In recent years, interest for petrochemical processes has been increasing, especially in refinement area. However, the high variability in the dynamic characteristics present in the atmospheric distillation column poses a challenge to obtain quality products. To improve distillates quality in spite of the changes in the input crude oil composition, this paper details a new design of a control strategy in a conventional crude oil distillation plant defined using formal interaction analysis tools. The process dynamic and its control are simulated on Aspen HYSYS dynamic environment under real operating conditions. The simulation results are compared against a typical control strategy commonly used in crude oil atmospheric distillation columns.
CCS335 _ Neural Networks and Deep Learning Laboratory_Lab Complete Record
Design and implementation of a control structure for quality products in a crude oil atmospheric distillation column
1. Research article
Design and implementation of a control structure for quality products
in a crude oil atmospheric distillation column
David Sotelo a
, Antonio Favela-Contreras a,n
, Carlos Sotelo a
, Guillermo Jiménez b
,
Luis Gallegos-Canales c
a
Tecnologico de Monterrey, Department of Mechatronics and Automation, Monterrey, Mexico
b
Tecnologico de Monterrey, Department of Chemical Engineering, Monterrey, Mexico
c
Tecnologico de Monterrey, Department of Electrical and Computer Engineering, Monterrey, Mexico
a r t i c l e i n f o
Article history:
Received 14 April 2016
Received in revised form
16 June 2017
Accepted 8 August 2017
Available online 23 August 2017
Keywords:
Distillation process control
Crude oil atmospheric distillation process
Interaction analysis
Process modeling
a b s t r a c t
In recent years, interest for petrochemical processes has been increasing, especially in refinement area.
However, the high variability in the dynamic characteristics present in the atmospheric distillation
column poses a challenge to obtain quality products. To improve distillates quality in spite of the changes
in the input crude oil composition, this paper details a new design of a control strategy in a conventional
crude oil distillation plant defined using formal interaction analysis tools. The process dynamic and its
control are simulated on Aspen ®
HYSYS dynamic environment under real operating conditions. The si-
mulation results are compared against a typical control strategy commonly used in crude oil atmospheric
distillation columns.
& 2017 ISA. Published by Elsevier Ltd. All rights reserved.
1. Introduction
Oil is the most important primary energy source worldwide. As
such, the exhaustion of this raw material serves as a motivation for
the oil industry to optimize its processes [1] and to obtain fossil
fuel products of higher quality.
In a petroleum refining plant (Fig. 1), crude oil components are
separated by the process known as fractional distillation [2]. The
process begins when crude oil enters into the atmospheric column,
where its products can be obtained in different fractions by in-
creasing the temperature. This is possible due to the fact that each
derivative product has a specific boiling point, classified in des-
cending order according to their volatility. In this article, multi-loop
PI/PID controllers are implemented in an atmospheric distillation
column. As in [3–9], the control strategy is proposed considering PI/
PID controllers in order to increase its feasibility to be implemented
in refining processes. Each controller is tuned in terms of the de-
sired response to the process specifications, using the Control
®
Station design tools based on the process open loop step response.
Crude distillates are obtained under quality standards in spite of
variations in the input load's composition. Also, the tower operates
at a constant temperature and fixed pressure, which ensure quality
oil fractions. The control structure's design is based on input-output
variable interactions, represented in a relative gain array (RGA).
Industrial processes are constantly subject to unplanned pro-
cess transients (e.g. process uncertainties, external disturbances
and sudden malfunctions) which can induce strong variations in
plant operating conditions [10]. The proposed control structure
aims to improve distillates' quality of the atmospheric distillation
column, avoiding relying on the expertise and intuition of its op-
erators and process engineers when dealing with changes in the
input crude oil composition. Input crude oil is composed of a
mixture of three types of oils, Maya, Isthmus and Olmec, char-
acterized as heavy, medium and light, respectively. The model
consists of four stages, preflash, atmospheric, stabilizer and va-
cuum where different pieces of equipment, such as condensers,
reboilers, heat exchangers, distillation columns and strippers, are
simulated in steady state and dynamic mode under real operating
conditions using Aspen ®
HYSYS software. Finally, to compare the
performance of the developed control strategy, the refinery plant
is also simulated under actual control structures commonly used
in crude oil distillation plants.
2. Process description
Crude oil is not directly usable. Therefore, separation of mix-
tures and solutions into its components is one of the fundamental
operations in the petrochemical industry. Distillation
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/isatrans
ISA Transactions
http://dx.doi.org/10.1016/j.isatra.2017.08.005
0019-0578/& 2017 ISA. Published by Elsevier Ltd. All rights reserved.
n
Corresponding author.
E-mail address: antonio.favela@itesm.mx (A. Favela-Contreras).
ISA Transactions 71 (2017) 573–584
2. accomplishes the separation of hydrocarbons in the refining pro-
cess. This is possible since crude products have different boiling
points.
In the present research, although the proposed control struc-
ture is implemented in the crude oil atmospheric distillation col-
umn, to present a more realistic conventional refinery plant, the
following stages are simulated:
1. Preflash: the more volatile components of the crude oil are re-
moved; this in order to avoid weeping in atmospheric distilla-
tion column.
2. Atmospheric: the crude oil is heated up to 338.4 °C by a furnace.
The distillation process operates at a constant pressure of
104 kPa, slightly above atmospheric pressure value. The crude
oil is fractionated into five cuts, e.g. naphtha, kerosene, diesel,
atmospheric gas oil (AGO) and atmospheric residue. The column
has physical components called strippers and pumparounds
interacting to control the main distillation process variables
such as temperature and pressure profiles.
