2017 American Control Conference

Sheraton Seattle Hotel

May 24–26, 2017, Seattle, USA

Steam Methane Reforming Furnace Temperature Balancing via CFD

Model-Based Optimization
Anh Tran, Andres Aguirre, Helen Durand, Marquis Crose and Panagiotis D. Christofides

Abstract— In this work, we introduce a furnace balancing steam into hydrogen, carbon dioxide and carbon monoxide in
algorithm that generates an optimized furnace-side feed distri- the presence of a nickel-based catalyst network. The process
bution that has the potential to improve the thermal efficiency takes place inside a unit that consists of two closed domains.
of reformers. The furnace balancing algorithm is composed of
three major components: data generation, model identification Typically, in one domain the fuel reacts with excess oxygen
and a model-based balancing scheme. Initially, we adopt a in air to generate the necessary thermal energy to drive the
computational fluid dynamics (CFD) model of an industrial- endothermic SMR process taking place in the other domain.
scale reformer developed in our previous work for the data The core unit of the SMR process is the steam methane
generation, as this model has been confirmed to simulate the reformer. In the steam methane reformer, the two closed
typical transport and chemical reaction phenomena observed
during reformer operation, and the CFD simulation data is in domains are referred to as the tube side and the furnace
good agreement with various sources in literature. Then, we side. The tube side is defined as a collection of all tubular
propose a model identification scheme in which the algorithm reforming reactors (which are referred to as reforming tubes),
is formulated based on the least squares regression method and the furnace side is defined by the remaining components
and basic knowledge of radiative heat transfer. Subsequently, of the reformer including the combustion chamber walls and
we create a model-based balancing scheme that is formulated
as an optimization problem within which the furnace-side feed flue-gas tunnels. In the steam methane reforming application,
distribution is the decision variable, and minimizing the sum of the key reactions of the SMR process are endothermic and
the weighted squared deviations of outer reforming tube wall reversible, and therefore, a reformer configuration that allows
temperatures from a set-point value for all reforming tubes is significant heat transfer to the tube side near the reforming
used as the objective function. CFD simulation results provide tube inlets is the appropriate design. Motivated by this, the
evidence that the optimized furnace-side feed distribution
created by the furnace balancing algorithm can reduce the top-fired furnace configuration is the most frequently used
temperature nonuniformity inside the combustion chamber, and reformer furnace configuration in industry.
therefore, allow the reformer thermal efficiency to be increased At chemical manufacturing plants that are designed to
without shortening the unit’s service life. produce high purity hydrogen on a commercial scale, the
plants’ efficiency is directly linked to the thermal efficiency
I. I NTRODUCTION of the reformer because it is the most energy-intensive unit.
Hydrogen is one of the most important raw materials Additionally, the reformer thermal efficiency is controlled by
for petroleum and chemical manufacturing industries. In the the temperature distribution inside the combustion chamber.
petroleum refining industry, many refining catalytic processes Specifically, a higher reformer thermal efficiency is observed
(e.g., the hydrotreating process, hydrocracking process and with a more uniform temperature distribution in the com-
hydrodesulfurization process) consume hydrogen. Specifi- bustion chamber [3]. The degree of nonuniformity in the
cally, the hydrotreating process, which requires hydrogen temperature distribution refers to the maximum variation in
to convert olefins to paraffins, and hydrocracking process, the radially averaged outer reforming tube wall temperature
which uses hydrogen to cleave sigma carbon-carbon bonds, between reforming tubes at a specified axial location [3],
serve the same purpose of increasing the hydrogen content [10], [13]. The degree of nonuniformity in the temperature
of the downstream process reactants, which allows these distribution, which may be measured by a system of infrared
downstream processes to have higher efficiency. The hy- (IR) camera sensors situated around an online reformer, can
drodesulfurization process utilizes hydrogen to remove thiol be as high as 110 K. A reformer that possesses a highly
compounds in the raw natural gas feedstock to prevent the nonuniform temperature distribution is referred to as an
catalysts of downstream processes from being poisoned. imbalanced reformer.
In an industrial setting, hydrogen is commonly produced A nonuniform temperature distribution is a critical op-
by the steam methane reforming (SMR) process, which is an erational problem, which is frequently mentioned in pub-
endothermic process that converts methane and superheated lished literature under various terms (e.g., a nonuniform
temperature distribution, temperature maldistribution and
Anh Tran, Andres Aguirre, Helen Durand and Marquis Crose are with temperature nonuniformity) [3], [10], [13]. The high degree
the Department of Chemical and Biomolecular Engineering, University of
California, Los Angeles, CA 90095, USA. Panagiotis D. Christofides is of nonuniformity in the temperature distribution forces the
with the Department of Chemical and Biomolecular Engineering and the reformer to be operated at suboptimal conditions at which
Department of Electrical Engineering, University of California, Los Angeles, the hydrogen production rate of the reformer is much less
CA 90095, USA. Email: pdc@seas.ucla.edu. Corresponding author:
P. D. Christofides. Financial support from the National Science Foundation than its design capacity. To understand the impact of the
and the Department of Energy is gratefully acknowledged. high degree of nonuniformity in the temperature distribution

