern.

2.6 SLUG FLOW PREDICTION

In a more recent study, Liu et al. (2022) developed hydrodynamics unified model for gas–

liquid two-phase slug flow. The study took into important parameters such as wall and

interfacial friction factors, slug translational velocity and average slug length). Furthermore,

the related parameters for liquid droplet and gas bubble entrainment are given. Accounting

for the gas–liquid interface shape, hydrodynamics models, i.e., the flat interface model (FIM)

and the double interface model (DIM), of liquid film in the slug body were applied and

compared with the experimental data. Their results showed that the predictions for the liquid

holdup and pressure gradient of the DIM were in agreement with experimental data better

than those of the FIM. Also, further comparison between the available experimental results

and Zhang’s model calculations shows that the developed DIM model correctly describes the

slug dynamics in gas–liquid pipe flow.

Kadri (2013) developed a predictive model for the average slug frequency in horizontal

gas/liquid pipe flow. The model proposed took into consideration the probability of slug

formation if slugs are triggered at the antinodes of a sinusoidal perturbation, along the pipe at
the frequency of oscillation of the interface. A slug is assumed to form if and only if triggered

at a space-time far enough from existing slugs. The author observed that with an increase

distance from the inlet, there was a decrease in the probability of slugs forming since the

downstream passage of existing slugs prevents the formation of new slugs. Predictions by the

model are compared with air/water, freon/water and air/oil measurements found in literature,

with a satisfactory agreement. However, the test results gotten showed that for high viscosity

liquids, the proposed model showed significant deviated from expected measurements.

Kadri et al. (2009) introduced a wave transition model from stratified flow to slug flow or

roll-wave regimes. The model was found to successfully predict the evolution of waves and

their transition into either slugs or roll-waves. It also predicted the formation time of slugs

and roll-waves. The model was developed to track the wave crest along the pipe with the

condition that if the crest overtakes the downstream wave end before hitting the top of the

pipe, a roll-wave is formed, otherwise a slug. To test and validate the model, experiments

were carried out taking measurements in air–water horizontal pipe flow facilities with

internal diameters of 0.052 and 0.06m. Furthermore, numerical calculations using a transient

one-dimensional multiphase flow simulator (MAST) which adopts a four-field model were

made.

2.7 ARTIFICIAL INTELLIGENCE IN SLUG FLOW PREDICTION

Sandnes et al. (2019) applied a data-driven approach to investigate the root causes of

slugging in a production system in a subsea field. They used the data to initiate supervised

learning to build predictive models for slugging. By training the models on actual production

data, the authors used feature importance ranking to identify likely drivers of slugging for

further root cause analysis. A selection of signals were investigated as possible drivers behind

slug severity. Focus was put on well specific signals such as pressures, temperatures and flow
rates, in addition to total flow rates, pipeline pressures and temperatures, and settings on the

topside facility. Total liquid rate, especially the water component, is isolated as an important

driver for slugging, while ruling out other signals believed to be important before the

analysis, such as production from individual wells. The slug severity measure was successful

in differentiating between low, moderate and high slugging.

Alhashem (2019) utilized proof-of-concept to assess and prove the practicality of using

supervised machine learning (ML) techniques to predict multiphase flow regimes in

horizontal pipes. The test flow was comprised of air, water, and oil. The input features used

were water cut (the percentage of water), gas superficial velocity, and liquid superficial

velocity. The predicted output was one of six possible flow regimes. The algorithms assessed

in the study were Decision Tree, Random Forest, Logistic Regression, Support Vector

Machine (SVM), and Neural Network Multi-Layer Perceptron (MLP). According to the

results, the best candidate for the dataset was to use the random forest algorithm with a high

accuracy of 90.8% and low training time (0.13 seconds) in case of increasing the size of the

data and features.

Shadloo et al (2020) applied Artificial Neural Network (ANN) modelling to predict the

pressure drop in gas/non-Newtonian liquid to investigate the impacts of physical and

operating conditions on two-phase flow. The design of the MLPNN with a 6-10-1 structure

was carried out using 511 data that were experimentally obtained. The training was done by

85% of these data that were selected randomly for network training and the validation process

was performed by 15% of the remaining data. The actual performance function of the

resulted neural network model was inspected by the rest of the data points. The estimation of

pressure drop using the optimum MLPNN has resulted in the values of 4.58%, 0.0025, 0.05,

and 0.99438 for AARD, MSE, RMSE, and R2 respectively. These statistical values and error
indices prove the reliability of the MLPNN for the prediction of pressure drop in pipes among

the other empirical correlations and various AI-based approaches.

Kim et al. (2020) predicted slug characteristics by several machine learning methodologies

with 2590 experimental data to overcome the limitation of selecting the proper model or

correlation. The random forest results revealed a better training performance and prediction

than that of a deep neural network and present a competitive prediction compared with the

existing correlations, indicating a great potential of utilizing the data-driven machine learning

methodology.