3. Stabilizer: the liquid product (naphtha) is treated through a
column to separate gas from oil, operating at a constant pres-
sure of 1 030 kPa. At this stage, combustible gas, LPG (liquefied
petroleum gas) and naphtha stabilized are obtained.
4. Vacuum distillation: using the non vaporized load of crude oil
from the atmospheric stage, the column operates at a constant
pressure of 2 kPa. At this complementary operation, the ob-
tained products are top, bottom, light liquid and heavy liquid
gas oil (LVGO and HVGO).
The studied refining plant is shown in Fig. 2. The input stream
consists of three kinds of crude oils: light, medium and heavy. The
entire plant has four distillation columns (preflash, atmospheric,
stabilizer and vacuum), modeled by trays. There are three con-
densers, located on top of the preflash, atmospheric and stabilizer
columns. In addition, there is a reboiler in the stabilizer column.
The strippers located in the atmospheric column are referred to as
Product-S while pumparounds are under PA legend.
To determine distillation efficiency between two adjacent
products, a comparison of ASTM D86 curves is done. The recovery
degree of the most volatile components is carried out in the light
fraction, whereas the heaviest components are in the heavy frac-
tion. The temperature difference in ASTM D86 curves defines the
quality of the separation between two consecutive fractions. This
is calculated by subtracting the temperature corresponding to 95%
ASTM D86 light fraction (LF) of the corresponding to 5% ASTM D86
of the heavy fraction (HF). If the temperature difference is positive
Fig. 1. Conventional refinery.
Fig. 2. Process flowsheet.
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584574
3. (gap), there is a good separation between adjacent products. On
the other hand, if the difference is negative (overlap), the se-
paration between consecutive fractions is poor [11].
In Fig. 3, ASTM D86 volatility curves of extracted products
in the atmospheric column are shown. Volatility curves are useful
for defining the Gap/Overlap between continuous products. The
Gap/Overlap values obtained in the simulation and their corre-
sponding limits in real processes are shown in Table 1. As it can be
seen, based on the ranges specified in [12,13], the results do not
exceed the limits; therefore, the separation of consecutive pro-
ducts is considered efficient.
3. Modeling and simulation
The refining plant is simulated using Aspen ®
HYSYS , based on
[14]. The oil supply modeling is described in Section 3.1. The se-
paration process is presented in Section 3.2. Separation of the
more volatile components takes place in the preflash column,
described in Section 3.2.1. The distillation process of studied sub-
products is conducted by the atmospheric column, described in
Section 3.2.2. LPG and stabilized naphtha are obtained by the
stabilizer column, presented in Section 3.2.3. The separation of
heavier components is performed in the vacuum column, de-
scribed in Section 3.2.4. Finally, geometric parameters to be in-
troduced in the model are described in Section 3.3.
3.1. Crude feed
Crude oil is the raw material for the distillation process;
therefore, its physical and chemical characteristics [15] must be
established. In crude oil modeling, characteristic data, mixture
proportions and oil pseudo components must be defined. In Aspen
®
HYSYS , modeling of the oil pseudo components is based on the
API (American Petroleum Institute) methodology, where their
density curve and volatility are required.
The modeled refining process receives a mixture composed of
three kind of oils, denominated Light Crude, Medium Crude and
Heavy Crude. Based on its API gravity and volatility curve, raw oil
mined in Mexico corresponds to Olmec, Isthmus and Maya re-
spectively. The input data [16] for crude feed modeling is: dis-
tillation curve, light analysis, density curve, API density and mass
fractions. Fig. 4 shows the load's TBP (True Boling Point) curves
[17] of the process. As shown, the Maya Crude with a FBP (Final
Boiling Point) of 947 °C corresponds to the heavier crude which
means that its density is higher than the others; the Isthmus and
Olmec Crudes have an FBP of 934 °C and 896 °C, respectively.
The raw oil mix composition defines the crude feed to the
preflash column with a new TBP distillation curve. The new FBP is
at 931 °C, which was expected, since the value of the final boiling
point must be within the 896–947 °C range.
Once the characterization of the three types of crude oils is
done, the pseudo components of the crude oil load are calculated
automatically by the Aspen ®
HYSYS software. Thus, based on Eq.
(1) and setting the inflow, the amount of naphtha, kerosene, die-
sel, AGO and residue to distill can be estimated.
Fig. 5 shows the mixed oil cut distribution presented in Table 2,
where the temperature at which each cut starts to boil is refered to
as the IBP (Initial Boiling Point), and the temperatures at which
each cut boils off completely is refered to as FBP (Final Boiling
Point). In the present research, the input oil is at a temperature of
232.1 °C and at a pressure of 517 kPa, considering an input stream
operating point of 99 000 barrels/day.
= · ( )Q f Q 1distilled r in
where Qdistilled indicates the liquid volumetric flow distilled
⎡⎣ ⎤⎦barrels day/ , fr
stands for the liquid volumetric fraction of total oil
and Qin indicates the input liquid volumetric flow ⎡⎣ ⎤⎦barrels day/ .
Fig. 3. ASTM D86 volatility curves.
Table 1
Gap/Overlap evaluation.
Admissible Simulation
Gap/Overlap
Naphtha-Kerosene ≥ °16.7 C °28.2 C
Kerosene-Diesel − °11 C a + °28 C °8.7 C
Diesel-AGO − °20 C a °0 C − °2.6 C
AGO-Residue − °30 C a − °10 C − °30.0 C
Fig. 4. Crude feed TBP curves.
Fig. 5. Cut distribution of crude oil.
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584 575
4. 3.2. Separation process
3.2.1. Preflash stage
The first stage in refining crude oil corresponds to the preflash
column, operating at a pressure of 167 kPa and a temperature of
42.7 °C. At this stage, the lighter components called PreFlash-Vap
are removed, producing Light Naphtha at the top and PreFlash-Liq
at the bottom.