978-1-5090-5992-8/$31.00 ©2017 AACC 4165

on the reformer thermal efficiency, the essential operational the reformer from producing hydrogen at the theoretical
constraints of the reformer should be mentioned. Specifi- reformer capacity. Then, a model identification scheme based
cally, the endothermic characteristics of the SMR process on a least squares linear regression method is formulated to
suggest that the reformer interior should be kept at high derive a data-driven model relating the outer reforming tube
temperature because a higher conversion of hydrogen can wall temperature (OTWT) distribution to FSF distribution
be accomplished by a higher outer reforming tube wall from the CFD data generated by the reformer CFD model.
temperature [6]. However, an increase in the outer reforming Subsequently, a model-based FSF redistribution scheme is
tube wall temperature of 20 K above the design temper- proposed. Specifically, minimizing the sum of the weighted
ature can reduce the reforming tube expected service life squared deviations of the OTWTs from a set-point value
of 100,000 hours by half [5]. In addition, when the outer for all reforming tubes and the distribution of the FSF are
wall temperature of the reforming tubes exceeds the design chosen as the cost function and the decision variable of the
value by a significant margin, the reforming tube wall might optimization problem, respectively. Finally, the performance
rupture, resulting in production and capital losses. Therefore, of the furnace balancing algorithm will be evaluated based on
when the temperature nonuniformity inside the combustion the percent reduction in the standard deviation of the OTWT
chamber is severe, the reformer must be operated at sub- distribution generated by the reformer CFD model with the
optimal conditions in which the overall average operating optimized FSF distribution with respect to that generated by
temperature of the reforming tubes is sufficiently smaller the reformer CFD model with uniform FSF distribution.
than the design value to compensate for the wide range of
II. F URNACE BALANCING ALGORITHM
the outer reforming tube wall temperatures. Additionally, it
is reported that an imbalanced reformer has noticeably lower A. Data generation
radiative heat transfer efficiency compared to a balanced The first step of the furnace balancing algorithm is data
reformer [3]. generation which employs a reformer CFD model to generate
The major cause of the temperature nonuniformity is an CFD data [11]. The reformer CFD model is created based
asymmetric distribution of the velocity of the combustion on the top-fired reformer designed by Selas Fluid Processing
products in the furnace side (commonly referred to as the Corporation [5]. The reformer has a width of ∼16 m,
furnace-side flow pattern) on which the temperature dis- length of ∼16 m and height of ∼13 m. The unit produces
tribution inside the combustion chamber depends [13]. On 2,800,000 Nm3 of high-purity hydrogen and 1,700,000 kg
the other hand, the asymmetric furnace-side flow pattern of superheated steam per day with an annual operating cost
is the result of the asymmetric geometry of the reformer of 62,000,000 dollars. Inside the reformer, there are seven
as the furnace-side outlet of the combustion chamber is parallel rows of 48 reforming tubes, inside which commercial
situated on the front side of the refractory wall as shown nickel-based catalyst pellets are used as packing material. At
in Fig. 1. Hence, the asymmetric furnace-side flow pattern the ceiling of the reformer, the rows of the reforming tubes
is an intrinsic characteristic of the reformer and cannot be are separated by eight rows of 12 burners. It is important
resolved without having to modify the reformer geometry. to note that 24 outer-lane burners which are adjacent to one
However, Oprins et al. [10] demonstrates that the severity of row of 48 reforming tubes are designed to be smaller and
the nonuniform temperature distribution under the influence to have lower capacity compared to the remaining 72 inner-
of the asymmetric furnace-side flow pattern can be reduced lane burners which are adjacent to two rows of 48 reforming
by adjusting the furnace-side feed (FSF) distribution. tubes. At the floor of the reformer, the rows of the reforming
Motivated by the above considerations, the present work tubes are separated by eight flue-gas tunnels. On each side
focuses on developing an optimization-based furnace bal- of the flue-gas tunnels, there is a row of 35 evenly spaced
ancing algorithm to reduce the degree of nonuniformity in rectangular extraction ports.
the temperature distribution inside the combustion chamber In CFD study, mesh generation is the most critical and
such that the overall efficiency of the hydrogen plant can time-consuming process because a CFD model created from
be increased without shortening the reformer service life. a poor-quality grid requires more computational resources
The furnace balancing algorithm consists of data generation, and a longer computing time to calculate a converged
model identification and optimization of the furnace-side simulation result, and the result is likely to carry a large
distribution. The data generation step utilizes a computational numerical error [2]. It is worth noting that the reformer
fluid dynamics (CFD) model of an industrial-scale reformer geometry is considered to be simple for CFD modeling as it
(e.g., that developed in our previous work [11], which has is made mostly of straight edges as shown in Fig. 1, and for
been confirmed to successfully simulate the typical transport systems with simple geometries, an unstructured meshing
and chemical reaction phenomena observed in reformers, approach offers no advantages compared to a structured
and the simulation results generated by the CFD model meshing approach [2]. Additionally, we have found that
have been verified to be in good agreement with typical for wall-bounded systems such as the reforming tubes, a
industrial data in published literature). Initially, the reformer CFD model created from a multiblock structured hexahedral
CFD model with a uniformly distributed FSF is employed to grid typically generates a simulation result closer to the
investigate the existing level of nonuniformity in the temper- corresponding experimental data compared to that generated
ature distribution in the combustion chamber that prevents by a CFD model built from an unstructured tetrahedral grid

