Article

Cite This: Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX pubs.acs.org/IECR

Simulation−Optimization-Based Design of Crude Oil Distillation

Systems with Preflash Units
Minerva Ledezma-Martínez,* Megan Jobson, and Robin Smith
Centre for Process Integration, School of Chemical Engineering and Analytical Science, The University of Manchester, Manchester
M13 9PL, U.K.
*
S Supporting Information

ABSTRACT: In energy-intensive crude oil distillation systems, with their dis-

See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

tillation units and heat recovery systems, preflash units can create opportu-
nities to reduce demand for fired heating and increase heat recovery. As holistic,
Downloaded via UNIV ESTADUAL DE CAMPINAS on July 4, 2018 at 23:20:50 (UTC).

systematic design methodologies are not available for design of crude oil
distillation systems with preflash units, this work proposes a simulation-based
design approach, where the objective is to minimize fired heat demand of the
system. The approach exploits interactions between the separation and heat
recovery systems, considering product quality and yield. Rigorous distillation
models in Aspen HYSYS, pinch analysis, and randomized optimization (a genetic
algorithm) are employed, via a MatLab interface, to minimize hot utility demand
by optimizing operating variables and one structural variable, namely, the feed
location of the preflash vapor. A case study demonstrates the approach,
showing that a preflash unit can reduce hot utility demand but that the system
is vulnerable to yield losses.

1. INTRODUCTION to identify cost-effective and energy-efficient design solutions.

Petroleum continues to be crucial in meeting global energy
demand, and crude oil distillation systems continue to play a
central role in petroleum refining. A typical crude oil distillation
system, as illustrated in Figure 1, comprises a preheat train;
crude oil distillation units with side strippers and pump-arounds;
and sometimes preseparation units, such as flash units and pre-
fractionation columns. These systems are energy-intensive: it is
estimated that 7−15% of the crude oil input is consumed in
refinery processes, of which 35−45% is used for crude oil
distillation.1
Preflash units can be useful for facilitating heat recovery within
the system and thus reducing demand for fired heat for crude
oil preheating prior to distillation. The preflash unit carries out
a partial separation: the vapor recovers some low-boiling com-
ponents and some material in the boiling range of light naphtha.
This vapor stream bypasses the fired heater, helping to reduce its
fuel consumption; the vapor may then be mixed with the stream
leaving the furnace or be fed to the main column at a suitable
location.
Global concerns about carbon emissions and pressure on
refining process economics encourage design of crude oil dis-
tillation systems that maximize process yield and minimize
energy consumption, and therefore also operating costs. Further-
more, the high capital and operating costs of these systems,
together with their considerable complexity, motivate develop-
ment of systematic approaches to develop optimized designs.
Crude oil distillation system design methods employ expe-
rience, trial-and-error, and heuristics, as well as process simu-
lations and heat recovery analyses, typically using pinch analysis,
Design methods that are systematic and employ optimization
effectively continue to be developed, but these have not focused
on the role of preseparation units, such as preflash units and
prefractionation towers.
The aim of this work is to develop a systematic approach to
design cost-effective, energy-efficient crude oil distillation systems
with preflash units, accounting for product quality constraints,
yield, and heat integration. The design methodology is developed
using simulation models in Aspen HYSYS v8.8; these models
are linked to MatLab R2016a through an interface that allows
communication between the two software packages.
2. PREVIOUS RESEARCH STUDIES
Several published studies consider design and optimization of
crude oil distillation systems and how to maximize the benefits
of including preflash units. Golden2 provides useful insights
into how key parameters, such as flash temperature and flashed
vapor feed location, affect the performance of the main distil-
lation column. Ji and Bagajewicz3 present a rigorous approach
for setting the inlet and outlet stream conditions when designing
a crude oil distillation unit with a preflash unit. The preflash
temperature and feed location of preflash vapor in the main
Special Issue: 2017 European Symposium on Computer-Aided
Process Engineering
Received: December 20, 2017
Revised: March 27, 2018
Accepted: April 3, 2018
Published: April 3, 2018