The condensing section is modeled by a partial condenser that
removes lighter components. In real plants, in order to reduce the
condenser temperature, a heat exchanger network is im-
plemented; however, this is modeled by a single exchanger (Q-1).
3.2.2. Atmospheric stage
A key stage in the refining process is the atmospheric distilla-
tion [18], operating at a pressure of 104 kPa and at a temperature
of 76.9 °C. At this stage, five extractions are performed by loading
crude treated in the bottom section of the preflash column: heavy
naphtha (labeled simply Naphtha), kerosene, diesel, atmospheric
gas-oil (AGO) and atmospheric residue (Residue).
A furnace (Heater ATM) is implemented, which receives refined
light oil at 228.4 °C to heat it up to 338.4 °C. This is possible re-
gards to −Q 3, which provides the amount of heat required. A
partial condenser at the top of the column is also implemented,
accumulating naphtha and processing water. Similar to the pre-
flash column, a single exchanger removes heat ( −Q 2) to reduce
the condenser temperature. The four pumparounds (PA-1, PA-2,
PA-3 and PA-4) are modeled as heat exchangers.
3.2.3. Stabilizing stage
In order to stabilize naphtha, a stabilizing column is added, op-
erating at a pressure of 1 030 kPa and at a temperature of 159.1 °C. At
this stage, using a pump (P-101), naphtha pressure increases from
104 to 1 825 kPa, and by means of a heat exchanger network (HEN-1),
naphtha temperature increases from 77.6 to 250 °C.
The stabilizer column works at a higher pressure than the rest
of the columns, and it receives 20,000 barrels/day of naphtha. At
the condensing section, there is a total condenser, producing LPG.
As in the previous columns, a heat exchanger Q-5 removes the
excess temperature. The reboiler is known as Kettle reboiler (Once-
through), where the liquid in the last tray is passed through a heat
exchanger. The heat injected by Q-6 generates vapors which are
returned into the bottom of the column. No evaporated product is
called Stabilized Naphtha.
3.2.4. Vacuum stage
The vacuum stage receives residue (bottom product) from the
atmospheric column, operating at a pressure of 2 kPa and at a
temperature of 112 °C. At this stage, evaporation of the heavier
components is achieved, which, at the same temperature but at
atmospheric pressure, had not been able to be removed [19].
Through a pump (P-102) and a heater (Heater Vac), atmospheric
residue (Residue) is extracted and maintained at the desired
temperature.
The vacuum column receives approximately 38,210 barrels/day
of Residue at a pressure of 234 kPa, and through a heater called
Heater Vac the temperature is mantained at 408.5 °C, avoiding
thermal cracking. The purpose is to carry out the distillation
without producing coke because the trays could get blocked up.
Considering the upper and lower part, four extractions (VacOver-
head, LVGO, HVGO and VacResidue) are performed. Pumparounds
PA-5 and PA-6 are modeled as heat exchangers [20,21].
3.3. Geometric parameters
The geometric parameters to be introduced in each of the main
columns (Preflash, Atmospheric, Stabilizer and Vacuum), and strippers
(Kero-S, Diesel-S and AGO-S) are described in Section 3.3.1. Finally,
geometric parameters for vessels ( Cond Cond Cond, ,Pfl Atm Stab and
RebStab) are described in Section 3.3.2.
3.3.1. Columns and strippers
In order to size a column [22], different volumetric flows must
be considered for each of the zones in the column; therefore, As-
pen ®
HYSYS recommends dividing the column by sections. As part
of the design, the number of internal flow paths must be proposed.
The column diameter corresponds to that of the largest of the
different sections. Both, the tray spacing and the height of the weir
are appointed from automatic calculations performed by the
software settings. The weir length is obtained by dividing the total
weir length (given by HYSYS) by the numbers of internal flow
paths Eq. (2). The pressure along the column is obtained by mul-
tiplying the maximum pressure increase per tray by the total
number of trays Eq. (3). Finally, adding the top pressure and the
pressure increase along the column, the bottom pressure is ob-
tained Eq. (4). Fig. 6 shows the internal parameters.
=
( )
Weir length
Total weir length
No. of internal flow paths 2
( )( )Δ = Δ ( )Total P Max P/Tray No. of trays 3
= + Δ ( )Bottom pressure Top pressure Total P 4
3.3.2. Condensers and reboilers
The total outflow is the sum of the liquid flows with zero phase
steam leaving the vessel Eq. (5). The residence time is an esti-
mating time required for material flow, stay and exit the vessel.
Considering these parameters and an approximated liquid level
[22], the vessel volume can be calculated using Eq. (6). Fig. 7 shows
the parameters to be considered.
= + ( )Total outflow flow 1 flow 2 5
Table 2
Crude oil fractions to be distilled.
IBP FBP Fraction Production
[ ]°C [ ]°C ⎡⎣ ⎤⎦barrels day/
Naphtha 70 180 0.157 15,543
Kerosene 180 240 0.105 10,395
Light Diesel 240 290 0.087 8613
Heavy Diesel 290 340 0.083 8217
AGO 340 370 0.048 4752
Residue 370 932 0.457 45,243
Fig. 6. Column parameters.
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584576
5. ( )=
( )
Volume
Total outflow Residence time
Liquid level 6
Tables 3 and 4 show the geometric parameters to be introduced
in the model for columns (both main and strippers) and vessels,
respectively. The results presented in Table 3 are reasonable. This
can be seen in the values obtained for the weir length for the
different columns, since according to [23], these must be between
60–85% of the diameter of the main column section. Furthermore,
the volumes specified in Table 4 are close to those in real plants;
e.g., typical atmospheric condenser volume is 90 m3
[23].
