Accepted Manuscript

Modeling and optimization of process fired heaters

Mojtaba Haratian, Majid Amidpour, Aliasghar Hamidi

PII: S1359-4311(17)37227-7
DOI: https://doi.org/10.1016/j.applthermaleng.2019.113722
Article Number: 113722
Reference: ATE 113722

To appear in: Applied Thermal Engineering

Received Date: 21 November 2017

Revised Date: 22 April 2019
Accepted Date: 3 May 2019

Please cite this article as: M. Haratian, M. Amidpour, A. Hamidi, Modeling and optimization of process fired heaters,
Applied Thermal Engineering (2019), doi: https://doi.org/10.1016/j.applthermaleng.2019.113722

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Modeling and optimization of process fired heaters

Mojtaba Haratian 1, Majid Amidpour2*, Aliasghar Hamidi 3

1- Department of Natural Resources and Environment, Science and Research Branch, Islamic Azad University, Tehran, Iran
2- Visiting professor, Department of Energy, Politecnico di Milano, Milano, Italy
3- Department of Chemical Engineering, Tehran University , Tehran, Iran

Abstract

In this study, a mathematical model associated with genetic algorithms (GAs) is developed to optimize the
design of process fired heaters where the objective function is the total annual cost. The mathematical model
considering operational and geometric constraints is used to design all subsections of a furnace including the
radiant chamber, the convection and stack sections. The proposed model allows the economic optimization using
the MATLAB genetic algorithm toolbox. This procedure is implemented to design and optimize fired heaters.
Two case studies are considered. First, by comparing the results with literature, it is shown that the developed
model can be successfully used in design with acceptable accuracy. In the first case study, based on TAC (total
annual cost) minimization, two optimization scenarios with two sets of decision variables using GA are applied
to determine the optimal economic design. It is demonstrated that by considering the tube diameter as a decision
variable (scenario B), the TAC minimization approach does not present a realistic optimal design, because the
crude oil pressure drop is out of a permissible range. Finally, by including the pumping cost into the operating
cost and modifying the objective function, another optimization approach based on MTAC minimization is
applied for the same optimization scenarios. In this case, scenario B results in the economic/realistic optimal
design by up to 2.48% cheaper than Original Design.
Keywords: Process fired heaters, Mathematical model, Economic optimization, Genetic algorithm, Modified
objective function.

1. Introduction

Fired heaters are widely used as hot utility systems in the petrochemical and oil refining industries. In a
fired heater, the process fluid flowing inside the tubes is heated by hot gases produced by the combustion of fuel.
The capacity of fired heaters ranges from 3 to 100 MW [1]. There are many different heater and coil
configuration combinations. Generally, the box heater type is used for large heat duty applications
(approximately 20 MW and higher) and the vertical cylindrical heater type is used for small and or medium heat
duty applications (below 20 MW) [2].

The implementation and use of several energy saving methods such as heat recovery and air preheating
techniques into the existing fired heaters can considerably improve the thermal and exergy efficiencies which
lead to a substantial reduction in fuel consumption of the units [3]. These approaches are more applicable to
retrofit for improving existing plants.

Masoumi and Izakmehri [4] developed a mathematical model to predict the performance of fired
heaters with variations in the operational conditions. They indicated that the furnace efficiency of a refinery in
Iran can be improved about 26% using combustion air preheating and excess air control methods. Al-Haj
Ibrahim and Al-Qassim [5, 6] simulated and analyzed the heat transfer mechanisms inside the process fired
heaters. Based on computer programming, they determined the effective gas temperature in the radiant section as
well as the temperature profiles and the heat transfer rate per layer in the convection section. In another research,
Al-Haj Ibrahim and Al-Qassim [7] compared direct and indirect methods for the thermal efficiency assessment
of process heaters. They showed that the direct method as an effective procedure is to be preferred for fast
efficiency calculation of process heaters. Gunasegran and Azarpour [8] conducted a numerical modeling and
sensitivity analysis for investigating the effect of the key parameters on the thermal performance of a cylindrical

*
Corresponding author E-mail: [email protected]

1
type of process heaters. By examining influential parameters including excess air, tube pitch, inlet air
temperature, fuel composition and fuel flow rate, they demonstrated that the air percentage is the most effective
parameter on the furnace efficiency. Chaibakhsh et al. [9] developed a mathematical model in Matlab simulink
environment to investigate the thermal performance of fired heaters. The proposed model, considering the
single-phase and two-phase flow patterns, is used to predict the outlet temperature of crude oil at different
operating conditions. Morales-Fuentes et al. [10] developed a fouling rate model for two phase flow conditions
to investigate the thermo-hydraulic performance of process fired heaters. They indicated that fuel consumption
and pressure drop are increased as the fouling builds up in fired heaters. In addition, the effect of operating
conditions such as inlet temperature of crude oil, excess air and steam injection on the fouling rate was
investigated by Morales-Fuentes et al. [11]. However, such developed models are frequently employed to
perform a detail assessment of the thermal behavior of fired heaters but are not very appropriate for optimization
design purposes.

Recently, many different numerical models, based on computational fluid dynamics (CFD) in the field
of design and analysis of process furnaces, have been developed. The CFD models are used for different
applications such as detail analysis of heat flux and temperature distribution on tubes [12, 13] and investigating
the effects of excess air and air preheating on the thermal performance [14]. Moreover, the CFD codes are
applied to evaluate the optimal burner design with respect to the global NOx production [15]. CFD analysis has
also been used to study the effect of replacing refinery fuel gas with hydrogen [16]. The application of CFD
models, especially for optimizing process designs, require several simulations with different arrangements which
definitely result in consuming large computing time and is not therefore very suitable for optimizing
applications.

Most of the practical optimization methods are used for improving process operations far more than for
the optimal design of a furnace. On the optimal design framework, less work has been done. Some commercial
research organizations such as Heat Transfer Research Inc. (HTRI) or Aspen Technology Inc., in recent years
have been developing reliable and optimal design procedures for fired process heaters based on computer
programs [17]. However, their methods are not available in the open literature. Jegla [18], introduced a quick
preliminary design procedure to obtain the optimum geometrical characteristics of the radiant chamber of a
process tubular heater from a total cost point of view. The conceptual design method allows to evaluate the effect
of the main design parameters such as geometrical dimensions of radiant zone and heat flux rate of radiant tubes
on the total annual cost. In another research, Jegla [19] presented a calculation method to determine an optimum
arrangement of the radiant tube coil system of process furnaces based on simplification of standard design
methods (Lobo-Evans and Belokon’s methods). Bahadori and Vuthaluru [20] developed simple-to-use
correlations (based on the analysis of available data in the literature) to design radiant and convection parts of
direct fired heaters. The excellent performance of the proposed procedure was demonstrated through a case
study. Mussati, Manassaldi et al. [21] developed a mathematical optimization model in General Algebraic
Modeling System (GAMS), using Mixed Integer Nonlinear Programming (MINLP) to get an efficient furnace
design. They found that the proposed procedure considering discrete and continuous decision variables and
various objective functions can be used to optimize the fired heater design with acceptable accuracy.
The aim of this work is to produce a new approach associated with genetic algorithms for the economic
optimization of operational and geometrical parameters of fired heaters. The geometrical parameters included
are: the number of passes, dimensions of the firebox (width and tube length), tube diameter. Also, the furnace
efficiency, radiant heat flux and excess air are considered as operating parameters. Furthermore, the effect of
increasing the number of free variables (i.e. degrees of freedom) on the optimal design is investigated. For these
purposes, a mathematical model based on the standard Lobo-Evans method is developed to design all
subsections of fired heaters, including the radiant chamber, convection section and stack. The proposed model
provides a constrained non-linear problem with multiple design variables. Then, the optimization scheme is
implemented by applying genetic algorithms to the model considering operational and geometrical constraints
and two sets of decision variables. GA technique is the most appropriate tool for nonlinear and multivariable
optimization problems. In this model, operational constraints such as allowable pressure drop on the oil side,
furnace draft, excess air and heat flux on tubes are considered for optimization design. Furthermore, the
geometrical constraints including the furnace dimensions, tube diameter, and coil arrangement are considered
based on engineering standards and pertinent relations [21-23]. Moreover, by considering the pumping cost into

2
the operating cost and modifying the objective function, the pumping cost effect is discussed for various
optimization scenarios.