© XXXX American Chemical Society A DOI: 10.1021/acs.iecr.7b05252

Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Figure 1. Crude oil distillation system.
column are addressed, while pinch analysis is applied to assess
minimum utility requirements. Errico et al.4 compare the perfor-
mance of a distillation system with and without a preflash unit,
considering product flow rates, product quality, and potential for
energy savings. A limitation of this study is that column operating
conditions and the preflash temperature are constant.
Wang et al.5 apply thermodynamic metrics to select the best
preseparation scheme for heavy crude oils. Nine predistillation
arrangements are explored, and the option with two preflash
units is found to perform best. Heat recovery is not explicitly
considered, so the results do not relate directly to demand for
fired heating; in addition, product quality specifications appear
to be only partially addressed, via stream or “cut” temperatures.
Benali et al.6 demonstrate that preflash units can bring benefits
in terms of exergy destruction, although their methodology for
adjusting the column operating conditions and for analyzing the
impact of the process changes on heat recovery opportunities is
not discussed.
Other researchers explore using optimization to improve the
performance of the crude oil distillation system using preflash
units. Al-Mayyahi et al.7 utilize multiobjective optimization
techniques to study the effects of single and multiple preflash
units on both energy consumption and yield. Their study inves-
tigates the vapor feed location and considers heat integration
with and without preflash units. The optimization variables
varied include the steam flow to the main column and the furnace
outlet temperature, as well as the vapor fraction from each flash
unit; significantly, pump-around duties and temperature drops are
not considered.
None of the methodologies discussed above provide a system-
atic design methodology for optimization of crude oil distillation
systems with preflash units that accounts for an extensive set of
operating variables, as well as heat integration and product quality
and yield constraints.
3. SIMULATION−OPTIMIZATION-BASED DESIGN
METHODOLOGY
This section introduces the proposed simulation−optimization-
based design methodology for the design of a crude oil distil-
lation system with a preflash unit. First, the Aspen HYSYS
B DOI: 10.1021/acs.iecr.7b05252
Article
simulation model is presented, along with the use of an inter-
face between Aspen HYSYS and MatLab. Second, heat integra-
tion is addressed. Next, the optimization approach is described.
The strong interactions between the crude oil distillation unit,
the preflash unit, and the heat recovery system make this a
challenging optimization problem, especially because both opera-
tional and structural variables are to be optimized.
3.1. Simulation Model and Aspen HYSYS−MatLab
Interface. The crude oil distillation process is modeled using
Aspen HYSYS; in such commercial simulation software, models
for crude oil characterization are well-established in industrial
practice, and rigorous distillation models have demonstrated
their potential to provide highly accurate representations of this
complex process. The models require the crude oil feed, the
process, and the column configuration to be fully defined.
At the design stage, it is appropriate to use heaters and
coolers, rather than heat exchangers, because this simplifies
process simulation and also because it allows the details of the
heat recovery system to be decoupled from process design.
Nevertheless, the minimum heating and cooling requirements
of a proposed design can be readily “targeted” using pinch anal-
ysis; this simplifies process simulation while still allowing evalu-
ation of utility demand.
A simulation file is created for two configurations: that without
and that with a preflash unit. The column structure−number of
stages in each section, number and location of pump-arounds,
draw and return stages for all side-strippers−is identical in both
cases, as are the feed, operating pressures, and product specifi-
cations. This work follows the approach in related studies8−10
by expressing the product quality in terms of points on the
boiling profile, namely, the temperatures at which 5 vol % (T5)
and 95 vol % (T95) of each product are vaporized, according to
the ASTM Standard D86.
For the configuration with a preflash unit, heating of the crude
oil is modeled using two heaters; one represents the heating
upstream of the preflash unit, and the other represents the pre-
heating of the flash liquid by heat recovery and fired heating. The
vapor leaving the flash unit is divided, using a stream splitter, into
five streams, each of which is connected to a different stage of the
main column, where there is one feed stage per section. The split
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
fractions are defined in MatLab such that all but one of these
five streams will have zero flow, i.e., that all the material is
directed to a single feed location. The extract of the MatLab code
included in the Supporting Information provides further detail.
The approach builds on the simulation−optimization tech-
nique of Caballero et al.11 for the design of distillation columns,
facilitated by an Aspen HYSYS−MatLab interface. The simula-
tion model in Aspen HYSYS is linked to MatLab R2016a via an
“automation” interface that allows the user to send inputs to
and collect outputs from the simulation software.12 Spreadsheets
within Aspen HYSYS are found to be very useful for viewing