Table 5 shows the top pressures for which the process is si-
mulated, and the theoretical values consulted in piping diagrams.
As it can be seen, these pressures are not far from those in an
actual process; the minimal difference is due to simplifications in
modeling the condensing part. Another important result reported
in Table 5 is the weir length, which, according to consulted sources
[23], should be in a range defined by the diameter of the main
section. Although only the required specifications in dynamic
mode to run the simulation are presented, the crude oil units were
carefully designed based on a deep literature review where geo-
metric parameters, pressures, temperatures and numbers of trays
are specified. More detailed explanation about dynamic simulation
setup is reported in [24].
4. Control structure
4.1. Variable pairing selection
System control design in crude oil distillation processes de-
pends on plant conditions. Consequently, control strategies exhibit
notable differences as much in complexity as in cost, and their
development relies on the relationship between manipulated and
process variables. Inappropriate selection of the manipulated or
controlled variables causes an inefficient control [25], even if an
advanced control strategy is implemented [26]. In this section,
with the purpose of designing an efficient control system, a
structured method for selecting manipulated and controlled vari-
ables is presented.
4.1.1. Static gain matrix
Proper selection of the control variables, as well adequate in-
stallation of the sensors, is fundamental in controlling a crude oil
distillation column [27]; however, many times in the industry,
variables are ignored during control strategy design, therefore
operational problems appear and its development is affected. The
loop sensitivity, showed in Eq. (7), measures how the sensor signal
responds in stationary state to changes in each manipulated
variable [28].
=
∂
∂ ( )
LS
S
M 7
i
j
where LS stands for loop sensitivity, Si indicates the controlled
variable i expressed in percentage of the maximum value, Mj
stands for the manipulated variable j measured in percentage of
the maximum value. Taking into account Eq. (7), static gain matrix
G, showed in Eq. (8), can be obtained by setting the column sen-
sitivity with respect to the boiling temperature ( =T Si i), at 5 % and
95 %, of each distillated product. The sensitivity is determined by
opening the valves corresponding to side extractions flow and
steam supply to strippers.
The static gain matrix G (Eq. (8)) describes the influence of each
opened valve on each boiling temperature, thus a control strategy
capable of reducing the effect that disturbances have on product
quality can be defined.
Fig. 7. Condenser parameters.
Table 3
Geometric parameters for the columns.
Diameter Tray
spacing
Weir
height
Weir
length
Bottom
pressure
[ ]m [ ]m [ ]m [ ]m ⎡⎣ ⎤⎦kPa
Preflash 3.200 0.6096 0.0508 2.5575 415.79
Atmospheric 6.248 0.6096 0.0508 4.9230 198.54
Kero-S 1.829 0.6096 0.0508 1.5335 185.95
Diesel-S 1.981 0.6096 0.0508 1.6930 187.02
AGO-S 1.219 0.6096 0.0508 0.9618 193.46
Stabilizer 2.591 0.6096 0.0508 2.0916 1 043.752
Vacuum 12.95 0.6096 0.0508 8.5200 16.21
Table 4
Geometric parameters for the condensers and reboilers.
Total
outflow
Geometry Orientation Volume Diameter Height/
Length
⎡
⎣
⎤
⎦m h/3 ⎡
⎣
⎤
⎦m3 [ ]m [ ]m
CondPfl 73.16 Cylindrical Vertical 24.39 2.746 4.119
CondAtm 278.476 Cylindrical Vertical 92.82 4.287 6.431
CondStab 91.08 Cylindrical Vertical 30.36 2.954 4.431
RebStab 96.06 Cylindrical Horizontal 32.02 3.007 4.510
Table 5
Comparison between simulation and real data.
Simulation Real
Top pressure [ ]kPa
Preflash 167 170
Atmospheric 104 105
Stabilizer 1 030 1 050
Vacuum 2 2
Weir length [ ]m
Preflash 2.56 1.9–2.7
Atmospheric 4.92 3.8–5.3
Kero-S 1.53 1.1–1.6
Diesel-S 1.69 1.2–1.7
AGO-S 0.96 0.7–1.0
Stabilizer 2.09 1.6–2.2
Vacuum 8.52 7.8–11.0
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584 577
6. ⎡
⎣
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎤
⎦
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
⎥
=
∂
∂
⋯
∂
∂
⋯
∂
∂
⋮ ⋮ ⋮
∂
∂
⋯
∂
∂
⋯
∂
∂
⋮ ⋮ ⋮
∂
∂
⋯
∂
∂
⋯
∂
∂ ( )
G
T
M
T
M
T
M
T
M
T
M
T
M
T
M
T
M
T
M 8
j m
i i
j
i
m
n n
j
n
m
1
1
1 1
1
1
where G indicates the static gain matrix containing n boiling
temperatures from m manipulated variables and Ti stands for the
boiling temperature i of distilled product at 5% and 95%, measured
in percentage.
4.1.2. Singular value decomposition
Singular value decomposition (SVD) of the static gain matrix,
presented in Eq. (8), is a useful technique to analyze multivariable
interactions. When applying this technique, the static gain matrix
is decomposed into three matrices, as it is showed in Eq. (9).
Σ= ( )G U V 9T
where U is an ortonormal matrix with dimension ×n n, its col-
umns are left singular vectors, Σ is a diagonal matrix with di-
mension ×n m, singular values, s.t. σ σ σ≥ ≥ ≥ 0m1 2 and V is an
ortonormal matrix with dimension ×m m, its columns are right
singular vectors.
4.1.3. Stability and decoupling condition
In pairing variables, the strongest interactions between input-
output variables should be considered. Additionally, the Nie-
derlinski Index is defined in Eq. (10), where negative values in-
dicate the existence of paired variables that affect system stability,
regardless of controller tuning [28].