2. Problem description

In this study, a cabin type fired heater with two distinct radiant and convection sections is considered to
heat up the process fluid. The conditions of the process fluid including the flow rate, the inlet/outlet temperatures
and the required heating duty are given. The fuel type or the heating value of the fuel is assumed to be known.
The main goal is to obtain the optimal economic design of the fired heater to provide the necessary heating duty
by minimizing the total annual cost. The total annual cost including the sum of the capital and operating costs is
minimized by means of GA.

In order to simplify the problem, the following assumptions are considered:

 Flue gases are assumed as ideal and gray gas with constant average heat capacity for each section
 The average flue gas temperature and the average heat flux are constant throughout the radiant section.
 The average gas temperature of radiant section is assumed to be equal to the temperature of the flue
gases leaving the radiant section.

Usually, the application of these assumptions is considered to be logical and extremely used for modeling
purposes. So, the following mathematical model can be proposed.

3. Fired heater modeling

In the present work, a simple model for a cabin type fired heater as shown in Fig. 1 is considered.

Flue gas

Stack

(Ts) Process fluid

Inlet (Ti)
Convection
tubes Crossover
Shield tubes (Tc)
Bridge wall (Tg)

Radiant
tubes Outlet
(To)

Air Fuel

Fig. 1. Flow sketch of a cabin type fired heater

Initially, the process fluid is heated through multi-pass tubes in the convection section and is followed by further
heating in the radiant section leaving the radiant tubes with the desired outlet temperature at the bottom segments
of the furnace. As can be seen in Fig. 1, the process fluid with a given flow rate (Mpf) is heated from an inlet
temperature (Ti) to a desired outlet temperature (To). The flue gas stream leaves the radiant section at the bridge
wall temperature (Tg). In most models, it is adequate to assume that the radiant section is a “well stirred
combustion chamber” and the average gas temperature is equal to the bridge wall temperature. In the convection
section, convective heat transfer dominates the heating process. The flue gases leave the convection section at
stack temperature (Ts). Shield section is the name given to the bottom rows of the convection zone. These tubes

3
which shield the remaining convection-section tubes from direct radiation, receive approximately the same heat
flux as the radiant tubes, since the higher convection transfer counteracts the lesser radiant due to lack of
refractory wall backing. The heat loss rate through the furnace walls is normally taken as 2% to 3% [22]. In this
work, it is assumed to be 2.5%. The heat balance for these processes in a fired heater may be expressed by Eq.
(A1).

In the following section, the modeling details for all subsections of a fired heater are presented. The
proposed model uses standard equations based on the Lobo-Evans method which can be found in more detail in
the Appendices A-C [21, 22].

3.1. Radiant section

Heat transfer in the radiant section of a fired heater occurs predominantly through radiation from the
flue gas (about 90%) but also significantly by convection. The combined effect is given by Eq. (A5). For a “well
mixed” radiant zone, the effective gas temperature is assumed to be equal to the bridge wall temperature, i.e. the
temperature of the gas leaving the radiant zone [5] (see Fig. 1). The effective gas temperature, as a key parameter
of operation which depends on the geometric dimensions of the radiation section, is required to perform the heat
transfer analysis. For this purpose, firstly, the radiant zone sizing must be performed. In this way, the geometric
dimensions of the radiant zone including the inside dimensions of the fire box, the number, diameter, and length
of radiant tubes and spacing between tubes are determined. Figure 2 shows a scheme of the tubes and box
configuration of a fired heater. As shown in Fig. 2, the dimensions L, W, and H are the inside dimensions of the
fire box. The formulation of the radiant zone sizing is described in the Appendix A.

Wc

Convection
tubes

Ceiling
tubes

Shield H
tubes

Wall side
tubes

W L

Fig. 2. Configuration of tubes and heating zones of a fired heater

Usually, the number of shield tubes will be the number of convection tubes per row [22]. In this model,
it is assumed that the number of convection tubes per row is equal to N pass, so the number of shield tubes is equal
to the number of the passes. A pass in the process fired heaters is defined as a flow circuit consisting of one or
more tubes in series (i.e. a single circuit from inlet to outlet) [23,24]. Pass layout, tube configuration and burner
arrangement in the fired heaters would be very different. Most cabin type heaters have four or six passes in a
single cell. Figure 3 shows a four-pass fired heater layout, as similar as mentioned in this model. The total
number of tubes in the radiant section is the summation of the number of wall side tubes (Nt wall), ceiling tubes
(Ntceil), and shield tubes (Ntshield). In addition, the number of shield tubes must be pair for assuring a symmetric
distribution of radiant tubes at each side of the radiant section. The arrangement of the burners shall be in
accordance with heater supplier’s standard design to give the most uniform tube wall temperature.

4
 Pass inlet # 3 Pass inlet # 2

Pass inlet # 4 Pass inlet # 1

Convection
Convection tubes (Ntcon)
section
Shield tubes (Ntshield)

Ceiling tubes (Ntceil)

Wall side tubes(Ntwall)

Radiant
section
Pass outlet # 4 Pass outlet # 1

Pass outlet # 3 Pass outlet # 2

Fig. 3. A four-pass fired heat layout

After radiant zone sizing, the effective gas temperature can be calculated by the standard Newton-Raphson
method [5, 21, 22].

3.2. Convection section

The heat transfer area (Acon) for the convection section is computed based on the overall heat transfer
coefficient-LMTD method. The logarithmic mean temperature difference (LMTD) based on the inlet and outlet
temperatures of flue gas and process fluid in convection section is calculated. The overall heat transfer
coefficient in the convection section depends on the total apparent gas film coefficient outside the tubes (hco) and
the convection heat transfer coefficient inside the tubes (hci). The equations of heat transfer in the convection
section are completely presented in the Appendix B.

3.3. Stack design requirement

In a natural-draft process heater, the main function of the stack is the generation of a slightly negative
pressure called draft inside the process heater. The stack height provides the draft required to maintain negative
pressure at arch (just below the convection section) to compensate for pressure losses in the burners (ΔPb),
convection section (ΔPc) and stack (ΔPs). The draft effect of the stack is computed by Eq. (C1). Typically, the
vacuum at arch (Parch) is considered around 0.2 (cm water gauge) [17]. The pressure drop across burners which
is reported around 0.63 (cm water), depends on the type of burner, fuel and flow rate of fuel [25]. The stack
calculations can be found in more detail in the Appendix C.

3.4. Pressure drop

In the hydraulic design of process heaters, the pressure drop of the fluids must be investigated. For flow
outside tubes (gas side), most of the pressure drop will occur in the convection section. The required equations
for estimating pressure drop across the tube bank of the convection section are represented in the stack design

5
section. The pressure drop in the radiant section will be small compared with the convection section and can
usually be neglected [17].

The tube side pressure drop for single-phase process fluid can be estimated by Eq. (B12). In this paper the
maximum allowable pressure drop of process fluid is assumed 1000 kPa [3, 7, 9].

It is noteworthy that the thermal design model based on Lobo-Evans method can be applied in single
and two-phase flow regimes [30]. In the computation of the pressure drop for two-phase flow, the use of Eq.
(B12) is less precise. However, to simplify the calculation of the pressure drop in two-phase flow, the application
of a friction coefficient of 0.0045 in Eq. (B12) for liquid-vapor hydrocarbon systems is highly accurate from an
engineering point of view [31].

4. Design optimization

In this study, design optimization for a typical fired heater with a given duty is considered to find the
best design variables such as geometrical parameters that satisfy the operational and economic criteria. It
involves some stages including a description of the decision variables, objective function, constraints, and
optimization methodology.

4.1. The decision variables, objective function and constraints

To find the optimal design of a fired heater, the total annual cost is defined as the objective function. It
includes the capital costs of heat transfer area and firebox structure as well as the operating cost. To compare the
obtained results to the literature approaches, the operating cost includes only the fuel consumption and operating
time. Afterwards, by modifying the objective function, the effect of pumping cost on the optimal design is
investigated in different optimization scenarios.
In this work, two sets of decision variables as inputs are considered. As the first scenario (case-A),
seven decision variables including the number of passes (Npass), furnace efficiency ( ), radiant duty fraction
(Rdf), radiant heat flux (Flux), width of firebox (W), tube length (La) and mass velocity in stack (Gs) are
considered. By adding two other design variables, including excess air (ex) and outer diameter of tubes (Do), the
second scenario (case-B) is analyzed. It is important to mention that the outer diameter of tubes and the number
of passes are discrete decision variables.

Table 1. Range of permitted variation of the decision variables for optimization

Decision variables Limitations
Npass 2, 4, 6
(%) 75 - 79
Rdf 0.65 - 0.75
Flux (kW/m2) 36 - 38
W (m) 4.5 – 6.5
La (m) 12 - 16
2
Gs (kg/m s) 0.75 - 1
ex (%) 20 - 25
0.073025, 0.0889,
Do (m)
0.1016, 0.1143, 0.127
Table 1 shows the range of permitted variation of the decision variables set with respect to the available
standards and references [21-23]. The upper and lower limits of the decision variables are the optimization
constraints. In addition to these constraints, some geometrical constraints such as Equations A7-A12 and
operational constraints including the permissible range of crude oil pressure drop and furnace draft (see sections
3.3 and 3.4) were taken into account in the mathematical model.
The total annual cost (TAC) function can be written as follows:

(1)

6
where Ccost and Ocost refer to the capital cost and operating cost, respectively. The capital cost, which is the sum
of the costs of radiant coil, convection coil and firebox, is calculated using the following equation:

(2)

where crcost and cccost are the unitary cost of heat transfer area in the radiant and convection section, respectively.
K1 and K2 are given values and they depend on the type of refractory, coil materials, insulation width, among
others [22]. CRF is the capital recovery factor.

The operating cost, which only includes the fuel consumption and operating time, is presented by:

(3)
where ocost and OT are the unitary fuel cost and the plant operating time, respectively. To include the pumping
cost in the operating cost function, the fuel equivalent consumption ( ) of the required pump power is
considered as follows:

(4)

where , and are the total efficiency of pump and motor, the power generation efficiency and the
pressure drop for process fluid flow, respectively. Therefore, the modified operating cost (considering the fuel
consumption and the pumping costs) and the modified total annual cost (MTAC) are computed from the
following equations:

(5)
(6)

4.2. Optimization algorithm

In this study, the design optimization from an economic point-of-view is performed via the application
of genetic algorithms (GAs). This technique is widely used to solve optimization problems, because it makes
easier to search for a global minimum. The GA is based on the evolutionary ideas of natural selection and
genetics, recommended to multivariable problems. In addition, the proposed model provides a constrained non-
linear problem with continuous and discrete design variables. The genetic search approach is highly applicable to
problems with search in a discrete design space (sometimes with a large number of variables). The
appropriateness of the genetic algorithm is due to the fact that the algorithm searches from a population of
variables to another population of variables (in contrast to the point-to-point search present in mathematical
programming methods). Then GA would probably find more accurate solutions. The mutation and crossover
operators will help GA to jump the discontinuity in the search space and lead to better exploration. In addition,
genetic algorithms act on coding of the decision variables rather than the variables themselves. This allows using
a composition of discrete and continuous variables [26, 27].

The optimization procedure is applied as shown in Fig. 4. After defining the number and bound limits
of decision variables and setting the parameters of GA in Matlab optimization toolbox, it begins with an initial
population of individuals, which may be generated in a random way. The individuals (i.e., solutions of the
problem) are identified from the values of the decision variables (see Table 1). The value of the objective
function (total annual cost) for each individual of the population is calculated by applying the mathematical
model of the fired heater (the Equations of Appendices A-C and Equations 1-6). In this way, in each generation,
the fitness of every individual in the population is evaluated. From this step, the GA operators (selection and
crossover) are started to execute. Based on the fitness evaluation, the healthiest individuals (i.e. the best
solutions) are selected as the parents to generate new individuals (children). The mutation operator is also
applied to introduce children with new features in the next generation. The particular characteristic of the new
individuals is that they are more healthy and smart which results in increasing the average fitness (fitness level)

7
in each generation. After some generations, based on the stopping criteria, the algorithm is ended and probably
represents the best solution of the problem. Additional description of genetic algorithms is available in many
references [28, 29].

Set the GA parameters

(Define the objective function, number Input variables & fixed
and Bounds limits of decision variables, parameters: (Mpf , Ti, To, Mfg,
see section 4.1) cc/Do, Qloss/Qf, and etc)

Generate the initial population

(Based on decision variables: Npass, Flux,
Rdf, ,… see Table 1)

Fired heater design

(Based on the Equations of
Appendices A-C)

Update the Evaluate the objective function

generation count (Fitness evaluation of each individuals in the
population, based on the Equations 1 to 6)

Generation of new population based on

the GA operators (selection, crossover
and mutation) for searching the optimum
values of decision variables

No Satisfy Yes The optimal design of

the stopping criteria? fired heater is
obtained

Fig. 4. Flow diagram of the fired heater design optimization using GA

In this study, first, the mathematical model of the fired heater design is developed in MATLAB
environment. Next, the genetic toolbox of MATLAB is used to find the optimal economic design of the fired
heater. The following setting for GA is applied:

- population size: 50 individuals;

- selection function: stochastic uniform;
- cross over fraction: 0.9;
- mutation function: adaptive feasible;
- maximum generations: 150;
- stall generations: 100
The algorithm generates 50 coincidence individuals in each step and eventually this leads to expiration of
optimal solution after 100 subsequent iterations.

8
5. Results and Discussion

In this section, the verification of the proposed model is performed and the optimization results are
presented. The model validation is conducted by employing the proposed mathematical model to design a fired
heater with design specifications taken from Ref. [21] (see Table 2). Then, the obtained results are compared
with the values corresponding to the literature. The optimization results of a fired heater through two case
studies, considering two sets of decision variables as scenarios A & B in each case, are presented. As the first
case study, the design optimization of the fired heater for both scenarios A & B, based on the minimization of
total annual cost (TAC) is performed with GA. In the second case study, the optimal design of the fired heater
for the same scenarios A & B is upgraded by incorporating the pumping cost into the operating cost and
modifying the objective function (MTAC). The design specifications for the model validation and all cases are
assumed the same and presented in Table 2.

Table 2. Design Specifications for all cases [21]

Parameter (unit) Symb. Value

Flue gas rate (kg/MJ) Mfg 0.4342
Center to center spacing of tubes (m) cc 0.2032
Process fluid flow rate (kg /s) Mpf 44.9812
Process fluid inlet temperature (K) Ti 466.483
Process fluid outlet temperature (K) To 630.37
Ambient temperature (K) Ta 310.92
Operating time (h/yr) OT 8000

5.1. Model validation

In this section, the validity and accuracy of the proposed model are investigated. It has to be noted that
the method suggested by Mussati et al. [21] has been used more as a simulator rather than an optimizer. Then,
the values of their results are not optimal. Accordingly, to compare the results of the two methods, the values of
the decision variables listed in Table 1 should be fixed to those given by Ref. [21]. These fixed parameters are
presented in Table 3.

Table 3. Fixed parameters used for the model validation

Parameter value
Npass 6
(%) 74.4
Rdf 0.68
Flux (kW/m2) 36913.583
W (m) 6.096
La (m) 12.192
Gs (kg/m2 s) 1
ex (%) 25
Do (m) 0.1143

In Table 4, the solution obtained for the proposed model, which is named Original Design, is compared
with the Mussati Design. As can be seen, there is a high consistency between the geometrical and operational

9
results. In Fig. 5, the costs’ detailed results of Original Design are compared with the values corresponding to the
literature [21]. The percentage deviation of the total annual cost as compared to the Mussati design [21] is about
0.08. The results show that the developed model can be successfully implemented to design refinery fired heaters
with acceptable accuracy.

Table 4. Comparison of Original Design with Mussati design [21] (model validation)

Mussati Design
Variable Original Design
Ref. [21]
Ntrad 96 96
Ntcon 108 108
Rowcon 18 18
Ntshield 6 6
Qf (MW) 28.3745 28.343
QA(MW) 21.9712 21.937
Qrad (MW) 14.9322 14.966
H (m) 7.62 7.615
L (m) 11.7348 11.735
2
Arad (m ) 404.5227 404.523
Tg (K) 1151.42 1155.03
Qcon (MW) 7.039 6.971
2
Acon (m ) 455.0881 455.088
Ts (K) 699.662 705.406
Ds (m) 1.8288 1.791
Hs (m) 19.92 20.776
TAC ($/yr) 1862800 1861360.7

2.0e+6
1862800 1861360.7
1.8e+6

1.6e+6 1549100 1547672.2

1.4e+6
Cost ($/yr)

1.2e+6

1.0e+6
Capital cost (Ccost)
8.0e+5 Operating cost (Ocost)
Total annual cost (TAC)
6.0e+5

4.0e+5 313661.5 313688.6

2.0e+5

0.0
Mussati Design Original Design

Fig. 5. The capital, operating and total annual costs of Original Design as compared to Mussati design [21]

It is worth mentioning that the dataset reported by Ref. [21] has been obtained based on Mixed Integer
Nonlinear Programming (MINLP) model implemented in General Algebraic Modeling System (GAMS). While
in this study the mathematical model has been solved in the MATLAB platform. The basic equations are based
on Lobo-Evans method and are similar in both models. Furthermore, for validating the proposed model, the input
data for both models are assumed the same. Consequently, similar results for Original Design and Mussati
Design should be expected. The next sections present the optimal design of a fired heater which is obtained by
applying Genetic Algorithms on the proposed mathematical model.

10
5.2. Case study I: Optimal design

Several optimization problems can be considered to further explore the relative advantages of the
proposed approach compared to the optimization design approach of Ref. [21]. Mussati et al. [21] used the
proposed MINLP model in design optimization of a fired heater and demonstrated a total annual cost reduction
to Original Design. However, they assumed the outer diameter and the excess air as fixed parameters in their
optimization model, which can considerably affect the total annual cost. In addition, they excluded the cost of
the pumping power from the objective function. Furthermore, it seems that some of the reported optimal values
(for convection section) do not exactly satisfy the thermal constraints. Hence, the modification of the
optimization approach may lead to an increase in the thermal surface area and consequently the total annual cost.

In this case, to demonstrate the ability of the proposed approach, two optimization scenarios with a
different number of decision variables are solved. For each scenario, the purpose of optimization is to minimize
the total annual cost (TAC) using a genetic algorithm. The required design data for these problems are the same
data as presented in Table 2.

5.2.1. Scenario A (Optimal Design A)

In this optimization scenario, the parameters listed in Table 1 except the outer diameter of tubes and the
excess air are considered as decision variables (7 decision variables). The values of the outer diameter of tubes
and the excess air are assumed to be fixed and equal to 0.1143 (m) and 25%, respectively. In fact, in this
approach the optimization design of Original Design is performed using the proposed developed model with the
genetic algorithm by seven decision variables. The comparison between Optimal Design A and Original Design
results is presented in Table 5.

Since the total flow rate of crude oil into the furnace as a design specification of a fired heater is
constant, according to equation (A1), the fluid velocity is increased with decreasing the number of passes.
Therefore, the total heat transfer coefficient of convection zone in Optimal Design A is increased by 9.7% as
compared with Original Design. In this approach, the heat transfer area in the convection section is slightly
higher than Original Design. In addition, the radiant duty fraction (Rdf) is reduced from 0.68 to 0.66 as
compared to Original Design. Therefore, Optimal Design A provides a convection section with a higher
contribution in heat transfer to crude oil (about 2%). In fact, by developing convection section which results in
higher heat transfer coefficient and heat transfer area for Optimal Design A, the furnace heat recovery is
increased. Consequently the furnace efficiency is increased and the fuel consumption is decreased. Besides,
Optimal Design A represents a radiant section with lower number of tubes, heat transfer area and firebox size
than Original Design. As a result, the capital cost of this section is significantly decreased for Optimal Design A
(see Fig. 6). Finally, this optimization approach provides a lower total annual cost, about 2%, and a higher
efficiency, about 1.6%, as compared with the traditional design method.

5.2.2. Scenario B (Optimal Design B)

The second optimization scenario, which is referred to as Optimal Design B, is solved based on all
design variables given in Table 1 (9 decision variables). In order to compare the capability of the optimization
design approaches, the solution of this case is given in the last column of Table 5.

In this case the lower bound on the pipe diameter and the excess air are reached and the overall heat
transfer coefficient in the convection zone again is increased compared to that of Original Design, while the
radiant duty fraction for both Optimal Design B and Original Design is the same. Then, the heat transfer area in
the convection zone for the Optimal Design B is reduced significantly. Due to the reduction in pipe diameter, the
number of tubes (Ntrad , Ntcon) in both sections is increased as well as HTAMrad. HTAMrad is defined as heat
transfer area of the radiant tubes per unit mass of process fluid, which is a key parameter of heat transfer in the
radiant zone. Moreover, the surface area of the tubes and the volume of the firebox for Optimal Design B are
lower.

11
 The costs detailed comparisons of Original Design with the optimal solutions obtained by applying
different optimization scenarios are presented in Fig. 6. As shown in Fig. 6 the operating costs (fuel
consumption) of the optimal solutions are the same and lower than that of the Original Design. On the other
hand, the capital cost for Optimal Design B is considerably lower than that of Optimal Design A and both of
them are lower than Original Design. From the above comparison, it can be seen that the total annual cost for
Optimal Design B is reduced by 3.4% and 1.4% compared with the Original Design and Optimal Design A,
respectively

Table 5. Comparison of optimization scenarios of case study I with Original Design

Original Case study I (TAC minimization)

Variable
Design Optimal Design A Optimal Design B
Ntrad 96 80 114
Ntcon 108 96 102
Rowcon 18 24 17
Npass 6 4 6
Ntshield 6 4 6
Do (m) 0.1143 0.1143 0.073025
ex (%) 25 25 20.19
77.4 78.96 78.58
Qf (MW) 28.343 27.769 27.769
QA(MW) 21.937 21.926 21.821
Qrad (MW) 14.966 14.429 14.787
Rdf 0.68 0.66 0.68
W (m) 6.096 5.872 5.412
H (m) 7.615 5.989 5.493
L (m) 11.734 13.391 15.209
Arad (m2) 404.523 384.681 397.787
Flux (kW/ m2) 36.997 37.51 37.174
Tg (K) 1155.03 1167.57 1168.82
Qcon (MW) 6.971 7.497 7.034
Acon (m2) 455.088 461.617 355.915
Uc (W/m2 K) 38.698 42.466 49.006
LMTD (C) 404.024 382.445 403.292
Ts (K) 705.406 671.02 698.14
Ds (m) 1.791 1.86 1.96
Hs (m) 20.776 25.824 21.68
D Pi (kPa) 173.65 549.72 2303.001
Atubing (m2) 893.102 875.192 776.358
Vfirebox (m3) 544.756 470.972 452.171
HTAMrad (m2/kg) 0.0571 0.0571 0.097
TAC ($/yr) 1861360.7 1823455.8 1798328.2
.

2000000 1861360.7 1823455.8 1798328.2

1547672.2 1516326.9 1516326.9

1500000
Cost ($/yr)

Series1
Capital cost
1000000
Series2
Operating cost
Series3
Total annual cost
500000
313688.6 307128.9 282001.3

0
Original
1.00Design Optimal
2.00Design A Optimal3.00 Design B
Case study I

Fig. 6. Comparisons of the optimal solutions costs of case study I with Original Design

12
 Also, the tube side pressure drop for both scenarios of case study I is investigated. In scenario B, by
reducing the pipe diameter the pressure drop is significantly increased (see Table 5). So that the pressure drop for
Optimal Design B is higher than the permissible limit (1000 kPa). Therefore, the optimal design of scenario B,
despite its lower TAC compared to scenario A, is not technically acceptable.

5.3. Case study II: Modified optimal design

To investigate the pumping power effect on the optimal design, the objective function is modified by
including the pumping cost in the operating cost (according to equations (4)-(6)). Next, based on the modified
objective function (MTAC), two optimization scenarios of case study I are solved. In scenario A, the
optimization problem based on MTAC as an objective function is performed using GA and seven decision
variables. In scenario B, by importing pipe diameter and excess air into the decision variables, the modified
optimization problem is solved. The optimal solutions of this case, referred to as Modified Optimal Design, are
presented and compared to Original Design in Table 6.

Table 6. Comparison of optimization scenarios of case study II with Original Design

Case study II (MTAC minimization)

Original
Variable Modified Optimal Modified Optimal
Design
Design A Design B
Ntrad 96 80 108
Ntcon 108 96 108
Rowcon 18 24 18
Npass 6 4 6
Ntshield 6 4 6
Do (m) 0.1143 0.1143 0.0889
ex (%) 25 25 20.19
77.4 78.69 78.58
Qf (MW) 28.343 27.769 27.769
Qrad (MW) 14.966 14.342 14.668
Rdf 0.68 0.66 0.67
W (m) 6.096 5.665 5.437
H (m) 7.615 5.989 6.464
L (m) 11.734 13.391 12.951
Arad (m2) 404.523 384.686 390.63
Flux (kW/ m2) 36.997 37.284 37.551
Tg (K) 1155.03 1172.21 1175.66
Qcon (MW) 7.115 7.511 7.153
Acon (m2) 455.088 461.623 390.63
Uc (W/m2 K) 38.698 42.52 45.26
LMTD (C) 404.024 382.66 404.59
Ts (K) 705.406 670.05 697.38
Ds (m) 1.791 1.79 1.94
Hs (m) 20.776 27.673 20.26
D Pi (kPa) 173.65 549.73 716.05
Atubing (m2) 893.102 875.203 808.84
Vfirebox (m3) 544.756 454.386 455.097
HTAMrad (m2/kg) 0.0571 0.0571 0.0765
MTAC ($/yr) 1864455.5 1833257.1 1818126.1

Table 6 shows the optimization process design considering the constraints on the decision variables in Table 1.
The furnace efficiency is increased to 78.7% and the heat released by the fuel (Qf) is reduced from 28.343 to
27.769 (MW). Consequently, the thermal performance for both scenarios A & B as compared with the Original
Design shows improvement. Moreover the tube side pressure drop for both Modified Optimal Design A & B is
in the permissible range. However, the total heat transfer area for Modified Optimal design B is lower than the
Modified Optimal design A. Therefore the modified total cost for scenario B is lower than for scenario A. In Fig.
7 the detailed comparison of the optimal solutions costs of this case (MTAC minimization) with Original Design

13
is presented. The Modified operating costs including the costs of fuel consumption and pumping for both optimal
solutions A & B are the same and lower than that of the Original design. It is indicated that the modified total
annual cost for Modified Optimal Design B is reduced by 2.5% and 0.8% compared to the Original design and
Modified Optimal Design A, respectively.

Capital cost Modified operating cost Modified Total Annual Cost

2100000
1864455.5 1833257.1 1818126.1
1800000
1550766.9 1526124.2 1529088.3
1500000
Cost ($/yr)

1200000
900000
600000
313688.58 307132.99 289037.84
300000
0
Original Design Modified Optimal Modified Optimal
Design A Design B

Fig. 7. Comparison of the costs of case study II with Original Design

To compare two optimization approaches (case studies I & II), the optimal values obtained for the main
variables are presented in Table 7. As can be seen, all operational and geometrical characteristics (such as the
number of tubes, passes, heating surfaces, furnace dimension, efficiency and so on) of Modified Optimal Design
A are similar to Optimal Design A. Also in scenario A for case studies I & II, by upgrading the convection
section as compared to the Original Design, the heat recovery and efficiency of furnace are increased.
Consequently, when the pipe diameter is assumed a fixed parameter (i.e. scenario A), both the design
optimization approaches (TAC and MTAC minimization) result in the same solutions.

Table 7. Comparison between the optimal solutions of the optimization approaches

Case study I (minimizing TAC) Case study II ( minimizing MTAC)

Variables Optimal Design A Optimal Design B Modified optimal solution A Modified optimal solution B

(7 decision variables) (9 decision variables) (7 decision variables) (9 decision variables)

Ntrad 80 114 80 108
Ntcon 96 102 96 108
Rowcon 24 17 24 18
Npass 4 6 4 6
Ntshield 4 6 4 6
Do (m) 0.1143 0.073025 0.1143 0.0889
ex (%) 25 20.19 25 20.14
Qf(MW) 27.769 27.769 27.769 27.769
QA(MW) 21.926 21.821 21.853 21.821
78.96 78.58 78.69 78.5
W (m) 5.872 5.412 5.665 5.437
H (m) 5.989 5.493 5.989 6.464
L (m) 13.391 15.209 13.391 12.951
D Pi (kPa) 549.723 2303.001 549.728 716.04678
Atubing (m2) 875.192 776.358 875.203 808.84
Vfirebox (m3) 470.972 452.171 454.356 455.097
HTAMrad 0.0571 0.097 0.0571 0.0765
TAC ($/yr) 1823455.8 1798328.2 1823459.9 1805364.8
MTAC ($/yr) 1833252.9 1839372.2 1833257.1 1818126.1

14
On the other hand, in scenario-B, the values of total annual cost (TAC) for both optimization approaches are
close together. However, the pressure drop on the tube-side for Optimal Design B (case study I) is higher than
the allowable value. While, the optimization approach based on MTAC minimization (case study II) modifies the
dimension and arrangement of tube coil such that the tube side pressure drop is reduced to below the permissible
value (1000 kPa). From the analysis of the above optimization approaches, it is reasonably concluded that when
the pipe diameter is a decision variable in the optimization procedure (i.e. scenario B), the pumping cost can be
an effective parameter to determine the economic optimal design of fired heaters. Fig. 8 summarizes the results
obtained in the optimization approaches. As can be seen, the Modified Optimal Design B obtained by
minimizing the modified total annual cost and nine decision variables is the best solution. The fuel consumption
of this optimal solution similar to Optimal Design A is decreased by 2% compared to Original Design. However,
the total annual cost (TAC) and the modified total annual cost (MTAC) for this case are lower than for Optimal
Design A. Moreover, the value of the heat transfer area to mass ratio in the radiant section for this case is higher
than for Optimal Design A (see Table 7).
1864455.5
1833252.9 1818126.1 MTAC
1861360.7
1.8e+6 1823455.8
1805364.8
TAC
1.7e+6

1.6e+6
1547672.2
Costs ($/yr)

1.5e+6
1516326.9 1516326.9 Fuel cost

1.4e+6

1.3e+6

1.2e+6

1.1e+6

1.0e+6
Original Design Optimal Design A Modified Optimal Design B
Case study I Case study II
Increasing the number of effective decision variables

Fig. 8. Comparison of the costs of optimal solutions and Original Design

6. Conclusions

A detailed mathematical model associated with Genetic Algorithms (GAs) was developed to determine
the economic optimal design of process fired heaters. The developed model considers the discrete decision
variables including the number of passes and tube diameter based on API standards as well as continues decision
variables including furnace efficiency, width of firebox, tube length, mass velocity in stack, radiant duty fraction
and heat flux. Furthermore, the model includes constraints on the operational parameters (pressure drop, excess
air and heat flux) and geometric design (furnaces dimensions) in order to achieve an optimal and reliable design.
The comparison of results with literature shows that the model is completely robust and realistic. Two case
studies, considering various sets of decision variables and objective functions, are successfully solved. The main
findings are as follows:
 The proposed model, through examining different case studies considering various objective functions
(TAC, MTAC) and several sets of design variables, demonstrates higher robustness and flexibility than
other models reported in the literature.
 Different optimization approaches, considering the tube diameter as a fixed parameter, show that the
pumping cost effect is very low on the optimal design and the optimization approaches (TAC & MTAC
minimization) result in the same solution.
 When the tube diameter is considered as a decision variable, the optimization approach based on
MTAC minimization must be used in order to control the pressure drop of crude oil and obtain a
realistic optimal design.
 Finally, considering nine decision variables and MTAC as an objective function, the economic/realistic
optimal design obtained is 2.48% cheaper than Original Design.

Appendix A: Radiant zone sizing

The heat balance for a fired heater is expressed by Eq. (A1).

15
 (A1)
where Qloss is the heat loss rate through the furnace walls which is about 2% to 3% of the heat generated by the
fuel (Qf) and Qs is the heat content of the gas leaving the convection section [22]. Here, the percent of heat loss
to the surrounding is assumed 2.5%. Qrad and Qcon are the heat transfer rates to the process fluid in radiant and
convection zones, respectively.

The heat duty of a fired heater (QA) can be computed as:

(A2)
where hi and ho are the enthalpies of inlet and outlet process fluid, respectively.

The radiant duty fraction (Rdf) is defined as the ratio of radiant heating (Qrad) to the heat absorbed (QA). The
radiant duty fraction can be initially assumed as 75% of the total heat duty [22].

The heat generated by the combustion process (Q f) is calculated as:

(A3)

where Mf , LHV and hF are the rate of fuel consumption, the lower calorific value and the furnace efficiency,
respectively.

For the given tube diameter (Di) and flow rate of the process fluid (Mpf), the number of required passes is
represented by Eq. (A4).

(A4)
where Ac is the cross section area of the tube, and Vpf are the density and velocity of the process fluid,
respectively.

Heat transfer in the radiant section of a fired heater is given by Eq. (A5) [22],

(A5)

where Tg and Tt are the effective gas temperature and the average tube skin temperature in the radiant zone,
respectively. In addition, , Acp, and F are defined as the tube surface absorptivity, the cold plan area and the
exchange factor, respectively.

The required number of tubes in the radiant zone (Ntrad) is computed as follows:

(A6)

where Do is the outer diameter of the tube and L is the exposed length of the tubes in the radiant zone which is
calculated as:

(A7)
where Ka refers to the unexposed length of the tube. In this model, La is the tube length which is considered as
input data between 12 and 16 m. Usually, the exposed length of the tube (L) is 0.45 m shorter than the tube
length [22, 23].

The total number of tubes in the radiant section is obtained as:

(A8)

16
The following constraints are presented to estimate the dimensions of the radiant zone [21]:

(A9)
(A10)
(A11)

(A12)

(A13)

(A14)
where WC is the width required to accommodate the number of the shield or convection tubes in each row (see
Fig. 2). kk1, kk2, and kk3 are constants which are related with tube diameter, distance to refractory, tube
arrangement, among others.

The heat content of the gas leaving the radiant zone (Qgc) can be correlated as a function of flue gas temperature
(Tg-oF) and excess air (ex) as:

(A15)

The physical properties of process fluid, based on the temperature of fluid (T pf) can be estimated as follows [7,
31]:

Specific Gravity: SG = SG15 - (5.93*10-4)*(Tpf-15)

(A16)
-3 -3
Heat capacity: Cp = (2*10 *Tpf-1.429)*SG+2.67*10 *Tpf+3.049
(A17)
Thermal conductivity: k = 0.49744 – 29.4604*10-5*Tpf
(A18)
Viscosity: ln(μ) = - 0.2207*ln2(Tpf) + 0.5052*ln(Tpf)-11.8201
(A19)
where SG15 is the specific gravity of crude oil at 15 ( C) .

Appendix B. Convection Section

Heat transfer in the convection section is given by,

(B1)
The logarithmic mean temperature difference is represented as:

(B2)

The cross-over temperature of process fluid from convection to radiant section (T c) can be determined from the
energy balance for the process fluid in the radiant zone. The stack inlet temperature (Ts) can also be computed
considering equations (A1) and (A15).

The overall heat transfer coefficient in the convection section depends on the total apparent gas film coefficient
outside the tubes (h co) and the convection heat transfer coefficient inside the tubes (h ci).

(B3)
The total apparent gas film coefficient is calculated as:

17
 (B4)
The following equations compute the individual heat transfer coefficients (h cc , hcr) that correspond to the
convective gas film and the flue gas radiation coefficients, respectively [7, 21].

(B5)

(B6)
(B7)
(B8)
where is the average specific heat capacity of flue gas, Tgf and Tgc are the average gas film
temperature and the average gas temperature in the convection zone and Gmax is the maximum flue gas rate at the
minimum cross section in the convection zone.

In Eq. (B4) f is a coefficient factor based on re-radiation from the refractory walls of the convection section. This
factor usually ranges from 6 to 15% of the sum convective gas film and flue gas radiation coefficients [7]. A
value of 10% is considered as a typical average.

For turbulent flow, 10000<Re<120000 and L/Do>60, the tube convection heat transfer coefficient, is represented
by [7]:

(B9)

where K, Pr and μ refer to the thermal conductivity, Prantdtl number and viscosity of the process fluid,
respectively. They are computed at the average temperature of the process fluid and μw is computed at the tube-
wall temperature. The required number of tubes (Ntcon) and rows (Rowcon) in the convection zone are presented
as follows:

(B10)

(B11)

The tube side pressure drop for single-phase process fluid can be estimated using the equation:

(B12)

where v, G, Di and f are respectively, mean specific volume, mass velocity of process fluid, internal diameter of
tube and fanning friction factor. Le is the equivalent length of tube which allows to describe all sources of
pressure losses inside the tube as a length of pipe.

Appendix C. Stack Calculation

The draft effect of the stack is represented by,

(C1)
where Dstack and Dfire box are the draft effects of the stack (above shield tubes) and the fire box, respectively.

The draft in inch of water can be estimated by [17, 30]:

(C2)

18
 where Hs is the height of stack or firebox, P0 is the atmospheric pressure, Ta is ambient temperature and Tga is
average flue-gas temperature in the stack or firebox.

The pressure drop across the convection section (ΔPc) can be estimated by:

(C3)
which Ph is the velocity-head in convection section and can be computed in inch of water using the equation:

(C4)

where G is the mass velocity of the combustion products in the convection section.

The pressure losses in the stack (ΔPs) corresponding to the stack entrance and exit, damper and stack friction can
be estimated in terms of velocity-head in the stack (Phs), height of stack (Hs) and inner stack diameter (Ds) by the
following equation:

(C5)

Nomenclature
Ac Cross section area of tube (m) Parch Vacuum at the arch of firebox (Pa)
Acon Convection heat transfer area (m2) Ph Velocity-head in the convection section (Pa)
Acp Cold plan area of tubes in firebox (m2) Phs Velocity-head in the stack (Pa)
Arad Radiant heat transfer area (m2) DPi Pressure drop inside of tubes (Pa)
AMrad Area to unit mass ratio in radiant zone (m2/kg) DPb Pressure drop across the burners (Pa)
cp Average specific heat capacity (kJ/kg K) DPc Pressure drop across the convection section (Pa)
Ccost Capital cost ($/yr) DPs Pressure loss in the stack (Pa)
cc Center to center spacing of tubes (m) QA Heat absorbed by the process fluid (MW)
cccost Unitary cost of convection section ($/m2) Qcon Heat transfer rate in the convection zone (MW)
crcost Unitary cost of radiant section ($/m2) Qeqp Equivalent fuel consumption of pumping power (MW)
CRF Capital Recovery Factor (yr-1) Qf Heat released by fuel combustion (MW)
Dfirebox Draft effect of the firebox (Pa) Qgc Heat content of gas leaving the radiant zone (MW)
Di Internal tube diameter (m) Qloss Heat loss rate through the furnace walls (MW)
Do Outer tube diameter (m) Qrad Heat transfer rate in the radiant zone (MW)
Ds Stack Diameter (m) Qs Heat content of gas leaving the convection zone (MW)
Dstack Draft effect of the stack (Pa) Re Reynolds number
ex Excess air (%) Rdf Radiant duty fraction
f Fanning friction factor Rowcon Number of tube-rows in convection section
F Exchange factor SG Specific gravity of crude oil
Flux Radiant heat flux (kW/m2) Ta Ambient temperature (K)
G Mass velocity (kg/m2 s) Tc cross-over temperature of process fluid (K)
Maximum of flue gas mass velocity in convection section
Gmax Tg Effective gas temperature (K)
(kg/hr.m2)
Gs Mass velocity of flue gas in stack (kg/m2 s) Tga Average temperature of flue gas in stack section (K)
h Enthalpy of process fluid (kJ/kg K) Tgc Average temperature of flue gas in convection section (K)
2
hcc Convective gas film coefficient in convection section (W/m K) Tgf Average gas film temperature in convection section (K)
hci Tube side film coefficient (W/m2 K) Ti Inlet temperature of process fluid (K)
Total apparent gas film coefficient in convection section
hco To Outlet temperature of process fluid (K)
(W/m2 K)
hcr Gas-radiation coefficient in convection section (W/m2 K) Tpf Process fluid temperature (K)
H Height of firebox (m) Ts Inlet stack temperature (K)
Hs Height of stack from above of the firebox (m) Tt Mean temperature of tube skin (K)
HTAMrad Heat transfer area per unit mass of oil (m2/kg) TAC Total annual cost ($/yr)
K Thermal conductivity (W/m K) Ntceil Number of ceiling tubes
Ka Unexposed length of tube (m) Ntconv Number of tubes in convection section
K1, K2 Numerical constants Ntrad Number of tubes in radiant section
kk1, kk2, kk3 Numerical constants Ntshield Number of tubes on shield section
L Length of furnace/exposed tube (m) Ntwalls Number of wall side tubes
La tube length (m) Uc Total heat transfer coefficient (W/m2 K)
Le Equivalent length of tube (m) Vpf Process fluid velocity (m/s)
LMTD Logarithm mean temperature difference ( C ) W Width of furnace (m)
LHV Low heating value (kJ/kg) X1, X2 Functions of excess air
Mf Fuel mass flow rate (kg/s) Greek symbols
Mfg Flue gas rate (kg/MJ) P Total efficiency of pump and motor
Mpf Process fluid mass flow rate (kg/s) g Power generation efficiency

19
MOcost Modified operating cost ($/yr) F Fired heater efficiency
MTAC Modified total annual cost ($/yr) ρ Process fluid density (kg/m3)
Npass Number of passes ρg Flue gas density (kg/m3)
ocost Unitary cost of fuel ($/J) a Absorptivity of a tube surface
Ocost Operating cost ($/yr) μ Viscosity of process fluid (N.s/m2)
OT Plant Operating Time (hr/yr) μw Viscosity of process fluid at the tube-wall temperature (N.s/m2)
Pa Ambient air pressure (Pa)
References
[1] R.K. Sinnott, and G.Towler, Chemical engineering design: SI Edition. Elsevier, 2009.
[2] Z. Jegla, J. Kohoutek, P. Stehlik, Design and operating aspects influencing fouling inside radiant coils of fired
heaters operated in crude oil distillation plants, proceeding of international conference on heat exchanger fouling
and cleaning, 2011.
[3] M. Shekarchian, F. Zarifi, M. Moghavvemi, F. Motasemi, T.M.I Mahlia, Energy, exergy, environmental and
economic analysis of industrial fired heaters based on heat recovery and preheating techniques, Energy
Conservation and Management, 71 (2013) 51-61.
[4] M.E. Masoumi, Z. Izakmehri, Improving of refinery furnaces efficiency using mathematical modeling,
International Journal of modeling and optimization, 1 (1) (2011) 74-79.
[5] H. Al-Haj Ibrahim, M. Al-Qasseimi, Calculation of radiant section temperatures in fired process heaters, Chemical
Engineering and Science, 1 (4) (2013) 55-61.
[6] H. Al-Haj Ibrahim, M. Al-Qasseimi, Matlab program computes thermal efficiency of fired heater, Chemical
Engineering, 52 (2) (2008) 61-69.
[7] H. Al-Haj Ibrahim, M. Al-Qasseimi, Simulation of heat transfer in the convection section of fired process heaters,
Chemical Engineering, 54 (1) (2010) 33-40.
[8] S. Gunasegran, A. Azarpour, Numerical analysis of influential parameters on the performance of vertical-
cylindrical refinery furnaces, 4th International Conference on Process Engineering and Advanced Materials,
Procedia Engineering, 148 (2016) 963-969.
[9] A. Chaibakhsh, N. Ensansefat, A. Jamali, R. Kouhikamali, Crude oil direct fired furnace model, Applied Thermal
Engineering, 83 (2015) 57-70.
[10] A. Morales-Fuentes, G. T. Polley, M. Picon-Nuñez, S. Martínez-Martínez, Modeling the thermo-hydraulic
performance of direct fired heaters for crude processing, Applied Thermal Engineering, 39 (2012) 157-162.
[11] A. Morales-Fuentes, M. Picon-Nuñez, G. T. Polley, S. Méndez-Días, Analysis of the influence of operating
conditions on fouling rates in fired heaters, Applied Thermal Engineering, 62 (2014) 777-784.
[12] Z. Jegla, J. Hajek, J. Vondal, Numerical analysis of heat transfer in radiant section of fired heater with realistic
imperfect geometry of tube coil, Chemical Engineering Transactions 39 (2014) 889-894
[13] Z. Jegla, J. Vondal, J. Hajek, Standards for fired heater design: An assessment based on computational modeling,
Applied Thermal Engineering 89 (2015) 1068-1078.
[14] E. Khodabandeh, M. Pourramezan., M.H. Pakravan, Effects of excess air and preheating on the flow pattern and
efficiency of radiative section of fired heater in a petroleum refinery, Applied Thermal Engineering 105 (2016)
537-548.
[15] M. Sarlej, P. Petr, J. Hajek, P. Stehlik, Computational support in experimental burner design optimization,
Applied Thermal Engineering 27 (16) (2007) 2727-2731.
[16] T. Weydahl, J. jamaluddin, M. Seljeskog, R. Anantharaman, Pursuing the pre-combustion CCS route in oil
refineries – The impact on fired heaters, Applied energy 102 (2013) 833-839.
[17] G. Towler, R.K. Sinnott, Chemical Engineering Design, Principles, Practice and Economics of Plant and Process
Design, Elsevier, 2012, pp. 793-795, 932-938.
[18] Z. Jegla, The conceptual design of a radiant chamber and preliminary optimization of a process tabular furnace,
Heat transfer engineering 27 (6) (2006) 50-57.
[19] Z. Jegla, Optimum arrangement of tube coil in radiation type of tabular furnace, Heat Transfer Engineering 29 (6)
(2008) 546-555.
[20] A. Bahadori, H.B. Vuthaluru, Novel predictive tools for design of radiant and convection sections of direct fired
heaters, Applied Energy 87 (7) (2010) 2194-2202.
[21] S. Mussati, J.I. Manassaldi, S.J. Benz, N.J. Scenna, Mixed integer nonlinear programming model for the optimal
design of fired heaters, Applied Thermal Engineering 29 (11) (2009) 2194-2204.
[22] J.R. Couper, W.R. Penney, J.R. Fair, S.M. Walas, Chemical process equipment, selection and design, 3 rd ed.,
Elsevier, 2012, pp. 202-210.
[23] Standard A. P. I. 560: Fired Heaters for General Refinery Services." American Petroleum Institute, Washington,
D.C. (2001).
[24] M. King, Process control: a practical approach, John Wily & Sons, 2016, pp. 300-305.

20
 [25] N. Wimpress, Method for predicting fired heater performance, Mchanical Engineering, 100 (1) (1978) 100-
100.345 E 47TH ST, NEW YORK, NY 10017: ASME-AMER SOC MECHANICAL ENG.
[26] F. Pettersson, H. Saxén, K. Deb, Genetic algorithm-based multicriteria optimization of ironmaking in the blast
furnace, Materials and Manufacturing Processes 24 (3) (2009) 343-349.
[27] C-Y. Lin, P. Hajela, Genetic algorithms in optimization problems with discrete and integer design variables,
Engineering optimization 19 (4) (1992) 309-327.
[28] R.L. Haupt, S.E. Haupt, Practical genetic algorithms, 2nd ed., John Wiley & Sons, Hoboken, 2004.
[29] D.E. Goldberg, Genetic algorithms in search, optimization and machine learning, Reading: Addison-Wesley
Longman Publishing, Co., Boston, USA, 1989.
[30] W.E. Lobo, J.E. Evans. Heat transfer in the radiant section of petroleum heaters, Trans. Am. Inst.
Chem. Engrs 35 (1939) 748-778.
[31] H. L Berman. FIRED HEATERS, Part I: Finding the basic design for your application; Part II: Construction
materials, mechanical features performance monitoring; Part III: How combustion conditions influence design
and operation; Part IV: How to reduce your fuel bill, Chemical Engineering 85 (1978), a four-part series, Mc
Graw-Hill.
[32] API, Design and operation of oil water separator. Publication 421, American Petroleum Institute API,
1990.

Table captions

Table 1. Limitations of the decision variables for optimization

Table 2. Design Specifications for all cases [21]

Table 3. Fixed parameters used for the model validation

Table 4. Comparison of Original Design with Mussati Design (model validation)

Table 5. Comparison of optimization scenarios of case study I with Original Design

Table 6. Comparison of optimization scenarios of case study II with Original Design

Table 7. Comparison between the optimal solutions of the optimization approaches

Figure captions
Fig. 1. Flow sketch of a cabin type fired heater

Fig. 2. Configuration of tubes and heating zones of a fired heater

Fig. 3 A four-pass fired heat layout

Fig. 4. Flow diagram of the fired heater design optimization using GA

Fig. 5. The capital, operating and total annual costs of Original Design as compared to Mussati Design

Fig. 6. Comparisons of the optimal solutions costs of case study I with Original Design

Fig. 7. Comparisons of the costs of case study II with Original Design

Fig. 8. Comparison the costs of optimal solutions and Original Design

21
Research highlights

 A modified optimization approach based on genetic algorithms is proposed for fired heaters.
 The effect of pumping cost on the optimal design is analyzed.
 Two cases are studied to illustrate the advantage of the proposed optimization model.
 The modified optimal design obtained is 2.48% cheaper than Original Design.

22