République Algérienne Démocratique et Populaire

Ministère de l’enseignement Supérieur et de la Recherche Scientifique

Université M’Hamed Bougara – Boumerdes
Faculté des Hydrocarbures et de la Chimie

Mémoire de fin d’études

En vue de l’obtention du diplôme: Master 2
Filière : Hydrocarbures
Spécialité : Commande Automatique

Evolving intelligent control of three-phase separator

level loop with slug flow compensation

Pressented by:
Supervised by:
MOKHATCHE Islem Prof. HABBI Hacen
HEDNA Abderrahmene

Année universitaire 2022/2023

Agenda 1 Introduction
Description & Modeling
4
Of the 10-V-255 Separator

2 Problem statement

Separator water level

5 control
( using evolving & classical methods )

3 Multi-phase separation in
oil production facilities

6 Conclusion
 Introduction

- The performance and operational efficiency of petroleum processes

are of utmost importance in
the oil and gas industry.
- Three-phase separators play a critical role in separating oil, water,
and gas in oil production
facilities.
- It is of substantial interest to keep high operational performance of
the separator
through proper controlling of the main control loops
- The project focuses on level control in a three-phase separator,
specifically studying the simulated
model of the industrial separator 10-V-255 in the GS1 center.
 Introduction

Problem statement
We have noticed during our internship time that effective
control of the control loops poses
significant challenges due to the dynamic nature of the
process and inherent complexities that
conventional control methods have limitations in dealing
with, knowing that 30% of emergency shutdowns are caused
by poor level control, and also the slug flow which is
considered as an input disturbance that has a huge impact on
the system’s stability. Our project aims to improve the
control structure using the (RECCo) control strategy, a
model-free control method that learns control
structure and parameters from data streams without prior
2 modeling of the plant.
Multi-phase separation in Oil
production facilities
Multi-phase separation

Multi-phase separation process

In the petroleum industry, the three-phase separation process separates crude oil, associated
gas, and water efficiently. This maximizes production efficiency and ensures compliance with
environmental regulations. The process can be split into for main zones

Flow
distribution Gravity/Coalescing Outlet zone
Inlet zone
zone Zone

1 2 3 4
4
 Multi-phase separation

Industrial separators:

-Petroleum separators are devices that separate the

gaseous and liquid components of a well stream
into different phases.

-They can have different configurations, such as

horizontal, vertical, or spherical, depending on the
flow conditions and design requirements.

5
Multi-phase separation

Separator’s control:
- In three-phase separators, it is essential to control the
water level, oil level, and gas pressure. Consequently,
there is typically a control loop for each of these
variables.
-PID is the conventional method used to control the levels
and pressure inside the separator. This method requires
tuning the PID parameters to achieve satisfactory
performance and stability.

6
Multi-phase separation

Slugging effect:
Large disturbances in the separator process can often be
attributed to slug flow, which is characterized by
intermittent flow variations where liquid flows in
concentrated masses along pipes. This phenomenon is
particularly noticeable in the first separator in the
system but can also propagate throughout the entire
system. Flow variations, such as slug flow, are common
in pipelines that connect oil wells to offshore or
onshore processing units. Slugs are formed by gas
pockets or liquid plugs in the pipe, resulting in
intermittent flow to the plant. The presence of slug flow
can lead to undesired flaring and potentially even plant
shutdown. Consequently, it necessitates larger
operating margins and may reduce the overall plant
7 throughput.
Description & Modeling
of the
10-V-255-separator
Description & Modeling

The surface area of a liquid can change based on its

level, and a calculation can be made using the
Equation , which applies the Pythagorean formula, to
determine the width of the liquid's surface. This
equation considers parameters such as half of the
width (β), the radius (r), the liquid level (h), and the
distance between the liquid level and the center of the
tank

9
Description & Modeling

If the liquid level surpasses the midpoint of the separator, it may result in a negative value for α.
However, this issue can be resolved through symmetrical considerations. The formula for
determining the width of the liquid surface allows us to calculate the area of the cross-section by
integrating line segments from the bottom of the separator to the liquid level. The area of the
liquid cross-section is called A(h).