=
∏ ( )=
NI
G
g 10i
n
ii1
where NI is the Niederlinski index, >NI 0 as a necessary and
sufficient condition for stability, G is the determinant of static gain
matrix and gii
are the diagonal elements of matrix G.
Moreover, singular value decomposition shows us the degrees
of freedom of the control strategy. In the diagonal matrix of sin-
gular values Σ, smaller values indicate a difficulty in control. The
Conditional Number, presented in Eq. (11), defines the degrees of
interaction among control loops, and it is set as the ratio between
the maximum and minimum singular values [28].
σ
σ
=
( )
CN
11
max
min
where CN is the conditional number, σmax indicates the maximum
singular value and σmin indicates the minimum singular value.
Therefore, the CN can be defined as a numerical index of the
system sensitivity. The higher the CN value, the more difficult it is
to decouple the interactions of the control loops. Based on [29], if
≥CN 20, the system is almost singular and decoupling is not
feasible.
4.1.4. Relative gain array
Relative gain array (RGA) is a mathematical tool useful for
identifying the strongest interaction between manipulated and
controlled variables. In this article, RGA is applied in order to de-
fine a control structure [30].
RGA is showed in Eq. (12):
⎡
⎣
⎢
⎢
⎢
⎤
⎦
⎥
⎥
⎥
Λ
λ λ
λ λ
=
…
⋮ ⋱ ⋮
… ( )12
n
n nn
11 1
1
where each element is obtained by means of Eq. (13) [31].
λ
λ
=
( )
′ ( )
=
′
= …
( )
→
−
→
−
G s
G s
g
g
i j n
lim
lim
, 1, ,
13
ij
s ij
All control loops opened
s ij
Remaining control loops closed
ij
ij
ij
0
0
An alternative way to obtain the RGA is through Eq. (14).
⎡⎣ ⎤⎦( )Λ λ= = ⊗ ( )
−
G G G 14ij
T
where ⊗ stands for Hadamard Product and −
G T
corresponds to the
transpose of the inverse matrix of G. The RGA pairing criterion
relies on:
The sum of the elements in each row or column is 1.
The control loops should be defined by relative static gains
closer to 1.
Control loops with relative static gains tending to infinity or
zero should not be defined [32].
4.2. Control structure definition
In the Section 4.2.1, a control strategy to achieve stability in
distillation columns is described. Then, in Section 4.2.2 control
strategies to regulate the liquid volumetric flow of distillates in
atmospheric distillation columns are presented. Finally, in Section
4.2.3 an efficient method for defining manipulated and controlled
variables is presented.
4.2.1. Stability in distillation columns
Stability in the pressure on the top of the column is achieved
once the liquid level and the pressure into the condenser are
regulated. In the atmospheric distillation column (Fig. 8), the li-
quid level in the condenser is usually controlled by a PID controller
LIC-100 which determines the reflux to the top of the column. This
control scheme is commonly used in the direct feed split control
scheme, when the heat input is limited or must be fixed and
distillate is manipulated directly to control the composition profile
[33,34]. In addition, the pressure in the condenser is regulated by
PIC-100 which establishes the extracted heat from the vessel [35].
These control strategies were also implemented in preflash and
stabilizer columns. For the side extractions, single feedback control
loops are commonly used. Hence, PID controllers LIC-1000, LIC-
1001 and LIC-1002 controls the liquid level of strippers Kero-S,
Diesel-S and AGO-S to maintain the liquid volumetric flow of Ker-
osene, Diesel and AGO respectively. Nowadays, this control strat-
egy is commonly used in refinery processes.
On the other hand, crude oil temperature, processed by the
preflash section, must be kept in a range of 290 to 370 °C, allowing
the phase change to occur without thermal cracking. In this case,
ATM Crude is set at 338 °C using a PID controller TIC-300 which
regulates the input thermal load to the furnace.
4.2.2. Distillates flow in the atmospheric column
In order to ensure that sufficient material is leaving the column,
PID controllers are implemented (Fig. 9). In a typical operating
scenario for the distillation process, a PID, FIC-300, regulates the
liquid volumetric flow of distilled naphtha, using the input power
of a pump P-100 as manipulated variable. In this work, for the
remaining side extractions, single feedback control loops are used.
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584578
7. Hence, PID controllers FIC-1000 and FIC-1002 regulate the liquid
volumetric flow of kerosene and AGO respectively, while LIC-1001
controls the liquid level of stripper Diesel-S to maintain the liquid
volumetric flow of Diesel. The manipulated valves from these
controllers are respectively V-100, V-102 and V-101. Looking for a
control of distilled Diesel effectively that rejects disturbances, a
cascade control strategy is proposed [36]. The FIC-1001 controller
generates the reference for the liquid level control system of the
stripper Diesel-S, based on the measurement of the liquid volu-
metric flow of Diesel.
4.2.3. Distillates quality in the atmospheric column
Control strategy definition for improving distillates quality in
the atmospheric column begins with the input-output interaction
analysis. The applied method consists of:
1. Compute the static gain matrix G (Eq. (8)) containing the
relation between manipulated and controlled variables in the
atmospheric distillation column.
2. Verify whether or not the process is stable using the Nie-
derlinski Index (Eq. (10)), NI 0. In case this condition stability
is not satisfied, reduce the number of pairing variables in the G
matrix considering those with lower absolute value. Once this is
done, verify the stability condition again.
3. Verify whether or not the process is decoupled based on the
Conditional Number (Eq. (11)), CN 20. In case this condition is
not satisfied, there exists at least one manipulated variable
strongly affecting two or more process variables, thus it must be
eliminated from G.