and storing results and for facilitating data transfer between
Aspen HYSYS and MatLab. For example, HYSYS spreadsheets
are useful for storing the current value of the objective function
and values of other variables that need to meet specified con-
straints (e.g., product quality specifications).
The column simulation is set up to allow its convergence.
In addition, a set of variables is defined, corresponding to the
design degrees of freedom. In this work, the variables are cooling
duty and temperature drop of each pump-around, flow rate of
stripping steam to all steam-stripped column sections, reflux
ratio, column feed inlet temperature, and preflash feed tem-
perature. One structural variable, the feed location in the main
column to which the flash vapor is sent, is included as a degree
of freedom.
For the case with a preflash unit, the flash temperature is an
important degree of freedom. Higher temperatures allow more
vaporization of the crude oil feed and thus more of the feed to
bypass the fired heater; heating only the preflash liquid reduces
the fired heating duty. On the other hand, if too great a fraction
of the crude oil feed bypasses the fired heater, there is a risk
that the total enthalpy of the two feed streams is too low to
allow the desired separation to be achieved.
The vapor leaving the flash unit is sent to a stream splitter
which facilitates the vapor stream to be sent to any one of
several potential feed locations. Initially, it is assumed that the
vapor is sent to the same stage as the main feed, i.e., is effec-
tively mixed with the crude oil leaving the fired heater.
Product quality specifications aim to ensure that the product
streams meet the requirements of downstream processing and
of the market. Some of these specifications can be defined
(as “specifications”) within the rigorous distillation model, but
the limited number of degrees of freedom of the column means
that not all products can be fully specified. Therefore, the
remaining product specifications are defined within MatLab as
inequality constraints, where compliance with specifications is
checked and, in the optimization, a penalty term is added to the
objective function. In line with industrial practice, where spec-
ifications are defined in terms of characteristic boiling temper-
atures within a tolerance ε (typically ±10 °C), the constraints
are defined accordingly.
When using the “automation” feature for direct simulation-
based optimization, there is a risk that the rigorous simulation
will not converge, either because the simulation has been poorly
initialized or because, for a particular set of inputs, the specifi-
cations cannot be met. If the simulation does not converge, the
optimization may not be able to proceed, or taking steps to
facilitate convergence, such as reinitializing the simulation or
increasing the number of iterations, may be computationally
intensive. Therefore, the automation code instructs the Aspen
HYSYS simulation to stop if it does not converge within the
specified maximum number of iterations. In the case that the
Aspen HYSYS simulation does not converge, a penalty is applied
C DOI: 10.1021/acs.iecr.7b05252
Article
to the objective function within MatLab. This penalty helps to
reject spurious results during the optimization.
3.2. Heat Integration. In practice, heat integration is
crucial for energy-efficient operation of crude oil distillation
systems. Heat integration, the recovery of heat between streams
requiring cooling and streams requiring heating, reduces the
need for hot and cold utilities. In this system, the crude oil feed
needs to be heated, usually from ambient conditions, to the
temperature of the desalter and then to the temperature of the
column feed; this “cold stream” is heated by other “hot” streams
in the system that require cooling; these include pump-arounds
and product streams.8
This work applies “pinch technology” to evaluate the mini-
mum heating and cooling requirements, after all opportunities
for heat recovery have been exploited.13 In particular, the grand
composite curve is used to assess the minimum hot utility
demand; in these systems, the temperatures at which heat is
needed require fired heating. This approach allows minimum
utility requirements to be calculated for each converged simu-
lation, without requiring more detailed analysis and design of
the heat recovery system. The disadvantage of this approach is
that the investment costs and complexity of the heat recovery
system that could achieve these utility targets are not considered.
In the context of the optimization of this complex distillation
system, designing a heat exchanger network for each possible
column design would require very significant computational
effort. Therefore, the design of the heat exchanger network is
not addressed in this work. Previous research14 proposes a relevant
approach.
Pinch analysis is carried out within MatLab R2016a using an
open source algorithm,15 where relevant stream data (supply
and target temperatures and duties) are extracted from Aspen
HYSYS for each proposed design.
This study is limited to the case in which the column
design−the number of sections and number of stages in each
section−is fixed, and to a given crude oil feed and a given set of
products, with associated quality specifications. The constraints
on product quality partially fix the distribution of the crude oil
into the various products; the option of including constraints
on product quantity, as well as quality, is also explored in this
work. The methodology focuses on reducing the fired heating