4166
 TABLE I
with the same number of subdomains [2], [4]. Therefore,
R EFORMER MESH QUALITY
the reformer mesh created for this study is a multiblock
structured hexahedral grid with ∼41,000,000 subdomains.
Based on the three mesh quality recommendations from the The reformer mesh Recommended range
creator ANSYS of ICEM CFD software, which are used to Min orthogonal factor 0.459 0.167 − 1.000
evaluate the quality of the reformer mesh. Table I shows that Max ortho skew 0.541 0.000 − 0.850
the reformer mesh quality is considered to be acceptable. Max aspect ratio 28.5 1.000 − 100.0
In our previous work, the transport and chemical reaction
phenomena observed inside the reformer during hydrogen
production are discussed at great length, and the modeling
describing the relationship between the OTWT distribution
strategies to create the reformer CFD model that require a
at a specified location away from the reforming tube inlets
moderate computational time are described in Lao et al. [4]
and the FSF distribution. The model identification is for-
and Tran et al. [11]. In particular, the reformer CFD model is
mulated based on a linear least squares regression method.
able to simulate the non-premixed combustion characteristics
Specifically, we assume that the average OTWT of the ith
of the furnace-side fuel stream by using the standard k − ϵ
reforming tube at a specified axial location, Ti (K), can be
turbulence model, finite rate/eddy dissipation (FR/ED) model
written as a linear combination of the FSF flow rates of all
and global kinetic models of methane/hydrogen combustion
burners in the reformer as follows,
in air [1], [8]. Additionally, the reformer CFD model is
95
!
capable of simulating radiative heat transfer between the
furnace-side mixture, combustion chamber refractory walls Ti = αij Fj (1)
j=0
and outer reforming tube walls by using an empirical cor-
relation between the furnace-side radiative properties and where Fj (kg s−1 ) is the FSF flow rate of the j th burner
temperature [7], Lambert Beer’s law, Kirchoff’s law and the and αij (K kg−1 s) is the empirical coefficient of the data-
discrete ordinate method [2]. Furthermore, the reformer CFD driven correlation representing the impact of Fj on Ti . In
model can simulate the SMR process in the tube side by this study, the outer-lane/ inner-lane burners and reforming
means of the standard k − ϵ turbulence model, FR/ED model tubes are indexed from 0th − 95th and 0th − 335th in the
and global kinetic model of the SMR process [12]. Moreover, specified patterns as shown in Fig. 1.
the reformer CFD model can also simulate the effects of the In high-temperature applications, thermal radiation is the
catalyst network on the tube side flow by employing a built- dominant mode of heat transfer. Olivieri et al. [9] show
in ANSYS porous zone function of which the parameters that radiative heat transfer accounts for 95% of the total
are estimated from the semi-empirical Ergun equation and heat transfer in the top-fired reformer, which suggests that
relevant typical industrial data [4]. Lastly, we note that the the OTWT distribution is controlled primarily by thermal
simulation results generated by the reformer CFD model radiation. The rate of energy transferred by thermal radi-
have been validated and confirmed to be in good agreement ation between two blackbodies at different temperatures is
with typical industrial SMR data in published literature, and commonly modeled as being proportional to the difference
therefore, can be considered as a good representation of in temperatures raised to the fourth power. Additionally, the
experimental data. radiative heat transfer rate is proportional to the radiation
intensity, which is inversely proportional to the distance
between the two blackbodies raised to the second power.
This indicates that the rate of thermal radiation quickly
decreases with an increasing distance between two black
bodies. Therefore, the influence of the FSF flow rate on the
OTWT is localized, which suggests that when the distance
between the j th burner and ith reforming tube is sufficiently
large, it can be assumed that the FSF flow rate of the j th
the burner has negligible impact on the OTWT of the ith
reforming tube. In this study, the distance between a specified
burner and reforming tube is defined as the distance between
the projection of the burner centroid and the projection of
the reforming tube centroid onto any 2-D horizontal cross-
Fig. 1. The isometric view of an industrial-scale, top-fired, co-current sectional plane.
reformer.
To quantitatively determine the local effect of the FSF flow
rate of each burner on the OTWT distribution, a cylindrical
B. Model Identification volume (as shown in Fig. 2) is used to represent the heating
The second step of the furnace balancing algorithm is zone of a burner in which the outer wall temperatures of
model identification which utilizes the CFD data collected the reforming tubes are dependent on the FSF flow rate
from the data generation step to create a data-driven model of the burner. Additionally, the concept of heating zone