4. Compute the RGA matrix (Eq. (14)).
5. Define for each manipulated variable its corresponding
controlled variable in order to set the control loops. This should
be done relying on the highest degree interaction.
Table 6 shows the manipulated and controlled variables in-
itially considered to control the distillates quality [24]. Thus, G0
(15) corresponds to the initial static gain matrix of the atmo-
spheric distillation column. Columns of the G0 matrix indicate the
manipulated variables ( OPj), while rows stand for the controlled
variables (Ti).
⎡
⎣
⎢
⎢
⎢
⎢
⎢
⎢
⎢
⎤
⎦
⎥
⎥
⎥
⎥
⎥
⎥
⎥
=
−
−
− − ( )
G
0.0179 0.1711 0.0061 0 0 0.0052
0.0109 0.9812 0.0061 0.0712 0 0.0049
0.0100 1.0600 0.0381 0.0198 0 0.0055
0.0076 0.8195 0.0139 0.1348 0 0.0052
0.0003 0 0.0142 0.0761 0 0.0312
0.0014 0 0 0.0075 0 0.0133 15
0
Before we apply RGA criterion, we must select the most re-
presentative variables to control destillates Gap/Overlap.
Analyzing the columns of G0 matrix, one can observe that the
controlled variables associated with OP102 show a magnitude zero.
This implies that these interactions can be neglected. On the other
hand, it is observed that OP100 is strongly correlated with more
than one controlled variable, TK05, TK95 and TD05. Thus, this ma-
nipulated variable should be also neglected to preserve the system
stability.
By observing the rows of G0 matrix, and considering that
Kerosene-Diesel Gap/Overlap can be controlled through TK95 or
TD05, one can conclude that TK95 must be neglected since OP105
affects TD05 6.25 times more. Then, considering that Diesel-AGO
Gap/Overlap can be controlled through TD95 or TA05, one must ne-
glect TA05 because OP101 affects TD95 2.44 times more.
Fig. 8. Atmospheric column under original control structure.
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584 579
8. The resulting static gain matrix Gp is showed in (16).
⎡
⎣
⎢
⎢
⎢
⎢
⎤
⎦
⎥
⎥
⎥
⎥
=
− − ( )
G
OP OP OP OP
0.0179 0.0061 0 0.0052
0.0100 0.0381 0.0198 0.0055
0.0076 0.0139 0.1348 0.0052
0.0014 0 0.0075 0.0133
T
T
T
T 16
p
K
D
D
A
104 105 101 106
05
05
95
95
With =NI 0.8996 and =CN 10.6, the system is stable and de-
coupled; this guarantees there is no impact among unpaired variables.
RGA is obtained based on Gp, and it is shown in ( )Λ Gp (17). By
this matrix, the input-output variables that present the largest
interactions are indicated in bold.
⎡
⎣
⎢
⎢
⎢
⎢
⎤
⎦
⎥
⎥
⎥
⎥
( )Λ =
−
− − −
−
( )
G
1.0669
1.1595
1.0404
0.9637
OP OP OP OP
0.0937 0 0.0268
0.0967 0.0495 0.0133
0.0026 0.0658 0.0228
0.0272 0 0.0091
T
T
T
T 17
p
K
D
D
A
104 105 101 106
05
05
95
95
Attending this pairing criterion and using PID controllers, the
proposed atmospheric column control scheme is shown in Fig. 10.
Results show that for controlling the Naphta-Kerosene Gap/Over-
lap, the TIC-104 controller manipulates the position of the steam
supply valve V-104 (OP104) in order to modify ASTM D86 5% Ker-
osene temperature (TK05).
Similarly, to maintain in range Kerosene-Diesel Gap/Overlap,
the TIC-105 controller is designed, the position of the steam supply
valve V-105 ( OP105) modifies ASTM D86 5% Diesel temperature
(TD05).
The TIC-101 controller is added for the control of Diesel-AGO
Gap/Overlap. This controller is different from the previous ones as
it generates the reference for the proposed cascade control strat-
egy (OP101). Thus, the distilled Diesel modifies the ASTM D86 95%
Diesel temperature (TD95).
Finally, the TIC-106 is implemented to control AGO-Residue
Gap/Overlap. The position of the steam supply valve V-106 (OP106)
modifies ASTM D86 95% AGO temperature ( TA95). The im-
plemented control structure in the entire refinery process is
shown in Fig. 11.
Fig. 9. Atmospheric column under proposed control structure.
Table 6
Manipulated and controlled variables initially considered.
Matrix G0 Variables description
Column Manipulated variable
1 OP104 Kero-S steam supply
2 OP100 Distilled Kerosene flow
3 OP105 Diesel-S steam supply
4 OP101 Distilled Diesel flow
5 OP102 Distilled AGO flow
6 OP106 AGO-S steam supply
Row Controlled temperature
1 TK05 Kerosene ASTM D86 5%
2 TK95 Kerosene ASTM D86 95%
3 TD05 Diesel ASTM D86 5%
4 TD95 Diesel ASTM D86 95%
5 TA05 AGO ASTM D86 5%
6 TA95 AGO ASTM D86 95%
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584580
9. Fig. 10. Atmospheric column final control structure.
Fig. 11. Control process flowsheet.
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584 581
10. 5. Simulation modeling and results
The purpose of this section is to present simulation modeling
and results obtained after implementing the proposed control
structure in the atmospheric column, described in Section 4.2.
As soon as the atmospheric column process becomes stable,
input flow changes are applied to the system in order to disturb it,
and evaluate the proposed control structure. The simulated dis-
turbances correspond to a typical composition changes that take
place in a refinery of Mexico. The objective of the control scheme
is to keep temperature and flow variables in specific reference
values which guarantee distillate quality.