demand of the system, without compromising the quality and
quantity of products. Therefore, the objective function is the
minimum hot utility demand, calculated using pinch analysis,
where penalty terms ensure that the optimal solution meets
product specifications and is also successfully converged when
simulated.
3.3. System Optimization. Experience showed that deter-
ministic nonlinear optimization techniques (such as f mincon in
MatLab) for optimizing the operating variables frequently led
to the optimization reaching a local optimum. Furthermore, the
vapor feed location introduces an integer variable into the
problem, which cannot be handled effectively by such an algo-
rithm. Therefore, randomized search methods, also known as
stochastic optimization or metaheuristics, are applied. These
methods are known to help to overcome the limitations of non-
randomized methods16 by “learning” about the problem during
the optimization and tailoring the search strategy accordingly.17
However, it is well-known that such algorithms find “near-
optimal” solutions, rather than the globally optimal solution.17
Two options from MatLab R2016a Global Optimization
Toolbox18 were tested, both simulated annealing (simulannealbnd)
and a genetic algorithm (gaoptimset). The latter was found to
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
be far more robust in reaching good solutions and therefore is
adopted.
The application of the genetic algorithm for optimization is
described as follows and is illustrated in Figure 2:
Figure 2. Overview of genetic algorithm.
(1) The search space is defined. In particular, upper and
lower bounds are defined for the optimization variables.
(2) The optimization parameters are selected, namely, the size
of the initial population size and the number of generations.
(3) The initial generation is created randomly by the
gaoptimiset algorithm.
(4) The algorithm evaluates the objective function for all indi-
viduals in the population, applying penalties as required.
(5) After evaluating the “fitness” of individuals, the algorithm
selects a subset of best-performing individuals.
(6) The algorithm populates the next generation using genetic
operators (crossover and mutation), applying the default
parameters; the objective function is evaluated again for
all individuals.
(7) If a termination condition is met (e.g., maximum number
of generations, or a very small change in the average
value of the objective function over a certain number of
generations), the solution corresponding to the “best”
solution is retrieved. This solution, including the corre-
sponding set of optimized variables, represents the best-
performing design, i.e., with lowest hot utility demand.
The objective function can be expressed mathematically as
n
minF(x) = f (x) + γ1|h(x)| + ∑ γj[max(0, gj(x))]2
j=1
h(x) = 0
g j (x ) ≤ 0
(1)
where x is the vector of optimization variables (i.e., cooling
duty and temperature drop of all pump-arounds, flow rate of
steam to the main column and to any other steam-stripped
sections, the feed temperature, and preflash temperature, if
applicable); F(x) is the overall objective function; f(x) is the
objective function before applying any penalty terms; h(x)
represents the set of equality constraints; gj(x) represents
inequality constraints, and γj is a set of scalar penalty factors
that scale the penalty according to the significance of the
constraint and the magnitude of the violation of the constraint.9
D DOI: 10.1021/acs.iecr.7b05252
Article
Product quality constraints are included as inequality con-
straints gj(x) in the objective function as follows:
T 5*i − ε ≤ T 5i ≤ T 5*i + ε i = 1, 2, ..., Nproducts (2)
T 95*i − ε ≤ T 95i ≤ T 95*i + ε i = 1, 2, ..., Nproducts
(3)
where T5i and T95i correspond to the T5 and T95 ASTM
Standard D86 temperatures for product i, where the lower
bound is ε less than the specified temperatures (T5i* and
T95i*, respectively) and the upper bound is ε greater than the
specified temperature. These specifications are conveniently
represented using the max function in MatLab:
n
g1(x) = ∑ [max(0, (T 5*i − ε) − T 5i )2 ]
i=1
+ [max(0, T 5i − (T 5*i + ε))2 ] (4)
n
g 2 (x ) = ∑ [max(0, (T 95*i − ε) − T 95i )2 ]
i=1
+ [max(0, T 95i − (T 95*i + ε))2 ] (5)
Each squared term ensures that a positive penalty is incurred
only when the inequality constraint is violated.19
In case 2, a third inequality constraint is added to ensure that
the volumetric flow rate of the atmospheric residue, mRES, is no
greater than that in the base case, m0RES:
0
g3(x) = max(0, mRES − mRES )
(6)
The rigorous simulations, heat recovery analysis, and process
optimization, applying a genetic algorithm, are combined in the
framework represented in Figure 3. Aspen HYSYS v8.8 is
applied for flowsheet simulation, while MatLab R2016a is used
to carry out pinch analysis on proposed solutions and to drive
the optimization. The approach is demonstrated in a case study
in the following section.
4. CASE STUDY
This case study aims to demonstrate the capabilities of the
proposed design methodology and to illustrate that a preflash
unit can bring significant energy savings, even when product
flow rates are constrained.
4.1. Case Study Description. The case study is based on
the data presented by Watkins.20 The Supporting Information
provides details of the crude oil assay in Tables S1 and S2,
design specifications (Table S3), column structure (Table S4),
base case product properties and flow rates (Table S5), and
base case stream data (Table S6). The oil characterization tool
in Aspen HYSYS v8.8 is used to “cut” the oil into a set of 25