4167
allows the data-driven model to account for the reformer
geometry characteristics (i.e., the burner and reforming tube
arrangements). It is important to note that a larger cylindrical
volume (i.e., that allows the FSF flow rate of a burner to
affect more surrounding reforming tubes) allows the model
identification to have more optimization variables, which
may make the data-driven model more accurate with respect
to the reformer CFD data at the cost of increased compu-
tational requirements. We have found that at rHZ ∼3.43
m (as shown in Fig. 2) the least-squares algorithm of
the model identification produces an acceptable data-driven
model within a reasonable computing time interval. The data-
Fig. 2. A representation of a burner heating zone which is created by the
driven model is designed to predict the OTWT distribution highlighted burner. The burner heating zone is represented by a shaded-blue
corresponding to a FSF distribution, and the data created cylindrical volume.
from the data-driven model is expected to be close to that
taken from a reformer CFD simulation result in the least
squares sense. The least-squares algorithm of the model C. Model-based balancing scheme
identification based on n sets of the reformer CFD simulation
data is formulated as follows, The third step of the furnace balancing algorithm is
⎛ ⎞2 the model-based balancing scheme which utilizes the data-
n−1
!! 335 !95
⎝Tk,m − driven correlation (Eq. 1) to determine an optimized FSF
min αkj Fj,m ⎠ (2)
αij ∈[0,∞) distribution which can reduce the degree of nonuniformity
m=0 k=0 j=0
in the OTWT distribution. Specifically, the furnace balancing
subject to
scheme is formulated as an optimization problem in which
αkj = 0 if dkj > rHZ (3a) minimizing the sum of the weighted squared deviations
αkj = αki if dkj = dki (3b) of all outer reforming tube wall temperatures from a set-
& 'β point value (TAV E ) is the objective function. The idea of
dkj
· αkj ≥ αki ≥ αkj if dkj > dki (3c) implementing the weighted squared deviations is critical
dki because the data-driven model is not designed to account
where Tk,m is the average outer wall temperature of the k th for the effect of the furnace-side flow pattern on the OTWT
reforming tube taken from the mth reformer CFD data set, distribution inside the reformer. Hence, in the effort of
Fj,m is the FSF flow rate of the j th burner derived from determining the appropriate weighting factors of the furnace
the mth reformer CFD data set, β = 6.0 is an empirical balancing scheme, the existing reformer CFD data is utilized
constant of the data-driven model estimated from the study to investigate the impact of the furnace-side flow pattern on
of the burner heating zone, dkj is the distance between the OTWT distribution. Fig. 3 shows that the combustion
the k th reforming tube and the j th burner and dki is the products flow down the reforming tube length, enter the flue-
distance between the k th reforming tube and the ith burner. gas tunnels through the extraction ports and move toward
In Eqs 3a−3c, the ranges of k, i and j are 0−335, 0−95 and the reformer outlets. The furnace-side flow pattern causes
0 − 95, respectively. The cost function (Eq. 2) of the model the wall temperature of the flue-gas tunnels to increase with
identification penalizes the deviation of the average outer decreasing distance from the reformer outlets as shown in
wall temperature of each reforming tube generated by the Fig. 4. We note that the minimum temperature of the flue-gas
data-driven model from that derived from the corresponding tunnels (1240 K) is higher than the maximum temperature of
reformer CFD data set. Eq. 3a suggests that if the k th the outer reforming tube wall (1183 K) [11], which suggests
reforming tube is not situated within the heating zone of the that the neighboring reforming tubes also receive additional
j th burner (Eq. 3a), the data-driven model will assume that radiative heating from the flue-gas tunnels. However, the
the FSF flow rate of the j th burner does not affect the k th additional radiative heating transferred to the neighboring
outer reforming tube wall temperature. Additionally, Eq. 3b reforming tubes along each row is nonuniform, and specifi-
indicates that if the distance between the k th reforming tube cally, the reforming tubes near the reformer outlets to receive
and the j th burner is equal to that between the k th reforming a higher rate of radiative heating from the neighboring flue-
tube and the ith burner, the data-driven model then presumes gas tunnels compared to the other reforming tubes that are
that the effects of the burners on the k th outer reforming tube situated further away. As a result, to account for the impact
wall temperature are the same. Similarly, Eq. 3c shows that of the nonuniform heating from the neighboring flue-gas
if the distance between the k th reforming tube and the j th tunnels on the OTWT distribution, the weighting factors in
burner is greater than that between the k th reforming tube the objective function are chosen as shown in Eqs. 5a, 5b
and the ith burner, the data-driven model then infers that the and 5c, so that the deviations of the estimated OTWT from
effects of the j th burner on the k th outer reforming tube wall TAV E are penalized more severely for reforming tubes that
temperature are weaker than those of the ith burner. are positioned further away from the reformer outlets so that

4168
the FSF can be redirected to the burners situated near the
refractory back wall.
In addition, the model-based balancing scheme is formu-
lated to account for typical industrial operational constraints.
In particular, it is unconventional for a FSF flow rate of a
burner to be regulated individually. In the reformer, every
set of two consecutive burners in a row of twelve burners
is assumed to be controlled by a flow regulator and to have
the same FSF flow rate (as shown in Eq. 5d). A common
flow regulator installed to adjust the FSF distribution is a
flow control valve. In this study, we assume that the FSF of
the burners is regulated by two distinct flow control valve Fig. 3. The velocity vector field of the furnace-side flow pattern in the
models, one of which is used for the inner-lane burners, vicinity of the 4th burner row in the reformer is constructed from the
reformer CFD data [11].
and the other valve model is implemented in the outer-lane
burners such that when all valves are fully open, the FSF flow
rate of the outer-lane burners is 60% of that of the inner-lane
burners. Even though in industrial practice a FSF distribution
is generated from a certain valve position distribution, the
decision variable of the model-based balancing scheme is
chosen to be the FSF distribution. Nevertheless, the furnace
balancing algorithm remains useful for industrial applications
that target to reduce the temperature nonuniformity in the
combustion chamber and to increase the reformer thermal
efficiency because for any FSF distribution, a valve position
distribution can be derived based on the characteristics
of flow control valves and the proportionality relationship
between the valve models. In this work, we assume that the
Fig. 4. The temperature contour map of the 4th flue-gas tunnel, which is
total FSF derived based on typical industrial data is kept situated directly under the 4th burner row in the reformer, is created from
constant at Ftot when an optimized distribution is evaluated the reformer CFD data in [11], and the reforming outlets are situated at the
(as shown in Eq. 5e). The model-based balancing scheme is bottom right corner.
formulated as follows,
⎛ ⎞2
!335 95
! tube wall temperature distribution and the limited detailed
min wk ⎝TAV E − αkj Fj ⎠ (4)
Fj ∈(0,∞),j=0,··· ,95 industrial data of reformer operational settings (e.g., although
k=0 j=0
the total FSF flow rate is available for public access, to the
subject to extent of our knowledge the data of the FSF distribution has
wk = 0.1 if (k mod 48) < 16 (5a) never been reported), we assume that the FSF is uniformly
wk = 0.3 if 16 ≤ (k mod 48) < 32 (5b) distributed among the inner-lane burners and among the
outer-lane burners. Hence, the reformer CFD simulation
wk = 0.6 if 32 ≤ (k mod 48) (5c)
result of the uniformly distributed FSF serves as a basis for
Fj = Fj+1 if j ∈ {0, 2, · · · , 94} (5d) this investigation. It is critical to recall that a given FSF
95
! distribution is the result of a combination of valve settings,
Fj = Ftot (5e) which regulates the amount of fuel and air delivered to each
j=0 group of two burners. Specifically, the uniform distribution
where wk is the weighting factor of the k th reforming tube of the FSF is accomplished by allowing all valves to be fully
and Fj is the optimized FSF flow rate of the j th burner. It is opened in which the feed flow rate of the outer-lane burners
important to note that wk is determined based on a trial- is 60% of that of the inner-lane burners because of the pro-
and-error approach until the largest reduction in standard portionality relationship of the valves discussed in Sec. II-C.
deviation of the estimated OTWT distribution is observed. The OTWT distributions along the reforming tubes that are
generated by the reformer CFD model implemented with the
III. S IMULATION RESULTS uniformly distributed FSF are shown in Fig. 5. It is reported
In this section, we illustrate the deployment of the furnace that the temperature nonuniformity pattern is independent
balancing algorithm to generate the optimized FSF distri- of the total FSF flow rate and of the axial location along
bution that leads to a reduction of nonuniformity in the the reforming tubes [3]. This information suggests that if
temperature distribution inside the combustion chamber. Due the OTWT distribution becomes more uniform at any given
to the lack of prior knowledge of the quantitative effect of the plane at a constant height in the reformer then the overall
asymmetric furnace-side flow pattern on the outer reforming temperature nonuniformity inside the combustion chamber is

4169
 1200
also reduced and the overall reformer thermal efficiency can
be improved. In the remainder of this section, we focus on

Outer wall temperature (K)

Standard deviation, σ (K)

15
the OTWT distribution that is 6.5 m away from the reforming 1150
max
tube inlets because the distribution has reasonably large Twall
values of the maximum temperature (1187.0 K), temperature ave
Twall
1100
range (40.0 K) and standard deviation (7.1 K). Then, the min
Twall
10

model identification developed based on the least squares σ

regression method is employed to create a data-driven model 1050
that relates the selected OTWT distribution to the FSF 5
distribution based on the available CFD data. Next, the data-
1000
driven model is utilized to predict the reformer behavior as 2 4 6 8 10 12
Distance down the reforming tube (m)
the model-based balancing scheme of Eqs. 4−5 searches for Fig. 5. OTWT distributions along the reforming tubes by the reformer
CFD model implemented with uniform FSF distribution (represented by the
the optimized FSF distribution to decrease the temperature black-filled symbols) and with the optimized FSF distribution (denoted by
nonuniformity inside the reformer. Subsequently, the opti- the red-hollow symbols).
mized FSF distribution is implemented within the reformer
CFD model, and the resulting CFD simulation data is used to
evaluate the performance of the furnace balancing algorithm and the FSF flow rates from the CFD data. Then, we
prior to applying the predicted control action to the online introduced the model-based balancing scheme that accounts
industrial-scale reformer. Specifically, the performance of the for typical industrial operational constraints while searching
furnace balancing algorithm is assessed based on the percent for an optimal FSF flow rate distribution based on the data-
reduction in the standard deviation of the OTWT distribution driven model. The CFD simulation results showed that the
generated by the reformer CFD model implemented with the optimized FSF distribution created by the proposed furnace
optimized FSF distribution with respect to that of the basis. balancing algorithm reduced the nonuniformity of the chosen
The proposed furnace balancing algorithm solved by Ipopt OTWT distributions along the reforming tubes, which al-
can generate an optimized FSF distribution and quantitatively lowed the reformer thermal efficiency to be improved without
estimate the reduction of the temperature nonuniformity shortening the reformer life service.
of the OTWT distribution within 24 hours using 80 cores
R EFERENCES
of UCLA Hoffman2 Cluster. Approximately 90% of the
[1] S. P. M. Bane, J. L. Ziegler, and J. E. Shepherd. Development of one-
required computing time is designated to the data generation step chemistry models for flame and ignition simulations. Technical
procedure of the furnace balancing algorithm. We would like report, GALCIT Report GALCITFM:2010.002, 2010.
to note that it is necessary to execute the furnace balancing [2] ANSYS, Inc. ANSYS Fluent Theory Guide 15.0. 2013.
[3] A. Kumar, M. Baldea, T. F. Edgar, and O. A. Ezekoye. Smart
algorithm for more than one iteration initially to allow the manufacturing approach for efficient operation of industrial steam-
data-driven model to gradually become a sufficiently accurate methane reformers. Industrial & Engineering Chemistry Research,
representation of the reformer CFD model, as is required to 54:4360 – 4370, 2015.
[4] L. Lao, A. Aguirre, A. Tran, Z. Wu, H. Durand, and P. D. Christofides.
derive the optimized furnace-side distribution. CFD modeling and control of a steam methane reforming reactor.
The simulation results shown in Fig. 5, which are gen- Chemical Engineering Science, 148:78 – 92, 2016.
erated after five iterations of the furnace balancing algo- [5] D. Latham. Masters Thesis: Mathematical Modeling of An Industrial
Steam Methane Reformer. Queen’s University, 2008.
rithm, suggest that the optimized FSF distribution reduces [6] J. S. Lee, J. Seo, H. Y. Kim, J. T. Chung, and S. S. Yoon. Effects of
the overall nonuniformity of the OTWT distributions and combustion parameters on reforming performance of a steam-methane
enhances the radiative heat transfer efficiency of the reformer. reformer. Fuel, 111:461 – 471, 2013.
[7] A. Maximov. Thesis for the Degree of Doctor of Science: Theoretical
Specifically, the maximum outer wall temperature and the Analysis and Numerical Simulation of Spectral Radiative Properties of
corresponding temperature range are decreased to 1183.4 K Combustion Gases in Oxy/air-fired Combustion Systems. Lappeenranta
and 31.4 K, respectively. The selected OTWT distribution University of Technology, 2012.
[8] D. G. Nicol. Ph. D. Thesis: A Chemical Kinetic and Numerical
becomes more narrow as the standard deviation is lessened Study of NOx and Pollutant Formation in Low-emission Combustion.
from 7.1 K to 5.2 K. The reduction in standard deviation of University of Washington, 1995.
the outer reforming tube wall temperature distribution due [9] A. Olivieri and F. Vegliò. Process simulation of natural gas steam re-
forming: Fuel distribution optimisation in the furnace. Fuel Processing
to the optimally distributed FSF with respect to the basis is Technology, pages 622 – 632, 2008.
27.8%. [10] A. J. M. Oprins and G. J. Heynderickx. Calculation of three-
dimensional flow and pressure fields in cracking furnaces. Chemical
IV. C ONCLUSION Engineering Science, pages 4883 – 4893, 2003.
[11] A. Tran, A. Aguirre, H. Durand, M. Crose, and P. D. Christofides.
In this work, we developed a furnace balancing algorithm CFD modeling of an industrial-scale steam methane reforming furnace.
that can reduce the temperature nonuniformity inside the Chemical Engineering Science, submitted.
[12] J. Xu and G. F. Froment. Methane steam reforming, methanation and
combustion chamber and thus improve the reformer thermal water-gas shift: I. intrinsic kinetics. AIChE Journal, 35:88 – 96, 1989.
efficiency. To this end, we adopted a comprehensive reformer [13] D. Zheng, B. Wu, J. Fleitz, R. Trajkovski, and C. Q. Zhou. CFD
CFD model created in our previous work and a model simulation of a hydrogen reformer furnace. In 2010 14th International
Heat Transfer Conference, pages 233 – 244. American Society of
identification scheme based on the least squares regression Mechanical Engineers, 2010.
method to approximate the relationship between the OTWT

4170