5.1. Liquid level in the condensers and pressure on the top of
columns
Table 7 shows pressure reference values required to keep dis-
tillation columns stable. Pressure and level variables in condensers
remain constant even though input crude oil changes.
5.2. Atmospheric distillation column
5.2.1. Temperature in ATM Crude
In spite of disturbances appearing in the process, crude oil
temperature remains stable at its reference value of 338.4 °C, with
336.7 °C as minimum, and 340 °C as maximum, and these tem-
peratures do not represent a possible thermal cracking. The con-
troller's performance under these circumstances is evaluated by
Table 7
Pressure values on condensers of actual distillation columns.
Reference value
Top pressure [ ]kPa
Preflash 167
Atmospheric 104
Stabilizer 1 030
Fig. 12. Simulation results.
Table 8
Gap/Overlap evaluation to input crude oil changes.
Cut Mean Variance Admissible
Gap/Overlap
Time out
of range ⎡⎣ ⎤⎦h
Original Proposed Original Proposed Original Proposed
Naphtha-Kerosene 27.6 28.1 4 −
e 2
6 −
e 5 ≥ °16.7 C 0 0
Kerosene-Diesel 10 8.5 0.61 0.44 − °11 C a + °28 C 0 0
Diesel-AGO À0.02 À2.5 2.75 0.73 − ° °20 C a 0 C 18.8 0.8
AGO-Residue À12.8 À28 18.1 1.01 − °30 C a − °10 C 8.29 2.15
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584582
11. analyzing the minimum and maximum overshoot, with values of
5.8 % and 6.2 %, respectively.
5.2.2. Pressure profile
Although the atmospheric column is disturbed by composition
changes, the liquid level in the condenser, regulated by the ex-
traction of distillate crude oil from the accumulator, remains
constant. In addition, the pressure on the top of the column is
controlled at its reference value of 104 kPa, by removing heat from
the condenser, with 104.1 as maximum, and 103.9 kPa as mini-
mum. Consequently, the unit's pressure profile, obtained by
measuring pressures in the condenser and side extractions (Ker-
osene, Diesel and AGO), remains stable.
5.2.3. Temperature profile
Since temperature profile stability in the atmospheric column
is one of the main control targets in the present work, temperature
sensors are located along the unit at the output of their trays.
Temperatures in output products of the column vary according to
their extraction point; the lower the extraction point, the higher
the temperature value, and this relationship is kept continuously,
T T T T TNaphtha Kerosene Diesel AGO Residue. This goal is achieved by
regulating input steam flow in the strippers and output distillate
flow from the column. By doing this, the temperature profile in the
atmospheric column remains constant, preserving product
composition.
5.3. Gap/Overlap in distillate products
In order to define the distillate quality achieved in consecutive
products, Gap/Overlap is evaluated in the presence of disturbances.
The original and the proposed control structure's performances
are observed once input flow composition changes.
The simulation results are shown in Fig. 12. The first three graphs
show the sequence of changes of flow and composition of the load
composed of a mixture of Olmec, Isthmus and Maya crude oils. The
following four graphs show the responses of the Gap/Overlap indexes
that were obtained respectively for the cuts of: kerosene-naphtha
(KER-NAP), diesel-kerosene (DIE-KER), atmospheric gasoil-diesel
(AGO-DIE) and residue-atmospheric gasoil (RES-AGO).
As it can be seen in Fig. 12, using the original control strategy in
the atmospheric distillation column (Fig. 8), distillates quality
varies considerably, hence Gap/Overlap AGO-DIE and RES-AGO are
out of the admissible range for a certain period. On the other hand,
under the proposed control system (Fig. 10), process variability is
reduced even when there are abrupt changes in crude oil com-
positions. Then, improvement is observed in:
Increasing plant stability.
Attenuating Gap/Overlap variability.
Reducing Gap/Overlap values out of range.
In order to compare the performance of the original and proposed
control schemes, Table 8 shows a summary of Gap/Overlap once
the system is disturbed. Using the proposed control scheme, the
reduction of the variance of the process in terms of the settling
time is evident, thus quality crude oil distillates is improved. In
addition, crude oil temperature, column pressure and temperature
profile are maintained at the right values.
6. Conclusion
The simulation results shows that the proposed control strat-
egy, using PID controllers, increases the distillates quality in a
typical refinery plant. Pressure and temperature profiles in dis-
tillation columns are maintained close to real operating values, in
spite of changes in the input crude oil composition.
Comparing the refinery process model of the present research
to previous simulations [37–39], the current virtual plant consists
of redesigned distillation columns; i.e., preflash, atmospheric,
stabilizer and vacuum.
In this paper we propose a cascade control strategy to regulate
the Diesel flow. Usually, distillates flows are regulated by single
feedback control loops; however, considering the time constants
of the process, a cascade structure is implemented, improving the
control system performance. In addition, by analyzing interactions
in the static gain matrix, we identify which variables should be
manipulated in order to control those that affect products quality.
Thus, measuring performance as a function of Gap/Overlap index,
we propose a control structure that ensures the distillates quality.
The proposed solution ensures that cut-oils are most of the time
inside of the admissible Gap/Overlap margins face to realistic
changes in the input crude oil composition.
In future work, the RGA analysis with multiple operating fre-
quencies will be applied to petrochemical processes in order to
improve the actual control system performance.
Acknowledgments
This work is supported by the Tecnológico de Monterrey and
the National Council for Science and Technology (CONACYT),
México. The authors also want to show great thanks to the re-
search group of Sensors and Devices of the School of Engineering
and Sciences for its support for the development of this work.
References
[1] Bothamley M. Drilling and production. Oil Gas 2004;102(45):47–55.
[2] Lawal SA, Zhang J. Actuator fault monitoring and fault tolerant control in
distillation columns. Int J Autom Comput 2016;3(11):865–70.
[3] Maghadea D, Patre B. Decentralized pi/pid controllers based on gain and phase
margin specifications for tito processes. ISA Trans 2012;51(1):550.
[4] Ali A, Majhi S. Pid controller tuning for integrating processes. ISA Trans
2010;49(1):70.
[5] Yeroglu C, Tan N. Classical controller design techniques for fractional order
case. ISA Trans 2011;50(1):461.
[6] Das S, Saha S, Das S, Gupta A. On the selection of tuning methodology of FOPID
controllers for the control of higher order processes. ISA Trans 2011;50(1):376.
[7] Vrančić D, S., Kocijan J, de Moura Oliveira PB. Improving disturbance rejection
of pid controllers by means of the magnitude optimum method. ISA Trans
2010;49(1):47.
[8] Maudi RK, Dey C. Performance improvement of pi controllers through dy-
namic set-point weighting. ISA Trans 2011;50(1):220.
[9] Vu TNL, Lee M. Independent design of multi-loop pi/pid controllers for in-
teracting multivariable processes. J Process Control 2010;20(1):922.
[10] Manenti F. Natural gas operations: considerations on process transients, de-
sign, and control. ISA Trans 2012;51(2):317.
[11] Chang A, Pashikanti K, Liu YA. Refinery engineering: integrated process
modeling and optimization. Wiley-VCH; 2012.
[12] Watkins R. Petroleum refinery distillation. Gulf; 1979.
[13] Sloley, A, Atmospheric distillation process fundamental concepts, 1; 2014. p. 1–20.
[14] Tech A. Aspen HYSYS operation guide. Aspen Technology Inc.; 2008.
[15] Bonell S, Oil and biotechnology: analysis of the state of the art and trends,
ACIMED [online] 19 (1); 2009. p. 0–0.
[16] Jechura J. Crude tower simulation - hysys v8.6. Tech. rep. Colorado School of
Mines; 2014.
[17] Wauquier J. El refino del petróleo: petróleo crudo, productos petrolíferos, es-
quemas de fabricación. Díaz de Santos; 2004.
[18] Fahim M, Al-Sahhaf T, Elkilani A. Fundamentals of petroleum refining. . Else-
vier Science; 2009.
[19] Rodrigo R, Velásquez A. Destilación atmosférica y al vacío en la industria del
petróleo. Igeniator 2013;3(5):51–84.
[20] Young BR, Tellez R, Westphalen DL. Towards integrated process and control
system synthesis for heat-integrated plants. Can J Chem Eng 2006;84(2):219–29.
[21] Young BR, Westphalen DL, Svrcek WY. Heat exchanger network dynamic
analysis. Asian Pacific Journal of Chemical Engineering Mineral Process 14 (3).
[22] Tech A. Aspen HYSYS dynamic modeling. Aspen Technology Inc.; 2008.
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584 583
12. [23] Towler G, Sinnott R. Chemical engineering design. Principles, practice and
economomics of plant and process design. Elsevier; 2008.
[24] Sotelo D. Análisis, diseño e implementación de una estructura de control que
garantiza calidad en los subproductos de una planta de destilación de crudo
en presencia de variaciones en su composición a la entrada. Tech. rep. ITESM;
2015.
[25] Fruehauf PS, Mahoney DP. Improve distillation-column control design. Chem
Eng Prog 1994;90(2):75–83.
[26] Acevedo J. Instrumentación y control avanzado de procesos. Editorial Díaz de
Santos; 2006.
[27] Cole JD, Yount KB. Applications of dynamic simulation to industrial control
problems. ISA Trans 1994;33(1):11–8.
[28] Luyben WL. Evaluation of criteria for selecting temperature trays in distillation
columns. J Process Control 2006;16(1):115–34.
[29] Salahshoor K, Mirzae A. Plant-wide control of a gas plant using a structured
method. Chin Control Decis Conf 2010;1(1):182–9.
[30] Bristol E. On a new measure of interaction for multivariable process control.
IEEE Trans Autom Control 1966;11(1):133–4.
[31] Montelongo-Luna JM, Svrcek WY, Young BR. The relative exergy array-a new
measure for interactions in process design and control. Can J Chem Eng
2011;89(3):545–9.
[32] Westphalen DL, Young BR, Svrcek WY. A controllability index for heat ex-
changer networks. Ind Eng Chem Res 2003;42(20):4659–67.
[33] Fruehauf PS, Mahoney DP. Distillation column control design using steady
state models: usefulness and limitations. ISA Trans 1993;32(2):157–75.
[34] Fruehauf PS, Mahoney DP. An integrated approach for distillation column
control design using steady state and dynamic simulation. Tech. rep., Aspen-
Tech technical articles (March 1997).
[35] Luyben WL. Design and control of stacked-column distillation systems. Ind
Eng Chem Res 2014;53(33):13139–45.
[36] Minh VT, Abdul AM. Modeling and control of a multi-component continuous
crude distillation column using labview. Int J Adv Res Electr Electron Instrum
Eng 2014;3(11):12921–7.
[37] Raimondi A, Favela-Contreras A, Beltrán-Carbajal F, Piñon Rubio A, de la Peña
Elizondo JL. Design of an adaptive predictive control strategy for crude oil
atmospheric distillation process. Control Eng Pract 2015;34(1):39–48.
[38] Luyben WL. Distillation design and control using aspen simulation. Wiley;
2013.
[39] Minh VT, Abdul AM. Modeling and control of distillation column in a petro-
leum process. Math Probl Eng 2009;2009(1):259–63.
D. Sotelo et al. / ISA Transactions 71 (2017) 573–584584