pseudocomponents and to calculate the physical and ther-
modynamic properties of each pseudocomponent (e.g., molec-
ular weight, density, and viscosity). These pseudocomponents,
together with the six real low-boiling components, represent
the crude oil mixture. The Peng−Robinson equation of state is
used to simulate the mixture.
The crude oil distillation system comprises an atmospheric
distillation unit with a condenser, three pump-arounds, one
steam-injected side stripper, and two reboiled side strippers.
The case study addresses the system with and without a preflash
unit upstream of the column. The main column operates at a
uniform pressure of 2.5 bar and has 41 theoretical stages,
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Figure 3. General optimization-based design methodology.
Figure 4. Crude oil distillation system with a preflash unit.
distributed in five sections, and numbered from top to bottom.
Figure 4 illustrates the crude oil distillation system, showing
details of the section numbers, distribution of stages in the main
column, locations of pump-arounds and of draw and return
streams, and stripping steam feed locations. Table S4 presents
the distribution of trays in the main column and side-strippers;
note that the numbering continues from 42 to 59 for stages
in the side strippers. Figure 5 illustrates how the flowsheet is
implemented in Aspen HYSYS; in particular, it shows how the
flash vapor stream may be directed to five different feed loca-
tions in the main column and how several spreadsheets are used
to facilitate data transfer to and from MatLab.
Two cases are defined: case 1, where only product quality
constraints are applied, and case 2, where both product quality
and product flow rate constraints are applied. For both cases,
the options of no preflash and of having a preflash are opti-
mized and compared.
E DOI: 10.1021/acs.iecr.7b05252
Article
The system processes 100 000 bbl day−1 (662.4 m3 h−1) of
Venezuelan Tia Juana light crude oil. The crude oil distillation
column produces five products: light naphtha (LN), heavy
naphtha (HN), light distillate (LD), heavy distillate (HD), and
residue (RES). The unoptimized base case design is derived from
a study by Chen.10 Table S6 in the Supporting Information
provides details about the process stream data for the unopti-
mized base case (with no preflash unit), and Figure S1 shows
its grand composite curve and minimum utility requirements.
Vapor leaving the preflash unit is initially mixed with the stream
leaving the fired heater; the mixture is sent to the feed stage in
the main column. Steam is utilized as a stripping agent in the
main column and in the HD stripper. The HN and LD strippers
use reboilers, rather than live steam. Product specifications are
expressed in terms of ASTM T5 and T95 (in °C).
4.2. Operating Conditions. The crude oil distillation sys-
tem has 11 operational variables and one structural variable
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Figure 5. Screenshot of Aspen HYSYS simulation.
Figure 6. Operational and structural variables for crude oil distillation system with a preflash unit.
(vapor feed location), as shown in Figure 6. The base case
operating conditions and product specifications are listed in
detail in Table S3 in the Supporting Information; the vapor
feed location is selected to be the main feed stage, in section 5
of the main column. The preflash temperature and vapor feed
location are two important operating variables that significantly
influence the performance of the crude oil distillation system.
Product quality specifications and base case product flow rates
(expressed in cubic meters per hour and kilomoles per hour)
are presented in Table S5 in the Supporting Information. The
study assumes a minimum temperature approach of 30 °C in all
heat exchangers for the heat recovery calculations.
Prior to process optimization, the initial case forms the
basis for sensitivity studies. These studies facilitate under-
standing of the system and performance trends, in terms of hot
utility demand, as each design variable is changed. The results
of the sensitivity studies help to define suitable bounds for
the optimization, considering the effect of design variables
on performance and on ease of convergence of the flowsheet
simulation.
F DOI: 10.1021/acs.iecr.7b05252
Article
4.3. Optimization Parameters. The aim of the optimiza-
tion is to identify a flowsheet configuration and corresponding
set of operating conditions that minimize the hot utility demand.
The optimization is carried out as described in section 3. The
parameters selected for the genetic algorithm18 are 500 for the
maximum number of generations and 100 for the initial popu-
lation; the “function tolerance”, the average change in value of
the objective required before the optimization is terminated, is
set to 1 × 10−10. These optimization parameters were selected
after running the optimization several times with different values
for population size and number of generations and considering
the best value of the objective function and the effect on the
computation time.
Three optimization runs are performed for each case. Multiple
runs help to give confidence that the optimization is effective and
robust; wide variations in the value of the objective function
could suggest that the optimization algorithm and/or parameters
are unsuitable.
4.4. Optimization Results. Two cases are considered:
case 1, where constraints relate to product quality only, and
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research Article