To determine the volume of the liquid, we can multiply the cross-sectional area by the length

10
Description & Modeling
Water level model:
The liquid level can be determined by utilizing mass balance

11
11
Description & Modeling
Oil level model:

12
 Description & Modeling

the model for oil level change :

The system for the oil level is transformed into the frequency domain for analytical purposes.
The model for oil level is linearized around a working point to create a transfer function. The
water level is assumed to be constant with a value of kw at the working point.

after simplification

13
 Description & Modeling
Gas pressure model:

The pressure of gas in the separator is influenced by the volume of gas available, mass balance, and
gas density. Temperature and flashing also affect gas pressure but are not considered in this
analysis. The model for gas pressure is established by defining the volume of gas available and the
derivative of gas volume.

A volume balance cannot be used in the calculation because the volume of the process is not
consistent. Instead, a mass balance can be used to model the process, using the ideal gas law and
the mass balance as the foundation for the model.

14
Description & Modeling

resulting is:
and is the volume dedicated as a reference level for the oil when oil level equal to oil level
setpoint ℎo = ℎo𝑠𝑝

15
Description & Modeling

Model of an automatic valve:

Valves are typically represented by a mathematical formula that includes a time delay and a time
constant. In this specific facility, liquid valves have a dead band value of 5 seconds and a time
constant of 6 seconds, modeled using a first-order dynamic equation. The assumption is made
that valves and pipes have no friction. The valve equation in the time domain for the nonlinear
valve is provided in Equation

16
Description & Modeling

Water valve model:

The transfer function of the water level control valve is of the form:

With

Knowing that:

17
Description & Modeling

Oil valve model:

18
Description & Modeling

Gas valve model:

There is no time delay modeled for the gas actuator, although there may be some small
delay present. However, this delay is negligible when compared to the time constant. but for
the valve being used in this specific case, the time constant is T g = 60 seconds.

With:

19
Description & Modeling
Control loops for the separator:

The transfer function of the water level regulation loop is given by:

The transfer function of the oil level regulation loop is given by:

The transfer function of the Gas pressure regulation loop is given by:

20
Water level control design & tuning for
the 10-V-255 separator
control design & tuning

This section provides a comparative study of different control design methods, including traditional and intelligent
techniques that have been applied in simulation to the 10-V-255 separator model.

Zeigler- SIMC PSO RECCo

Nichols

1 2 3 4
22
control design & tuning

Conventional Ziegler-Nichols design method:

It is a popular method for tuning P, PI, and PID controllers, it
starts by zeroing the integral and differential gains and then
raising the proportional gain until the system is unstable
allowing you to use the ultimate gain Kcr and the ultimate period
Tcr to calculate the PID parameters
-Applying it to the transfer function of the water level loop will
result in:
Kcr=1144 and Tcr=40s
Using the table, we find:
𝑘𝑃 = 0.6K_cr 𝑘𝑃 = 686,4
𝑇𝑖 = 0.5T_cr 𝑇𝑖 = 20s
23 𝑇𝑑 = 0.125T_cr 𝑇𝑑 = 5s
control design & tuning

Internal model-based (SIMC) design method:

-The Skogestad’s Internal Model Control method is a two-step

procedure for smooth PID controller tuning that has already
found widespread industrial usage and has been shown to work
well on a wide range of processes

- The first step of the SIMC procedure is to obtain a first or

second-order plus delay model. The second step is to derive
model-based controller settings

24
control design & tuning

Internal model-based (SIMC) design method:

-Our transfer function is a first order with integrator with τ_2=6s

θ=5 s k'=
The calculation of kc, Ti, Td, using the table yields:

-The closed-loop time constant (τc) should be set to proper

value as to achieve good performance of the system response in
closed loop
This allows us to determine the following
-To achieve a good balance we chose τc=5s parameters:
𝑘𝑐 = 516,52
25 𝑇𝑖 == 40s
𝑇𝑑 = 6s
control design & tuning

Particle Swarm Optimization based control

design method:
-Particle Swarm Optimization is a population-based optimization
algorithm inspired by the collective behavior of bird flocking or
fish schooling. It is commonly used to solve optimization
problems where the goal is to find the optimal solution in a
search space.

-PSO is known for its simplicity and ability to handle both

continuous and discrete optimization problems

26
control design & tuning

PSO implementation for PID tuning:

-First, we start by creating a Simulink model that contains our water level control loop (in our case
‘normalppl.slx’) and then we integrate it into our matlab code using this command:

-Now we need to set the lower and upper bounds for our P and I and D parameters, the bigger the range the more
time it takes for the particles to navigate the entire range, we chose:

-The last thing is set the number of iterations and particles, the higher the value of these two the more precise the
search will be but it will take more computational power and time:

- Applying the PSO algorithm to tune the water level PID controller parameters gives the following results: Kp =
520.6332 Ti = 1s Td = 1000s
27
control design & tuning

ANYA:
ANYA (Adaptive Neuro-Fuzzy Inference System) is a type of
fuzzy rule-based system that
combines fuzzy logic and neural networks. It is a hybrid
approach that incorporates the
learning capabilities of neural networks with the interpretability
and reasoning abilities of
fuzzy logic

28
 control design & tuning

Robust evolving cloud-based method:

The RECCo controller consists of three parts: reference model,

evolving law, and adaptation law. It is based on the ANYA fuzzy
rule-based system with non-parametric antecedents. The
controller adapts its gains and adds new fuzzy rules based on
incoming data samples. The RECCo controller can use different
forms of defuzzification for the output control.

29
control design & tuning

The reference model:

Choosing the right reference model is vital for the adaptive
system. It should match the uncontrolled process and plant
dynamics. The reference model sets the desired trajectory and
helps minimize tracking errors. It can be a simple linear model
or other types based on the process.

In this case we define simple first order linear reference-model

as:

30
control design & tuning

The evolving law:

The evolving law focuses on adding new clouds (rules) based on

local density calculations. New samples are associated with the
cloud having the highest density or a new cloud is added if the
density is below a threshold.

31
control design & tuning

The adaptation law:

The consequent part of the RECCo controller uses PID-type
control, with each cloud having its own PID parameters. The
parameters are initialized differently for the first cloud and
subsequent clouds. Only the PID parameters of the selected
cloud are adapted, while others remain constant. The adaptation
is achieved using predefined gains and normalized local density.
The control variable is calculated using the PID-type controller
equation. The adaptation of one parameter aims to correct the
offset error of the current operating point. Discrete-time integral
and derivative of the tracking error are calculated.

30
control design & tuning

Application to water level control loop:

33
 control design & tuning

Control results:
This section simulates the control loop using the results of the different PID tuning methods described above
and evaluates their performance in handling slugging, a complex phenomenon in hydrocarbon mixture
treatment separators. Slugging is characterized by sudden and severe variations in the inlet flow rate of the
three-phase separator.

34
 control design & tuning

Level reference
tracking results:

The responses of the water level control loop in the

absence of disturbance obtained for the different
tuning methods are presented and compared in this
figure and table

35
 control design & tuning

Slugging effect
compensation results:

The results indicate that the PID-ZN and PID-

SIMC tuning methods are less effective in
attenuating the slugging effect. These methods
exhibit peaks in the responses and take longer to
compensate for them. On the other hand, the PID-
PSO and RECCo methods show better
performance, despite a maximum peak of (1-
10)% during disturbance rejection in the water
level control loop.
36
 Conclusion
This work proposes a reconfiguration of the control system for the 10-V-255 three-phase separator in the GS1

center of the GEA oil complex. The objective is to effectively regulate variables such as water level, oil level,

and gas pressure, particularly compensating for the slugging effect in turbulent flows. The study involves

understanding the separator's operation, physical modeling, and designing regulation strategies using PID

controllers. The RECCo strategy, which is model-free and parameter-free, demonstrates superior

performance in water level control, effectively compensating for the slugging effect. The results suggest the

potential application of the RECCo controller in industrial automation, offering flexibility and good

performance without requiring a plant model. Implementing the proposed control configuration on the DCS

system would be beneficial from an industrial perspective.

37
THANK YOU
FOR YOUR
ATTENTION