Predicting two phase fluid flow regime using machine learning

Major Project -3 Mid Report

submitted by

Rajbhushan Takhenchangbam Sharma

(R870220150)
Subhasis Chakra(R870220182)
Anshuman Singh (R870220024)
Saurabh Gupta (R870220129)
in
partial fulfillment of the requirements for the award of the degree

Bachelor of Technology
in

Applied Petroleum Engineering

with specialization in

Upstream

Under the guidance

of
Mentor- Dr. Seim Timung

Energy Cluster, School of Advanced Engineering

UPES, Dehradun-248007
14, March 2024

CERTIFICATE
1
This is to certify that, the Major Project-3 Mid report entitled, “Predicting two phase fluid
flow regime using machine learning” submitted by Rajbhushan Sharma (R870220150),
Subhasis Chakra(R870220182), Anshuman Singh (R870220024) and Saurabh Gupta
(R870220129) of Semester: VIII, for the award of the degree of Bachelor of Technology
in Applied Petroleum Engineering with specialization in Upstream at Energy cluster,
UPES, Dehradun, India, is a Bonafide record of the Project-3 work carried out by the
students under my supervision and guidance. The content of the report, in full or parts, has
not been submitted to any other Institute or University for the award of any other degree or
diploma.

Name Signature of Student

Rajbhushan Sharma

Subhasis Chakra

Anshuman Singh

Saurabh Gupta

Mentor -

Dr. Seim Timung

Associate Professor
Dept. of Petroleum and Earth Sciences
School of Engineering
University of Petroleum and Energy Studies
Dehradun, Uttarakhand-248007

Place: UPES, Dehradun

Date: 14, March 24

DECLARATION BY THE STUDENT

2
We hereby declare that the Project-3 Mid-review Report submitted to Energy Cluster, UPES,
by us, contains the record/results of the work that was purely carried out by us during my
Semester VIII, under the mentorship of Dr. Seim Timung. It does not contain any material
that was previously published or written by another person as well as any material which has
been accepted and submitted for the award of any other Degree or Diploma to any other
University or Institute.

Date: 14, March 2024

Place: UPES

Rajbhushan Sharma (R870220150)

Subhasis Chakra(R870220182)
Anshuman Singh (R870220024)
Saurabh Gupta (R870220129)

Energy Cluster
School of Advanced Engineering
UPES
Dehradun-248007

3
 ACKNOWLEDGEMENTS

We wish to express our deep sense of gratitude and indebtedness to Dr. Seim Timung,
Associate Professor, UPES, Dehradun, for his inspiring guidance, constructive criticism, and
valuable suggestion throughout this project work. We also extend our sincere thanks to all
our friends who have patiently helped us in accomplishing this undertaking.

Date: 14 March 2024

Place: UPES

Rajbhushan Sharma (R870220150)

Subhasis Chakra(R870220182)
Anshuman Singh (R870220024)
Saurabh Gupta (R870220129)

Energy Cluster
School of Advanced Engineering
UPES
Dehradun-248007
 CONTENTS

Certificate i

Acknowledgment iii

Contents iv

List of tables v

List of figures vi

CHAPTER No. TITLE PAGE-NO

1. Introduction 9 – 10
2. Literature Review 10 – 11
3. Knowledge Gap 11 – 12
4. Objectives 12 – 13
5. Materials & Methods 13 – 17
6. Results & Discussions 17 – 21
7. Conclusion 21 – 22
References 22 – 23
 LIST OF TABLES

Table No Table Title Page No

1. Input Parameters 14

6
 LIST OF FIGURES

Fig. No Figure Title Page No

1. Buffer zone(purple) 13
2. Iris flower 14
3. Bubbly, slug, annular flow regimes in air-water
vertical flow 14 4.
Stratified plug, kerosene Dispersed and water
4. Dispersed flow in horizontal pipe 15
5.

7
 SUMMARY

In multiphase flow, two or more fluids (gas, liquid, or solid) coexist within a conduit. Their
spatial arrangement defines the flow regime, influenced by factors like fluid types, their
volume fractions, and pipe orientation. Common gas-liquid regimes include bubbly flow,
where small gas bubbles are dispersed in a continuous liquid. Understanding these regimes is
crucial in optimizing pipeline design and efficiency across various industries, especially oil
and gas production.
In our project we use a machine learning model called SVC.SVC (Support Vector Classifier)
is a machine learning model that learns from labeled data to categorize new, unseen data
points. In short, you feed SVC training data with known classifications, and it learns to
identify patterns that help predict the category of new data.

For the data feed, we have compiled databases on superficial liquid velocity, superficial gas
velocity, viscosities, densities for the phase in the multiphase flow corresponding to the
observed flow regimes from various relevant sources for our work. After the model has been
trained, we aim at getting a fairly accurate result.

8
 CHAPTER 1
INTRODUCTION

In the world of petroleum engineering, understanding how fluids behave when they flow
through pipes or wells is crucial. Sometimes, these fluids can be a mix of different things,
like oil and gas, or oil and water. When two different substances flow together, it's called a
"two-phase fluid flow." Within this, there are different ways the fluids can move, which we
call "flow regimes." These regimes range from bubbly flow, where small bubbles of gas are
mixed in a liquid, to more complex patterns like slug flow or annular flow.

To help predict how these fluids will flow, scientists and engineers have traditionally used
formulas and rules based on their experience. But now, with the rise of machine learning,
there's a new tool in their toolbox. Machine learning is like teaching a computer to recognize
patterns in data. Instead of relying solely on human-made rules, we can feed lots of
information about fluid behavior into a machine learning model and let it learn the patterns
by itself.

The machine learning model that we have used for our work is called "Support Vector
Classifier" or SVC. It's like a smart sorting machine. The model works by studying the data
given to it. The training data is their study guide. This guide includes examples: the data
itself (think features of an object) and the correct category label (like "cat" or "dog" for
pictures). By analyzing this guide, the SVC model identifies patterns that differentiate the
categories. These patterns become a decision boundary, a mental map to separate the
categories. So, just like the student summarizing key differences between cats and dogs, the
SVC model builds a separation rule. Finally, when presented with new, unseen data (like an
unfamiliar animal picture for the student), the SVC model uses this decision boundary to

9
predict the category, just like the student applying their knowledge to classify a new animal.
Remember, the quality of the study guide matters! Just as a bad guide can lead to poor exam
performance, poor training data can lead the SVC model astray. Overall, SVC models learn
through a process very similar to studying - identifying patterns and building decision
boundaries based on the information they are given.

CHAPTER 2
LITERATURE REVIEW

 Research has also been done that looks at how air-water flows in microchannels.

P. Hanafizadeh,M. H. Saidi, A. Nouri Gheimasi, S. Ghanbarzade(2011) studied the

flow of air-water (two-phase flow) upwards in tiny vertical pipes (mini pipes) using
high-speed cameras. The pipes were different sizes (2mm, 3mm, and 4mm in
diameter) and lengths (25cm, 27cm, and 31cm). By observing the flow patterns, they
aimed to better understand the physics of this phenomenon. They observed several
different flow patterns happening at the same time depending on how fast the air and
water were moving. Based on their observations, they created a flow pattern map that
predicts which flow pattern will occur based on the air and water speeds. This map is
similar to what other researchers have found, which helps confirm the accuracy of
their findings.

 Hideo Ide, Akira Kariya Saki, Tohru Fukano also conducted similar study on how air
and water flow together (two-phase flow) inside tiny channels (Mini channels) under
constant temperature (isothermal). The channels come in two shapes: circular tubes
of varying diameters (1mm, 2.4mm, and 4.9mm) and rectangular channels that are
much wider than they are tall (aspect ratio 1:9.The key aspects they measured include
the patterns the air and water form (flow regimes), the pressure difference along the
channel (pressure drop), and the relative amount of air compared to water (void
fraction).

10
  M. Sharma, P. Ravi, S. Ghosh, G. Das, P.K. Das studied the flow behavior of
kerosene and water mixtures (kerosene-water flow) as they make a U-turn (return
bend) in horizontal pipes. They tested two different bend shapes (U-shaped and
rectangular) and looked at how the flow direction entering the bend (upwards,
downwards, or horizontally) affected the outcome. This involved extensive
experiments to understand hydrodynamics, which is the study of fluids in motion,
within these return bends.

CHAPTER 3
KNOWLEDGE GAP

One significant knowledge gap in the realm of predicting two-phase fluid flow regimes using
machine learning in upstream petroleum engineering is the absence of a streamlined
computational approach that can accurately predict flow conditions or regimes solely based
on input parameters characterizing the flow. While extensive research and experiments have
been conducted to study various types of multiphase flows and identify different flow
regimes, there lacks an easy and computationally efficient method to predict these regimes
directly from flow conditions.

Existing methods often rely on empirical correlations or mechanistic models that may not
adequately capture the complexity and variability of real-world flow regimes, especially
under diverse reservoir conditions. Moreover, these traditional approaches may require
extensive computational resources or manual intervention, limiting their scalability and
applicability in practical petroleum engineering applications.

The development of a computational framework leveraging machine learning techniques

holds promise in addressing this knowledge gap. By training machine learning models on
comprehensive datasets encompassing a wide range of flow conditions and corresponding
flow regimes, it is possible to establish predictive models capable of accurately forecasting
flow regimes based solely on input parameters such as fluid properties, flow rates, and
11
geometrical configurations. However, the design and validation of such models require
further research to ensure their robustness, accuracy, and generalizability across different
reservoirs and operational scenarios.

CHAPTER 4
OBJECTIVES

The objectives in training our model in the multiphase flow regime are to:

1. Feeding the model enough database which includes (superficial liquid flow rates, the
superficial gas flow rate, viscosities,), densities) for studying the flow pattern, to give
us the appropriate flow types (flow regimes) when the initial flow conditions are
prompted in the model.

2. After the model has been trained, accuracy of the model will be tested using the
actual experimental data.

12
 CHAPTER 5
MATERIALS AND METHODS

SVC stands for Support Vector Classification. It's a machine learning

algorithm used to classify things into different categories. SVC separates data
margins and classify them.

 Finding the Margin: SVC looks at the data points and tries to find the widest
possible margin between the two categories. This margin acts like a buffer
zone, making it less likely for new data points to be misclassified.

Fig.1 Buffer zone(purple)

 Support Vectors: The data points that fall closest to the dividing line on
either side are called support vectors. These points are crucial for defining the
best separation line.
13
  New Data Classification: When a new data point comes along, SVC
calculates its position relative to the dividing line established by the support
vectors. Based on which side of the line it falls on, SVC predicts the category
of the new data point.

The Iris flower data set is a classic example used in machine learning for supervised
classification. This dataset consists of 150 samples from three Iris species: Iris setosa,
Iris virginica and Iris versicolor. Each flower is described by four features: sepal
length, sepal width, petal length, and petal width, all measured in centimeters.

Fig. 2 Iris flower

The goal of using this dataset is to build a model that can predict the species of an Iris
flower based on its features.
----------------

DIFFERENT PHASES AND FLOW REGIMES

----------------

14
 Fig.3 Bubbly, Slug, Annular flow regimes in air-water vertical flow.

Fig.4 Stratified, Plug, Kerosene Dispersed and water Dispersed flow in horizontal pipe

----------------

MODEL INPUT DATA

----------------

Sl.No. Vg(m/ Vl(m/s) Diameter Density Density Viscosity Viscosity Contact Flow
s) of of phase of phase of phase of phase Angle type
pipe(mm) 1(kg/m3) 2(kg/m3 1(Pa.s) 2(Pa.s)
)

Table.1 Input Parameters

15
Fig.5 Sample data set (Flow regime map for air-water flow in 1mm microchannel)

CHAPTER 6
RESULTS AND DISCUSSIONS

16
 CHAPTER 7
CONCLUSIONS

 We can see a propagation of pressure front from the (4,4) block which corresponds to
the injection site.

 The pressure propagates throughout all directions of the cells, except on the
boundaries.

 The initial pressure of 4000psi has risen to about 4500 psi, here represented by the
red cells and those cooler places representing a relatively low-pressure jones. This
could imply that the injected fluid has yet to reach that jones.

17
 REFERENCES

[1] P. Hanafizadeh,M. H. Saidi, A. Nouri Gheimasi, S. Ghanbarzade(2011) studied the flow of

air-water (two-phase flow) upwards in tiny vertical pipes (mini pipes) using high-speed
cameras.

[2] Hideo Ide, Akira Kariya Saki, Tohru Fukano(2006) also conducted similar study on how air
and water flow together (two-phase flow) inside tiny channels (Mini channels) under
constant temperature (isothermal).

[3]

[4] Oyerinde, A. A., Alabi, B. A., & Adegbenga, O. A. (2019). Using Jupyter Notebook for
reservoir simulation: A case study. Advances in Petroleum Exploration and Development,
10(6), 255-267.

[5] Oyerinde, A. A., Alabi, B. A., & Adegbenga, O. A. (2019). A Python-based Jupyter
Notebook for reservoir simulation. Advances in Petroleum Exploration and Development,
10(3), 143-154.

[6] Al-Shammari, M., Alfarham, E. G., & Al-Shammari, T. J. (2019). Jupyter Notebook-based
reservoir simulation tools: A review. Journal of Petroleum Engineering Education, 58(4),
241-249.

18
19