case 2, where both quality of products and the flow rate of the Table 2 presents results for case 1 related to product quality
atmospheric residue (RES) are constrained. Details of the opti- in terms of ASTM T5 and T95 (in °C). It may be seen that the
mization results are provided in Tables S7 and S9 (for case 1)
and Tables S8 and S10 (for case 2). Optimization runs took Table 2. Product Quality of Optimized System: Case 1
between 4 and 6 h of CPU time on an HP desktop PC with
Intel Core i5 processor running at 3.20 GHz and 16 GB of optimization optimization
base case results no preflash results with preflash
RAM. ASTM (°C) ASTM (°C) ASTM (°C)
4.4.1. Case 1: Crude Oil Distillation System without and product T5% T95% T5% T95% T5% T95%
with a Preflash Unit, Where Constraints Relate to Product
LN 27 110a 25 110 25 110
Quality Only. A summary of optimization results and bounds
HN 143 196a 133 196 133 196
for operational and structural variables obtained for the crude
LD 218a 300a 218 300 218 300
oil distillation system with and without a preflash unit is pre-
HD 308 354a 305 354 302 354
sented in Table 1. (Note that pump-around duties are negative
RES 363 755 354 753 353 753
a
Table 1. Optimization Results and Bounds for the Crude Oil Specified in HYSYS.
Distillation System: Case 1
product quality constraints are all met within the allowed range
optimization of temperatures (±10 °C).
results
Table 3 shows the results for the optimized product flow
base lower upper no with rates and vapor for case 1. These results confirm that the
variablea,b units case bound bound preflash preflash
main steam kmol h−1 1200 900 1800 900 1190
flow rate Table 3. Product and Vapor Flow Rates: Case 1
−1
HD steam kmol h 250 200 375 200 211 flow rate base optimized no optimized with
flow rate (m3 h−1) case preflash preflash
PA1 duty MW 12.8 6 14 6.0 6.8
LN 102 101 101
PA2 duty MW 17.8 6 18 9.8 8.3
HN 87 88 89
PA3 duty MW 11.2 6 12 11.9 8.8
LD 128 126 123
PA1 ΔT °C 30 20 50 48.6 25.5
HD 54 48 49
PA2 ΔT °C 50 15 60 33.8 23.0
RES 292 299 301
PA3 ΔT °C 20 10 40 39.9 39.3
column inlet °C 365 350 385 350 354 product yields for most of the valuable products (LN, HN, and
temperature
flash °C 115 110 230 − 230
HD) change relatively little, which is a consequence of the
temperature product quality being constrained and therefore the distribution
reflux ratio 4.17 3.0 4.5 3.7 3.0 of the crude oil feed into products being constrained. However,
vapor feed 5 1 5 − 3 the 2.4% and 3% increases in the flow rate of the atmospheric
locationc residue (RES), compared to the unoptimized base case, repre-
minimum hot MW 58.3 43.4 35.9 sents a loss of more valuable products. Therefore, case 2
utility
addresses this problem by adding a constraint on the flow rate
a
PA: pump-around. bΔT: pump-around temperature drop. cNumber of the atmospheric residue, in line with previous studies.9
of section in main column. 4.4.2. Case 2: Crude Oil Distillation System with Constraints
on Both Product Quality and Product Flow Rates. Again, the
values in HYSYS; the MatLab code takes this into account crude oil distillation system is optimized without and with a pre-
when defining upper and lower bounds on the duties.) flash unit. The optimization results for case 2 are summarized in
Table 1 shows that the initial conditions of the base case Tables 4−6, providing details of operating conditions and flow-
were far from optimal. More importantly, Table 1 shows that sheet structure, product quality, and product and vapor flow rates.
the minimum hot utility demand could be reduced by 17% by As shown in Table 4, in case 2 the use of a preflash unit again
introducing a preflash unit to the flowsheet. It is also note- leads to a significantly lower minimum hot utility demand
worthy that the optimizer selected section 3 for the feed location (18%). However, for both configurations, without and with a
of the preflash vapor. (This result was obtained in all successful preflash unit, when the residue flow rate is constrained, the
optimizations.) It was observed that location for the vapor feed minimum hot utility demand increases significantly, by 6% and
was always chosen to be in the section with the temperature 5%, compared to the optimized flowsheets without product
closest to the preflash temperature. For the solution shown in flow rate constraints (i.e., case 1). This increased demand
Table 1, the optimized feed location corresponds to stage 18 in for fired heating is consistent with the higher column inlet
section 3 of the main column. It may be observed that the temperature of case 2 (362 and 383 °C), compared to case 1
column inlet temperature is 4 °C higher in the case that the (350 and 354 °C). The increased column feed temperatures
flash vapor bypasses the furnace; this result indicates that, with achieve greater vaporization of this stream, compensating for
a preflash, the crude oil feed needs to be hotter, to compensate both an increased vapor fraction entering the column at a
for the lower enthalpy of the preflash vapor when it enters the relatively low temperature (240 °C) and the need to vaporize
column. Table 1 also shows that the optimizer maximizes the more of the feed in order to meet the flow rate constraint on
temperature of the flash unit (to the upper bound of 230 °C), the atmospheric residue. Consequently, and aligned with pre-
indicating that the benefits of allowing some material to bypass vious work,9 demand for fired heating increases.
the fired heater outweigh the drawbacks of feeding relatively It is noted that, as in case 1, the flash temperature selected by
cold vapor to the column. the optimizer is at the upper bound of the range (240 °C),
G DOI: 10.1021/acs.iecr.7b05252
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research Article

Table 4. Optimization Results and Bounds for the Crude Oil preflash, i.e., allowing a fraction of the feed to bypass the fired
Distillation System: Case 2 heater, tends to reduce the amount of high-temperature heat
recovered. In particular, heat recovery from pump-arounds
optimization
results 3 and 2 is reduced in favor of rejecting lower-grade heat from
pump-around 1. This trend would also reduce the temperature
base lower upper no with
variablea,b units case bound bound preflash preflash to which the heat recovery system could preheat the crude oil
feed before it enters the fired heater.
main steam kmol h−1 1200 900 1800 1247 1195
flow rate In Table 5 it may be seen that the product quality constraints
HD steam kmol h−1 250 180 375 188 180 are all met within the allowed range of temperatures (±10 °C),
flow rate as are those in case 1. Table 6 confirms that the residue flow
PA1 duty MW 12.8 6 14 7.4 8.5 rate is unchanged, indicating that there is no loss of valuable
PA2 duty MW 17.8 6 18 9.8 8.8 products.
PA3 duty MW 11.2 6 12 14.0 13.0 Figure 7 presents the grand composite curves for the two
PA1 ΔT °C 30 20 50 31.9 38.8 optimized cases with a preflash unit, where case 1 considers
PA2 ΔT °C 50 15 60 34.9 31.6 product quality constraints and case 2 considers both product
PA3 ΔT °C 20 10 40 39.9 21.0 quality and product flow rate. The grand composite curves
column inlet °C 365 350 385 362 383 show that the minimum hot utility demand increases when
temperature
product flow rates are considered. In both cases, the existence
flash °C 115 110 240 − 240
temperature of several pinches (or near-pinches) indicates that heat recovery
reflux ratio 4.17 3.0 4.5 4.1 3.1 is maximized at a wide range of temperature levels: that is, the
vapor feed 5 1 5 − 3 column operating conditions have been optimized to maximize
locationc heat recovery. This is a strength of the methodology, where
minimum hot MW 58.3 46.6 37.9 each proposed solution is assessed in terms of its net heating
utility demand, after heat recovery, rather than by considering heat
a
PA: pump-around. bΔT: pump-around temperature drop. cNumber recovery only after the design of the system. The results in
of section in main column. Figure 7 and Tables 1 and 4 highlight the trade-offs between
the fired heating demand and the yield of valuable products.
Table 5. Product Quality of Optimized System: Case 2 Table S8 provides stream data for the optimized system applying
optimization optimization the constraint on the atmospheric residue flow rate.
base case results no preflash results with preflash
ASTM (°C) ASTM (°C) ASTM (°C)
5. CONCLUSIONS
product T5% T95% T5% T95% T5% T95%
A systematic design optimization approach is proposed for crude
LN 27 110a 25 110 25 110 oil distillation systems with a preflash unit, applying a rigorous
HN 143 196a 133 196 135 196 simulation model and using pinch analysis to determine the
LD 218a 300a 218 300 218 300 minimum hot utility demand of the heat-integrated system.
HD 308 354a 307 354 303 354 The methodology accounts for industrially relevant constraints
RES 363 755 361 754 362 754 related to product quality and yield and the main operating cost
a
Specified in HYSYS. of the distillation system, that of fired heating. Especially
because of the rigorous simulations involved, the optimization
Table 6. Product and Vapor Flow Rates: Case 2 is relatively computationally intensive, requiring 4 to 6 h of
flow rate base optimized no optimized with CPU time per optimization run.
(m3 h−1) case preflash preflash The case study confirms that introducing a preflash unit,
LN 102 101 101 while also optimizing the column operating conditions, can bring
HN 87 88 89 significant improvements in the hot utility demand. However,
LD 128 126 123 the preflash clearly impacts on the separation performance of
HD 54 55 58 the column, and thus potentially could reduce the value added
RES 292 292 292 by the distillation system to the crude oil, for example, if more
of the crude oil leaves the column in the residue stream. These
indicating that there are benefits for the heat recovery system results highlight the trade-off between yield and energy
(and few penalties for the separation performance) of using a demand, and suggest that the objective function should capture
preflash unit. It is also observed that the stream flow rate to the both aspects, for example, as net profit, as in previous studies.8
HD side-stripper is at the minimum value; a lower requirement The problem formulation described in this manuscript keeps
for stripping steam is consistent with the removal of lighter the design of the distillation column (number of stages in each
material from the column feed. section) fixed in all cases. Future work will aim to address this
Again, the flash vapor is directed to the section of the column limitation; the change in flow rate and composition of the
with the temperature that is most similar to that of the vapor. feeds after introducing a preflash unit, would logically require a
This result implies that the optimization search space could be different distribution and/or number of stages in the distillation
narrowed appropriately, with corresponding reductions in compu- unit, and would also significantly affect the cost of the column
tation time. because of changes in required column diameter. Furthermore,
The greater duties of the pump-arounds in case 2 indicate future work will take into account column design, together
that the higher feed temperature also allows more heat to be with the use of preseparation units, both preflash and prefrac-
recovered in the heat recovery system. It may be observed from tionation units, taking advantage of recent developments in this
Tables 1 and 4 that, in both cases 1 and 2, introducing a area.9
H DOI: 10.1021/acs.iecr.7b05252
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Figure 7. Grand composite curves of optimized crude oil distillation system with preflash.
Other limitations of the work are the use of pinch analysis,
rather than addressing detailed aspects of design and costing of
the heat recovery system. Also, the relatively high computa-
tional demand of this study, which could certainly be reduced
by using software that is more computationally efficient than
MatLab, points to the possibility of adapting recent develop-
ments in the use of surrogate models for distillation system
optimization.8,9
■
ASSOCIATED CONTENT
*
S Supporting Information
The Supporting Information is available free of charge on the
ACS Publications website at DOI: 10.1021/acs.iecr.7b05252.
Data on crude oil properties and characterization, distri-
bution of theoretical stages in the main column and side
strippers, base case operating conditions, base case product
quality specifications and flow rates, process stream data
for the base case and two optimized cases and grand
composite curve with utility targets, and optimization
results for the case study (PDF)
■
AUTHOR INFORMATION
Corresponding Author
*E-mail: minerva.ledezmamartinez@manchester.ac.uk.
ORCID
Minerva Ledezma-Martínez: 0000-0003-2929-7808
Megan Jobson: 0000-0001-9626-5879
Robin Smith: 0000-0003-1832-8592
Notes
The authors declare no competing financial interest.
■
ACKNOWLEDGMENTS
The authors acknowledge the financial support for the devel-
opment of this work from the National Council of Science and
Technology (CONACyT) Mexico, Grant Number 381926.
■
REFERENCES
(1) Szklo, A.; Schaeffer, R. Fuel Specification, Energy Consumption
and CO2 Emission in Oil Refineries. Energy 2007, 32, 1075−1092.
I DOI: 10.1021/acs.iecr.7b05252
Article
(2) Golden, S. W. Prevent Pre-distillation Drum Foaming. Hydro-
carbon Process. 1997, 76, 141−153.
(3) Ji, S.; Bagajewicz, M. Design of Crude Fractionation Units with
Preflashing or Prefractionation: Energy Targeting. Ind. Eng. Chem. Res.
2002, 41, 3003−3011.
(4) Errico, M.; Tola, G.; Mascia, M. Energy Saving in a Crude
Distillation Unit by a Preflash Implementation. Appl. Therm. Eng.
2009, 29, 1642−1647.
(5) Wang, Y.; Hou, Y.; Gao, H.; Sun, J.; Xu, S. Selecting the
Optimum Predistillation Scheme for Heavy Crude Oils. Ind. Eng.
Chem. Res. 2011, 50, 10549−10556.
(6) Benali, T.; Tondeur, D.; Jaubert, J. N. An Improved Crude Oil
Atmospheric Distillation Process for Energy Integration: Part II: New
Approach for Energy Saving by Use of Residual Heat. Appl. Therm.
Eng. 2012, 40, 132−144.
(7) Al-Mayyahi, M. A.; Hoadley, A. F. A.; Rangaiah, G. P. Energy
Optimization of Crude Oil Distillation Using Different Designs of Pre-
Flash Drums. Appl. Therm. Eng. 2014, 73, 1204−1210.
(8) Ochoa-Estopier, L. M.; Jobson, M. Optimization of Heat-
Integrated Crude Oil Distillation Systems. Part I: The Distillation
Model. Ind. Eng. Chem. Res. 2015, 54, 4988−5000.
(9) Ibrahim, D.; Jobson, M.; Guillén-Gosálbez, G. Optimization-
Based Design of Crude Oil Distillation Units Using Rigorous
Simulation Models. Ind. Eng. Chem. Res. 2017, 56, 6728−6740.
(10) Chen, L. Heat-integrated Crude Oil Distillation System Design.
Ph.D. Thesis, The University of Manchester, Manchester, U.K., 2008.
(11) Caballero, J. A.; Milán-Yañez, D.; Grossmann, I. E. Rigorous
Design of Distillation Columns: Integration of Disjunctive Program-
ming and Process Simulators. Ind. Eng. Chem. Res. 2005, 44, 6760−
6775.
(12) Aspen HYSYS Customization Guide, v2006.5; Aspen
Technology Inc.: Burlington, MA, 2007.
(13) Smith, R. Chemical Process: Design and Integration, 2nd ed.;
Wiley: Chichester, U.K., 2016.
(14) Smith, R.; Jobson, M.; Chen, L. Recent development in the
retrofit of heat exchanger networks. Appl. Therm. Eng. 2010, 30, 2281−
2289.
(15) Morandin, M. Pinch Analysis cascade calculation, 2014. http://
uk.mathworks.com/matlabcentral/fileexchange/47743-cascade-m (ac-
cessed Nov 27, 2014).
(16) Osman, I. H.; Laporte, G. Metaheuristics: A bibliography.
Annals of Operations Research 1996, 63, 511−623.
(17) Osman, I. H.; Kelly, J. P. Meta-Heuristics: An Overview. In
Meta-Heuristics Theory and Applications; Osman, I. H., Kelly, J. P., Eds.;
Springer: Boston, MA, 1996.
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research Article

(18) MatLab, version R2016a; Global Optimization Toolbox User’s

Guide; The Math Works Inc.: Natick, MA, 2017.
(19) Biegler, L. T. Optimization of Chemical Processes, 2nd ed.; Edgar,
T. F., Himmelblau, D. M., Lasdon, L. S., Eds.; McGraw Hill: New
York, 2003.
(20) Watkins, R. N. Petroleum Refinery Distillation, 2nd ed.; Gulf
Publishing Company: Houston, TX, 1979.

J DOI: 10.1021/acs.iecr.7b05252
Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX