Processes: Optimization of Reaction Selectivity Using CFD-Based Compartmental Modeling and Surrogate-Based Optimization
Processes: Optimization of Reaction Selectivity Using CFD-Based Compartmental Modeling and Surrogate-Based Optimization
Processes: Optimization of Reaction Selectivity Using CFD-Based Compartmental Modeling and Surrogate-Based Optimization
Article
Optimization of Reaction Selectivity Using
CFD-Based Compartmental Modeling and
Surrogate-Based Optimization
Shu Yang, San Kiang, Parham Farzan and Marianthi Ierapetritou *
Department of Chemical and Biochemical Engineering, Rutgers, The State University of New Jersey,
98 Brett Road, Piscataway, NJ 08854, USA; yang.shu.public@gmail.com (S.Y.); san.kiang@gmail.com (S.K.);
parhamfarzan@gmail.com (P.F.)
* Correspondence: marianth@soe.rutgers.edu; Tel.: +1-848-445-2971
Received: 19 November 2018; Accepted: 24 December 2018; Published: 29 December 2018
Abstract: Mixing is considered as a critical process parameter (CPP) during process development due
to its significant influence on reaction selectivity and process safety. Nevertheless, mixing issues are
difficult to identify and solve owing to their complexity and dependence on knowledge of kinetics
and hydrodynamics. In this paper, we proposed an optimization methodology using Computational
Fluid Dynamics (CFD) based compartmental modelling to improve mixing and reaction selectivity.
More importantly, we have demonstrated that through the implementation of surrogate-based
optimization, the proposed methodology can be used as a computationally non-intensive way for
rapid process development of reaction unit operations. For illustration purpose, reaction selectivity
of a process with Bourne competitive reaction network is discussed. Results demonstrate that we can
improve reaction selectivity by dynamically controlling rates and locations of feeding in the reactor.
The proposed methodology incorporates mechanistic understanding of the reaction kinetics together
with an efficient optimization algorithm to determine the optimal process operation and thus can
serve as a tool for quality-by-design (QbD) during product development stage.
1. Introduction
In chemical synthesis, many important reactions can be accompanied by undesired side-reactions.
This leads to wastes and affects product quality. Therefore, incorporating knowledge of mixing
can substantially improve reaction selectivity and yield, in addition to enhancing process safety.
Furthermore, due to a growing variety of reactors, characterization of mixing has become vital in
process development [1,2]. To achieve optimal selectivity and yield, appropriate modeling of the mass
transport process in reactors is critical [3]. Nevertheless, owing to the complexity of mass transport in
turbulent flow, analyzing mixing remains a difficult problem.
Frequently, mixing in reactor is approximated by residence time distribution (RTD) analysis, where
residence time is experimentally measured through a tracer test. This approximation has been proved
to work relatively well, however “RTD is not a complete description of structure for a particular reactor or
system of reactors” [3]. Therefore, when reaction with high conversion rates are considered, analysis
solely based on RTD can lead to significant error [4]. Based on local sensors, RTD characterization
of tracer test can be improved by mixing time measurement [5,6]. However, considering potential
bias and the requirement for specific equipment for tracer tests, RTD and mixing time measurement
have become less preferable comparing to the more resource-effective benchmark reaction method [7].
Therefore, competitive reaction systems have become the standard for mixing analysis [7–9]. Among
the competitive reaction systems, the “well-documented and highly reliable” [7] Bourne reaction and
Villermaux reaction is most commonly adopted [10]. Nevertheless, benchmark reaction tests provide
only the input-output relationship, without detailed description of the process dynamics. Therefore,
developing optimal process operation experimentally remains challenging.
With rapid advances in computer technology, computational fluid dynamics (CFD) has become
a powerful tool to study mixing. Comparing with the experimental methods, it provides detailed
understanding of mixing phenomenon in a timely-efficient manner without requiring meticulous
choice of equipment and sensors. In reaction engineering, CFD has been employed to study mixing
in chemical reactors and bio-reactors. Mixing of liquid-liquid system [11], solid-liquid system [12] or
non-Newtonian fluid [13] are studied with CFD, which agrees well with experimental data. In the
presence of complex chemical reactions [14] or biological metabolism [15] CFD is implemented to
provide detailed understanding of the mass transfer process. Complex hydrodynamics, mass transfer,
heat transfer and reaction kinetic can be satisfactorily captured by CFD, making it a powerful tool in
process design.
As a result, CFD has been implemented to optimize mixing by improve reactor design. Studies
reported in the literature mainly focus on improving geometrical attributes of reactors. Researches
that directly integrate detailed CFD simulation with optimization algorithms has successfully
improved reactor design [16–20]. Due to the complexity of CFD simulations, the chosen optimization
algorithms are often meta-heuristic, such as genetic algorithm (GA) [16–18] and particle swarm
methodology [20]. They can address complex black-box problems, such as reactor geometry [16,17]
or impeller configuration [18], but require a large number of function evaluations, which leads to a
large number of computationally expensive CFD runs. To improve the computational efficiency of
direct CFD simulations, hybrid methods have been proposed that replace direct CFD evaluation with
simpler data-driven models [21–25], which are then integrated with GA. Successful implementation
have been reported in the literature that use neural network [22,23], Gaussian process [24] and radial
basis function (RBF) [25]. However, building confidence in those data-driven models requires large
number of CFD runs, and balancing computational efficiency with accuracy is non-trivial [26].
Apart from improving mixing through optimized design, to the best of the authors’ knowledge,
there is no work that improves mixing by optimizing dynamic process operation based on CFD
simulations. The main reasons for the limited implementation is the complexity of CFD simulations.
To optimize dynamic process operation, more decision variables should be considered. Since GA
would suffer from “curse of dimensionality” [27], significantly more CFD runs would be required,
leading to increasing computational expense and degrading performance.
In this work, a framework is developed trying to leverage CFD simulations to optimize process
operation. Firstly, CFD-based compartmental model is built to replace direct CFD simulations. Comparing
to the data-driven models implemented for reactor geometry optimization, compartmental models are
finite-volume physical models, which provide satisfactory accuracy while requiring significantly less CFD
runs. Secondly, a surrogate-based optimization algorithm using radial-basis function is implemented,
which has been proven to be more efficient than GA [28]. This work offers a compact and systematic
framework for improving reaction selectivity with a numerically efficient Quality-by-Design (QbD)
tool. In a case study, Bourne reaction is employed as a benchmark, which serve as a more explicit
quantification for the efficiency of mixing. The proposed framework is compared with process
operations optimized based on ideal mixing model, suggesting great improvement by leveraging CFD
simulations. By integrating CFD-based compartmental model with surrogate-based optimization, the
proposed framework has shown a great potential for fast process development.
that in stirred tank reactors the flow field are considered independent from chemical reactions. As a
result, the fluid dynamics data can be used for different reactions, which could contribute to rapid
process design and cost reduction. The result from CFD simulation is used to develop a compartmental
model, which would be discussed in Section 2.1. Comparing to direct CFD simulation, computational
complexity of compartmental model is significantly reduced. Therefore, the optimal process design
can be determined numerically without prohibitive computational expense. The process selectivity is
then optimized by integrating the compartmental model with surrogate-based optimization, as will be
discussed in Section 2.2.
Adopting the divergence theorem and replacing Ci with the volume-averaged concentration Ci,Kj ,
Equation (2) is modified to Equation (3), where Vj is the volume of control volume Kj , and Sj is the
surface of control volume Kj . The mass flux N consist of convection and diffusion, where the diffusion
Processes 2019, 7, 9 4 of 20
is modelled by Fick’s law with diffusion coefficient D. The source term Ri models the homogeneous
consumption and generation of species i, which include chemical reactions and micro mixing. Due to
the assumption of homogeneous compartments, the volume integral of source term in Equation (5) is
modified as follows:
dCi,K j I Z
Vj + Ni,Kj ·ndS = Ri dV, (3)
dx Sj Kj
dCi,K j Z Z Z
Vj = −− v·Ci,Kj ·n dS + D ·∇C ·n dS + Ri dV, (4)
dx Sj Sj Kj
Z
Ri,Kj = Vj · Ri,Kj . (5)
Kj
It should be recognized that the homogeneous assumption depends upon the Damköhler number
(Da) in each compartment, which is a strong function of grid size, as will be discussed in sub
Section 2.1.3. Moreover, in this work diffusion mass transfer between different compartments are
neglected, which is also justified in sub Section 2.1.3 based on an analysis of the Péclet number (Pe).
The dominance of convective mass transfer would simplify 4 into Equation (6), where Qjk denotes the
flow rate from control volume j to control volume k.
dCi,Kj
Vj = −Ci,Kj ∑ Q j,k + ∑ Ql,j ·Ci,Kj + Vj · Ri,Kj . (6)
dt k l
Equation (6) is a finite-volume mixing model parameterized by mass flow rate Q and compartment
volumes V. Since CFD is also based on fine-volume models, parameters of this mixing model can be
effectively determined from CFD simulations, based on the topology mapping and merging strategy
proposed by Bezzo et al. [33].
Firstly, a steady-state CFD simulation should be developed and calibrated to yield a good
prediction of flow field inside the reactor. Then the computational cells of CFD mapped to the
same compartments are group into ensembles. Cells in the same ensemble are aggregated together,
where, the volumes and mass flow rate and summed together to determine the parameters of each
compartment. To achieve a good balance between computational complexity and model accuracy, the
resolution of the compartmental model is determined based on a grid independence test as will be
discussed in sub Section 2.1.3. For illustration purposes, a well-known pair of parallel competitive
Bourne reactions is studied. As will be discussed later in the case study, this reaction system is
composed of a first-order decay and a parallel second order coupling reaction [38].
In this work the grid independence test is performed in two steps. First the Damköhler number
(Da) and Péclet number (Pe) are analyzed, as suggested in Equations (7) and (8), which are critical to
justify the compartmentalization of model discussed in sub Section 2.1.2.
k1 C A + k2 Cb C A
Da = < 1, (7)
uC A /L
Lu
Pe = > 1. (8)
D
This analysis determines the lower and upper bounds of length scale for the compartments, which
could serve as a starting point for the grid independence test. Then the initial choice of length scale is
improved in an iterative manner. Simulations based on compartmental models with decreasing length
scale are tested. When the simulation results are no longer changing with the increasing grid density,
the model resolution can be considered as sufficient.
It should be mentioned that the optimal grid density depends on reaction kinetics. If the
time constant of chemical reactions is significantly larger than that of mixing, this process could
be considered mixing-insensitive and perfect mixing assumption could be adopted without harming
model precision. By contrast, for fast reaction, the deviation caused by perfect-mixing assumption
could be significant, which require higher resolution. If the characteristic time scale of reaction is
in orders of magnitude less than micro-mixing, which is in the order of 10−3 s [11], micro-mixing
would dominate chemical reaction. As a result, the rate law of chemical reaction should be replaced
with micro-mixing models. Although for different reaction kinetics we can use the same steady state
solution from CFD, if new reaction kinetics are used, grid independence test should be conducted with
the updated model.
Subject to:
[y R , y A ] = ϕ n, t1 , f 1 , t2 , f 2 , . . . , t Np , f Np , (10)
∑ tm = T, ∀m = 1, . . . , Ns , (11)
tm , f m ≥ 0, ∀m = 1, . . . , Ns , (12)
n ∈ [0, Nc ], (13)
where PR denotes the price of desired product R while yR denotes its yield. PA represents the unit cost
of raw material A and its consumption is denoted as yA . Both yR and yA are calculated through the
simulation ϕ based on the compartmental model. Addition point n and addition rate profile which is
defined by t1 , f 1 , t2 , f 2 . . . tNs , fNs are parameters of this simulation. The first set of constraints describe
the simulation based on the compartmental model. The second constraint represents that the total time
span of all stages is pre-defined as T.
Notice that the number of stages is introduced as a parameter instead of decision variable. This is
based on the difficulty of penalizing the monotonic increase of the number of stages. By allowing extra
degrees of freedom, an increasing number of stages is always preferred. Reaction selectivity would
always benefit from higher degree of freedom provided by the increasing Np , unless computational
expense of solving this optimization problem is taken into consideration. However, this is beyond the
scope of this paper. The duration of process T is defined as a parameter, which is usually determined
in the production scheduling stage. It is recommended to define T similar to the timescale of mixing in
mixing controlled processes to maximize time efficiency of reactors.
3. Case Study
In this section, a case study of a semi-batch process inside a dual-impeller stirred tank reactor
is studied. The duration of the whole process is 150 s, in which a fed-batch process is analyzed and
optimized. A well-known pair of competitive reactions [38] is introduced to study the influence of
mixing on reaction selectivity. The overall objective for the optimization problem is to maximize
Processes 2019, 7, 9 7 of 20
process productivity, which is defined as the revenue from selling the products minus the total cost of
raw materials injected. In this case study, process designed according to CFD-based compartmental
model and perfect-mixing model are compared to illustrate the effectiveness of this methodology.
Furthermore, constant
Processes 2019, 7, feeding
x FOR PEER REVIEW rate design is compared with time-varying feeding rate to demonstrate 7 of 20
that this framework can further improve reaction selectivity by enabling dynamic design.
3.1. Reactor Setup
3.1. Reactor Setup
This study was carried out in a 74 L baffled stirred vessel agitated with a Rushton impeller and
This study
a pitched bladewas carried
turbine, asout in a 74 Linbaffled
illustrated Figurestirred
1. Thevessel agitated
diameter of thewith a Rushton
vessel is 0.5 m,impeller
and theand a
liquid
pitched
level isblade
0.4 m turbine, as illustrated
from the bottom ofinthe Figure 1. The
vessel. Thediameter
Rushtonof the vessel
impeller is 0.5 m, and0.14
is assembled the m
liquid level
below the
ispitched
0.4 m from the bottom of the vessel. The Rushton impeller is assembled 0.14
blade turbine, whose blade angle is 45°. The agitation system is operated at 12 rpm m below the pitched
blade turbine, whose
anticlockwise, whichblade
drivesangle is 45◦ . The agitation
fluid downwards from the system
pitched is operated
blade turbine at to
12the
rpm anticlockwise,
Rushton impeller.
which drives fluid downwards from the pitched blade turbine to the Rushton impeller.
Figure 1. Geometrical dimension of the two-impeller stirred tank. Where (A) stands for the baffles, (B)
Figure 1. the
represents Geometrical dimension
pitched blade ofand
impeller the (C)
two-impeller stirred
denotes the tank.
Rushton Where (A) stands for the baffles,
impeller.
(B) represents the pitched blade impeller and (C) denotes the Rushton impeller.
3.2. Chemical Kinetics
3.2.To
Chemical
study Kinetics
the influence of mixing on reaction selectivity, a well-documented pair of parallel
competitive
To study the reactions
Bourne influenceisof
integrated.
mixing onThis reaction
reaction system is composed
selectivity, of a first-order
a well-documented pairdecay and
of parallel
a competitive
parallel second orderreactions
Bourne coupling,iswhere A is a This
integrated. diazonium salt
reaction (diazotized
system 2-chloro-4-nitroaniline)
is composed and
of a first-order decay
Band
is pyrazolone (4-sulphophenyl-3-carboxypyrazol-5-one).
a parallel second order coupling, where A is a diazonium R denotes the desirable
salt (diazotized product which is a
2-chloro-4-nitroaniline)
dyestuff, and S is the unwanted product of decomposition. The rate constants at 40 ◦ C are k = 10−3 s−1
and B is pyrazolone (4-sulphophenyl-3-carboxypyrazol-5-one). R denotes the desirable
1 product
and k2 =is7000 m3 kmol −1 s−1 at a PH = 6.6 [38]. Both reactants are dissolved in aqueous solution.
which a dyestuff, and S is the unwanted product of decomposition. The rate constants at 40 °C are
k1 = 10−3 s−1 and k2 = 7000 m3 kmol−1 s−1 at a PH = 6.6 [38]. Both reactants are dissolved in aqueous
k1
solution. A→ S, (14)
AA+⎯B →
kS,
2 (14)
R (15)
3.3.InFlow
this Field
reaction system, the advantage of defining objective function in the form of productivity is
Simulation
pronounced. Considering that the desired reaction happens faster than side reactions, infinitely slow
feedingCFD would simulation
always beispreferred
adopted iftowe
solve
want for
tovelocity
maximize field
the inside this reactor
ratio between based
desired on the
product andphysical
side
property of the solvent. The influence of the feeding pipe over
product. Based the time scale of mixing, the duration of process is set as 150 s.the flow pattern is ignored. Since the
flow rate of injection pipe is 10 order smaller than that of the bulk flow inside the reactor, influence
2
3.3.
ofFlow
reagentFieldinjection
Simulation
over the flow pattern is ignored.
CFD A steady state CFD
simulation simulation
is adopted is conducted
to solve with
for velocity theinside
field commercial code of
this reactor Ansys
based onFluent 16.0 [58].
the physical
property of the solvent. The influence of the feeding pipe over the flow pattern is ignored. Sincemulti-
The Reynolds-Averaged-Navier-Stocks (RANS) equation was numerically solved with the
reference frame (MRF) method.
2 To close the equations, k-epsilon turbulence model with
flow rate of injection pipe is 10 order smaller than that of the bulk flow inside the reactor, influence of standard
wall functions was adopted. The velocity field is shown in Figure 2.
reagent injection over the flow pattern is ignored.
It should be noted that it is necessary to calibrate CFD simulations so that can be confidently
A steady state CFD simulation is conducted with the commercial code of Ansys Fluent 16.0 [58].
utilized. Since this case study is used for demonstration purposes, due to the lack of experimental
The Reynolds-Averaged-Navier-Stocks (RANS) equation was numerically solved with multi-reference
data, validation of the numerical simulation is not conducted. This will be addressed in future work
frame (MRF) method. To close the equations, k-epsilon turbulence model with standard wall functions
where this method is applied.
was adopted. The velocity field is shown in Figure 2.
Figure 2. Simulated vector plot of velocity field inside the stirred tank (m/s).
Figure 2. Simulated vector plot of velocity field inside the stirred tank (m/s).
It should be noted that it is necessary to calibrate CFD simulations so that can be confidently
utilized. Since this case
3.4. Compartmental study isand
Modeling used forIndependence
Grid demonstration purposes, due to the lack of experimental data,
Test
validation of the numerical simulation is not conducted. This will be addressed in future work where
To develop compartmental models from steady state CFD simulation, computational cells
this method is applied.
extracted from CFD are aggregated based on a predefined grid. The grid is defined by evenly
dividing
3.4. the reactor
Compartmental in radial,
Modeling axial
and Grid and angular
Independence Testdirections. The grid density of each direction is
determined based on the grid sensitivity test proposed in Section 2.1.3.
ToTodevelop compartmental
justify the perfect-mixing models from steady
assumptions in eachstate CFD simulation,
compartment, computational
local Damköhler numbercells
(Da)
extracted from CFD are aggregated based on a predefined grid. The grid is defined by evenly dividing
is analyzed. As suggested Figure 2, the bulk velocity inside the stirred tank is in the order of 10−2 m/s.
theBased
reactoroninEquation
radial, axial
(7), and
the angular directions.
upper bound The grid
of length density
scale of each
in each direction isshould
compartment determined
be 1m.
based on the grid sensitivity test proposed in Section 2.1.3.
Furthermore, to justify neglecting diffusive mass transfer, Péclet number (Pe) is analyzed according
to Equation (8). Since diffusion coefficient in aqueous solutions are in the order of 10−9 m2/s, the lower
bound of compartment length scale is 10−7 m. It can be concluded that since in single phase turbulent
Processes 2019, 7, 9 9 of 20
To justify the perfect-mixing assumptions in each compartment, local Damköhler number (Da)
is analyzed. As suggested Figure 2, the bulk velocity inside the stirred tank is in the order of 10−2
m/s. Based on Equation (7), the upper bound of length scale in each compartment should be 1 m.
Furthermore, to justify neglecting diffusive mass transfer, Péclet number (Pe) is analyzed according to
Equation (8). Since diffusion coefficient in aqueous solutions are in the order of 10−9 m2 /s, the lower
Processes 2019, 7, x FOR PEER REVIEW 9 of 20
bound of compartment length scale is 10−7 m. It can be concluded that since in single phase turbulent
flow
flow convective
convectivemass masstransfer
transferraterate
is usually several
is usually ordersorders
several of magnitude higher than
of magnitude higher thatthan
of diffusion,
that of
compartmental model canmodel
diffusion, compartmental be safelycanadopted
be safelyinadopted
most single phase
in most stirred
single tankstirred
phase reactors.tank reactors.
Starting
Starting from
from the
the upper
upper bound
bound indicated
indicated byby the
the analysis
analysis of of Damköhler number, length
Damköhler number, length scale
scale of
of
the
the compartments is decreased to test the optimal grid density as discussed in Section 2.1.3. In
compartments is decreased to test the optimal grid density as discussed in Section 2.1.3. In this
this
work,
work, grid
grid independence
independence test test is
is performed
performed by by simulating
simulating the the injection of diazonium
injection of diazonium at at the
the top
top free
free
surface
surface of liquid near the wall. Considering that the injection should be fast enough to show mixing
of liquid near the wall. Considering that the injection should be fast enough to show mixing
effect,
effect, but
but not
not excessively
excessively fastfast so
so that
that the
the pyrazolone
pyrazolone is is instantly
instantly depleted
depleted andand the
the mixing-sensitive
mixing-sensitive
coupling
coupling is dominated by the decaying, the feeding rate of diazonium solution is set
is dominated by the decaying, the feeding rate of diazonium solution is set as
as 0.5
0.5 mL/s,
mL/s,
scaled
scaled from the work of Nienow [38]. The distribution of different chemical species at the end
from the work of Nienow [38]. The distribution of different chemical species at the end ofof
process
process is is predicted
predicted and and monitored.
monitored. The The total
total number
number of of compartments
compartments used used for
for the
the grid
grid sensitivity
sensitivity
test varied from
test varied from384384(8(8× ×8 ×86:×axial
6: axial × radial
× radial × angular)
× angular) to 2352to (122352
× 14(12 × 14
× 14). The×predicted
14). The amountpredicted of
amount of chemical species varies with the number of compartments
chemical species varies with the number of compartments and approaches asymptotic values as and approaches asymptotic
values
shown as in shown
Figure 3.in In
Figure
good3.agreement
In good agreement with the
with the scaling scalingconvergence
analysis, analysis, convergence
is achievediswith achieved
1920
with (12 × 16 ×
(12 × 16 × 10) compartments for all species, which corresponds to Da < 0.1. For fast model
1920 10) compartments for all species, which corresponds to Da < 0.1. For fast model
development,
development, Damköhler
Damköhler number
number can can served
served asas an
an efficient
efficient criterion
criterion for
for defining
defining grids
grids [30].
[30]. Further
Further
results presented in this paper are based on this discretization
results presented in this paper are based on this discretization scheme. scheme.
(a) (b)
0.012 0.047
Diazo Pyrazolone
0.046
Amount of Pyrazolone in Tank (mol)
0.0115
Amount of Diazo in Tank (mol)
0.045
0.011
0.044
0.0105 0.043
0.042
0.01
0.041
0.0095
0.04
0.009 0.039
0 500 1000 1500 2000 2500 0 500 1000 1500 2000 2500
Number of Compartments Number of Compartments
Figure 3. Convergence
Convergenceofofpredicted
predicted(a)
(a)Diazonium
Diazoniumand
and(b)
(b)Pyrazolone
Pyrazolonedistribution with
distribution number
with of
number
compartments.
of compartments.
conditions where the feeding rate changes dynamically are studied and optimized. Dynamic policies
comprised of 2 and 3 feeding stages are discussed. Furthermore, by optimizing process design with
perfect-mixing assumption, traditional design is compared with this proposed methodology.
0.08
Feed near Rushton Impeller
Yield of Desired Product (mol)
0.06
0.04
0.02
0
0 50 100 150
Time (s)
Figure
Figure 4. Simulated
4. Simulated yield
yield of of
thethe desired
desired product
product when
when addition
addition location
location is the
is at at the bottom
bottom corner
corner and
and
near
near Rushton
Rushton impeller.
impeller.
The location
The locationforfor
feeding
feedingis is
then numerically
then numerically optimized
optimized with
with the proposed
the proposedcompartmental
compartmental model.
model.
AsAs
suggested
suggestedin in
Table
Table1, 1,
it was
it was found
foundthat irrespective
that irrespective of of
thethenumber
numberof of
stages, thethe
stages, maximum
maximum
productivity is is
productivity reached
reached when
when reagents areare
reagents injected
injectedat at
thethe
tiptip
of of
Rushton
Rushtonimpeller,
impeller,which
whichsuggests a a
suggests
higher mixing efficiency.
higher mixing efficiency.
Table 1. Comparison
Table 1. Comparison of optimal feeding
of optimal location
feeding forfor
location (a)(a)
constant rate
constant feeding
rate (b)(b)
feeding two-stage dynamic
two-stage dynamic
feeding policy
feeding and
policy (c)(c)
and three-stage feeding
three-stage policy.
feeding policy.
Optimal
Optimal Injection
Injection Location
Location
Reagent
Reagent Injection
Injection PolicyPolicyHeight
Height (m) (m) Radial Position
Radial Position (m)(m)
Constant Rate Feeding 0.1–0.13 0.22–0.25
Constant Rate Feeding 0.1–0.13 0.22–0.25
Two-stage Dynamic Feeding 0.1–0.13 0.22–0.25
Two-stage Dynamic Feeding 0.1–0.13 0.22–0.25
Three-stage Dynamic Feeding
Three-stage Dynamic Feeding 0.1–0.13
0.1–0.13 0.22–0.25
0.22–0.25
ProcessesThe
2019, optimal
7, 9 feeding rate profiles determined for different reagent injection policies are
11 of 20
illustrated in Figure 5. It can be concluded that the proposed methodology favors a decreasing
feeding rate profile, which leads to an increased process productivity. The reason behind this
profile, which leads
productivity boostto
is an increased
studied process
through productivity.
process dynamics.The
Asreason behind
illustrated this productivity
in Table boost is
2, approximately 4%
studied through process dynamics. As illustrated in Table 2, approximately
increase in process productivity is achieved by implementing a dynamic operation.4% increase in process
productivity is achieved by implementing a dynamic operation.
3 Constant rate
Feeding Rate of Diazo (mL/s)
Two-stage
2.5 Three-stage
1.5
0.5
0
0 50 100 150
Time (s)
Figure5.5.Optimal
Figure Optimalfeeding
feedingrate
rateprofile
profilesolved
solvedfor
forconstant
constantrate,
rate,two-stage
two-stageand
andthree-stage
three-stagereagent
reagent
injectionpolicies.
injection policies.
Table 2. Comparison of optimal process productivity when (a) constant rate feeding (b) two-stage
Table 2. Comparison of optimal process productivity when (a) constant rate feeding (b) two-stage
dynamic feeding policy and (c) three-stage feeding policy are adopted.
dynamic feeding policy and (c) three-stage feeding policy are adopted.
Reagent
ReagentInjection Policy
Injection Policy Optimal
OptimalProcess
ProcessProductivity
Productivity($)
($)
Constant
Constant Rate
Rate Feeding
Feeding 6162.90
6162.90
Two-stage Dynamic
Two-stage DynamicFeeding
Feeding 6410.43
6410.43
Three-stage Dynamic
Three-stage DynamicFeeding
Feeding 6411.76
6411.76
Simulatedtrajectories
Simulated trajectoriesofofchemical
chemicalspecies
specieswhenwhendifferent
differentinjection
injectionpolicies
policiesare
areemployed
employedisis
illustrated in Figure 6. It is suggested that through employing dynamic policies, the yield ofofdesired
illustrated in Figure 6. It is suggested that through employing dynamic policies, the yield desired
productisisnot
product notsignificantly
significantlyenhanced
enhanced(Figure
(Figure6c),
6c),which
whichcan canbebeexplain
explainbybythetheway
wayweweformulate
formulatethis
this
problem. Since the price of the desired product is 10 times as high as the price
problem. Since the price of the desired product is 10 times as high as the price of diazonium, through of diazonium, through
theeffort
the effort
to to maximize
maximize thethe overall
overall profit,
profit, sufficient
sufficient diazonium
diazonium is fedistofed to exhaust
exhaust pyrazolone,
pyrazolone, whichwhich
lead
lead to similar yield of
to similar yield of the product.the product.
Nevertheless,dynamic
Nevertheless, dynamicfeedingfeedingrate
ratecan canimprove
improveprocess
processproductivity
productivitythrough
throughreducing
reducingthethe
wasteofofraw
waste rawmaterial.
material.As Assuggested
suggestedininFigureFigure6a,6a,considerable
considerableamount
amountofofdiazonium
diazoniumisiswasted
wastedifif
constant rate feeding policy is adopted. When reactant is fed at a constant rate,
constant rate feeding policy is adopted. When reactant is fed at a constant rate, with the consumption with the consumption
ofofpyrazolone,
pyrazolone,diazonium
diazoniumwould wouldinevitably
inevitablyaccumulate
accumulatefaster,
faster,which
whichleadleadtotomaterial
materialwaste
wastethat
that
compromiseseconomic
compromises economicperformance.
performance.By Byadjusting
adjustingfeeding
feedingraterateasaspyrazolone
pyrazoloneisisdeleted,
deleted,maximum
maximum
process profit can be achieved.
process profit can be achieved.
Processes 2019, 7, x FOR PEER REVIEW 12 of 20
Processes 2019, 7, 9 12 of 20
(a)
Constant rate
0.015 Two-stage
Three-stage
0.01
0.005
0
0 50 100 150
Time (s)
Amount of Pyrazolone in Tank (mol)
(b)
0.08
Constant rate
0.06 Two-stage
Three-stage
0.04
0.02
0
0 50 100 150
Time (s)
(c)
Yield of Desired Product (mol)
0.08
0.06
0.04
Constant rate
0.02 Two-stage
Three-stage
0
0 50 100 150
Time (s)
Figure 6. Simulated trajectories of (a) Diazo (b) Pyrazolone and (c) Desired product when optimal
feeding policies solved with different number of stages are employed.
7
Compartmental model
Feeding Rate of Diazo (mL/s)
0
0 50 100 150
Time (s)
FigureFigure
7. Optimal feeding
7. Optimal rate
feeding profile
rate profilesolved withCFD-based
solved with CFD-based compartmental
compartmental modelmodel and perfect
and perfect
mixingmixing model.
model.
Optimal feeding rate profiles determined with different methodologies are illustrated in Figure 7.
It is suggested that process design based on perfect-mixing assumption would lead to faster feeding
at earlier period of process. Therefore, a high quantity of diazonium is accumulated in the earlier
stage of process (Figure 8a). As a result, the undesired decomposition of diazonium is accelerated,
which compromise reaction selectivity (Figure 8d). Moreover, without considering the insufficient
consumption of pyrazolone due to imperfect mixing, the overall yield of desired product is hindered,
which further reduced product productivity.
Processes 2019, 7, x FOR PEER REVIEW 14 of 20
Processes 2019, 7, 9 14 of 20
(a)
0.01
0
0 50 100 150
Time (s)
Amount of Pyrazolone in Tank (mol)
(b)
0.08
Compartmental model
0.06 Perfect mixing model
0.04
0.02
0
0 50 100 150
Time (s)
(c)
0.08
Yield of Desired Product (mol)
0.06
0.04
0.02
Compartmental model
Perfect mixing model
0
0 50 100 150
Time (s)
Figure 8. Simulated trajectories of (a) Diazonium (b) Pyrazolone (c) Desired product when different
methodologies are employed.
4. Discussion
Figure 8. Simulated trajectories of (a) Diazonium (b) Pyrazolone (c) Desired product and (d)
undesired product when different methodologies are employed.
Mixing in turbulent flow can significantly influence reaction selectivity, therefore systematic
analysis of mass transfer inside reactors is crucial for process design. Despite the increasing
4. Discussion
computational power available for gaining understanding of the mixing process, in-silico process
design Mixing
can stillin
beturbulent
inefficientflow can and
in time significantly influence
cost. In this reaction
proposed selectivity,
framework, thereforerepetitive
by replacing systematic
analysisCFD
dynamic of mass transfer
simulation withinside reactors ismodel,
compartmental crucialprocess
for process
designdesign. Despite the
can be conducted in aincreasing
timely
Processes 2019, 7, 9 15 of 20
and economically more efficient manner. Moreover, in this work surrogate-based optimization is
implemented to numerically optimize process productivity. Comparing to genetic algorithm, which is
most widely adopted in engineering design, surrogate–based method requires less simulations to find
the optimal process design. As a result, the overall time efficiency can be significantly improved.
Rößger and Richter [25] have thoroughly compared state of the art CFD-based optimization methods
based on a 2-parameter design problem. For that specific problem it has been demonstrated that direct
optimization based on CFD simulation would require 1000 CFD simulations, which is computationally
prohibitive. Hybrid methods based on data-driven models can reduce the required number of CFD
simulations. However, as have been discussed, a large number of simulations are required to sufficiently
train the data-driven model and prevent over-fitting. The reported computational time is 6.5 h running
parallel on a 20-core Intel Xeon E5 V2 2.8 GHz processor.
In this manuscript, to optimize the process operation, 8 decision variables are considered. By
implementing an efficient iterative SBO algorithm, 150 iterations are sufficient for the algorithm to
converge to a good solution. Moreover, since each iteration is based on the compartmental model, which
only require a single CFD simulation to construct. As presented in Table 4, single simulations were
performed to determine the computational effort for one design evaluation. The overall computational
time is 3 h, running on a single-core of an Intel Xeon E5 V3 workstation. It can be concluded that
by implementing CFD-based compartmental modeling, significant time saving could be achieved if
same hardware system is applied. In active pharmaceutical ingredient (API) plant design where a
large number of simulations are conducted, computational expense could be reduced from weeks
to h through the implementation of the proposed methodology. Moreover, considering that CFD
is not freely available and requires special know-how, by cutting back dynamics CFD simulations,
implementing compartmental modeling is economically beneficial. This approach has shown a good
potential to characterize mixing in all the reactors in an API plant.
Table 4. Comparing computational expense between the proposed methodology and direct CFD simulation.
Despite the reduced model complexity, in this work the grid density is determined so that
inhomogeneity inside reactor is sufficiently captured. Comparing to traditional process design
strategies which are based on perfect-mixing assumption, better reaction selectivity could be achieved
through compartmental model. As has been discussed in sub Section 3.5.3, by replacing traditional
process design strategy with this proposed compartmental model, more than 10% increase in process
profit has been achieved (Table 5).
Table 5. Comparison between optimal operating conditions solved with perfect mixing model and the
proposed methodology.
Furthermore, this proposed framework allows the design of dynamic operations. As illustrated in
sub Section 3.5.2, by adapting feeding rate in time to account for the depletion of raw materials, material
waste can be substantially reduced, which in turn leads to improved process productivity. Considering
that for complex reaction networks commonly encountered in organic synthesis, process dynamics
could be very complicated. By enabling the design of dynamic operations to fit the evolving chemical
processes, this proposed framework has exhibited a great potential of productivity improvement.
Processes 2019, 7, 9 16 of 20
5. Conclusions
CFD is a powerful tool to study mixing in chemical processes, and as such it has been applied
widely for numerical optimization of reactor designs. Nevertheless, implementation of CFD to improve
mixing through optimization of process operation is limited. A key reason is the computational
complexity of CFD models. Even though some data-driven models are used to replace CFD to reduce
time expense, to build confidence in the data-driven models requires large amount of CFD simulation.
The implementation of meta-heuristic algorithms further increased the computational expense.
This paper has shown possible ways to address the inherent difficulty with two improvements.
First, instead of using data-driven models to represent CFD simulation, compartmental model is
implemented. Since compartmental model are built based on first-principle mass balance equations, it
requires significantly less CFD simulations to drive, making it a better compromise between computational
complexity and model accuracy. Second, GA is replaced with surrogate-based optimization based on
RBF, which has been proven to be more efficient than GA. It should be noted that this surrogate-based
optimization algorithm does not give global optimality guarantees, same as GA. The goals should be to
find “good” or near optimal solutions when enough resources are provided.
The surrogate-based compartmental model optimization presented a compact and efficient
structure, which can be easily implemented in other scenarios. Since in compartmental model
space is discretized, which leads to mixed-integer nonlinear programing problems, surrogate-based
optimization is an efficient way to address it. Moreover, compartmental model is built based on
steady-state hydrodynamic solution from CFD, which is independent from models inside each
compartment. Therefore, changing models in compartments does not require extra CFD simulations.
The proposed methodology allows for dynamic process operation design, which has shown a
great potential of productivity improvement. Moreover, dynamic optimization of process operation
improves process flexibility and agility to adjust to a more dynamic market. Finally, the developed
simplification of complex transport model has the potential for advanced process monitoring and
control for reactors with limited instrumentation, such as single-use bioreactors and fermenters.
It should be noted that, to leverage the result of CFD simulation, compartmental models are
constructed based on steady-state hydrodynamic solution. However, this simplification is based on
the assumption that the flow field is independent from chemical reaction. This assumption would
limit the application of the proposed method in reactive flows. In addition, experimental calibration of
CFD simulation is necessary build confidence in the predictive ability of the proposed methodology.
Author Contributions: Under supervision of M.I. and S.K., S.Y. was responsible for development of research
methodology and optimization framework, model development, optimization and analysis of the results. P.F.
contributed in model development.
Funding: This research received no external funding.
Acknowledgments: The authors gratefully acknowledge financial support from Gilead Science Inc. We also
appreciate the discussions and technical inputs from Chiajen Lai of Gilead Science Inc.
Conflicts of Interest: The authors declare no conflicts of interest.
Abbreviations
CPP Critical process parameters
CFD Computational fluid dynamics
QbD Quality by design
RTD Residence time distribution
GA Genetic algorithm
RBF Radial basis function
CMC Chemistry, manufacturing, and controls
ANN Artificial neural network
RANS Reynolds averaged Navier-Stocks
Processes 2019, 7, 9 17 of 20
References
1. Plutschack, M.B.; Pieber, B.; Gilmore, K.; Seeberger, P.H. The Hitchhiker’s Guide to Flow Chemistry(II).
Chem. Rev. 2017, 117, 11796–11893. [CrossRef]
2. Gobert, S.R.L.; Kuhn, S.; Braeken, L.; Thomassen, L.C.J. Characterization of Milli- and Microflow Reactors:
Mixing Efficiency and Residence Time Distribution. Org. Process. Res. Dev. 2017, 21, 531–542. [CrossRef]
3. Fogler, H.S. Essentials of Chemical Reaction Engineering; Pearson Education: Upper Saddle River, NJ, USA,
2010.
4. Gresch, M.; Brugger, R.; Meyer, A.; Gujer, W. Compartmental Models for Continuous Flow Reactors Derived
from CFD Simulations. Environ. Sci. Technol. 2009, 43, 2381–2387. [CrossRef]
5. Nienow, A.W. On impeller circulation and mixing effectiveness in the turbulent flow regime. Chem. Eng. Sci.
1997, 52, 2557–2565. [CrossRef]
6. Rosseburg, A.; Fitschen, J.; Wutz, J.; Wucherpfennig, T.; Schluter, M. Hydrodynamic inhomogeneities in
large scale stirred tanks–Influence on mixing time. Chem. Eng. Sci. 2018, 188, 208–220. [CrossRef]
7. Levesque, F.; Bogus, N.J.; Spencer, G.; Grigorov, P.; McMullen, J.P.; Thaisrivongs, D.A.; Davies, I.W.; Naber, J.R.
Advancing Flow Chemistry Portability: A Simplified Approach to Scaling Up Flow Chemistry. Org. Process.
Res. Dev. 2018, 22, 1015–1021. [CrossRef]
8. Aubin, J.; Ferrando, M.; Jiricny, V. Current methods for characterising mixing and flow in microchannels.
Chem. Eng. Sci. 2010, 65, 2065–2093. [CrossRef]
9. Commenge, J.M.; Falk, L. Villermaux-Dushman protocol for experimental characterization of micromixers.
Chem. Eng. Process. 2011, 50, 979–990. [CrossRef]
10. Reckamp, J.M.; Bindels, A.; Duffield, S.; Liu, Y.C.; Bradford, E.; Ricci, E.; Susanne, F.; Rutter, A. Mixing
Performance Evaluation for Commercially Available Micromixers Using Villermaux-Dushman Reaction
Scheme with the Interaction by Exchange with the Mean Model. Org. Process. Res. Dev. 2017, 21, 816–820.
[CrossRef]
11. Cheng, D.; Feng, X.; Cheng, J.C.; Yang, C. Numerical simulation of macro-mixing in liquid-liquid stirred
tanks. Chem. Eng. Sci. 2013, 101, 272–282. [CrossRef]
Processes 2019, 7, 9 18 of 20
12. Liu, L.; Barigou, M. Experimentally Validated Computational Fluid Dynamics Simulations of Multicomponent
Hydrodynamics and Phase Distribution in Agitated High Solid Fraction Binary Suspensions. Ind. Eng. Chem. Res.
2014, 53, 895–908. [CrossRef]
13. Reinecke, S.F.; Deutschmann, A.; Jobst, K.; Hampel, U. Macro-mixing characterisation of a stirred model
fermenter of non-Newtonian liquid by flow following sensor particles and ERT. Chem. Eng. Res. Des. 2017,
118, 1–11. [CrossRef]
14. Warmeling, H.; Behr, A.; Vorholt, A.J. Jet loop reactors as a versatile reactor set up—Intensifying catalytic
reactions: A review. Chem. Eng. Sci. 2016, 149, 229–248. [CrossRef]
15. Farzan, P.; Ierapetritou, M.G. Integrated modeling to capture the interaction of physiology and fluid dynamics
in biopharmaceutical bioreactors. Comput. Chem. Eng. 2017, 97, 271–282. [CrossRef]
16. Foli, K.; Okabe, T.; Olhofer, M.; Jin, Y.C.; Sendhoff, B. Optimization of micro heat exchanger: CFD, analytical
approach and multi-objective evolutionary algorithms. Int. J. Heat Mass Transf. 2006, 49, 1090–1099.
[CrossRef]
17. Uebel, K.; Rößger, P.; Prüfert, U.; Richter, A.; Meyer, B. CFD-based multi-objective optimization of a quench
reactor design. Fuel Process. Technol. 2016, 149, 290–304. [CrossRef]
18. Chen, M.; Wang, J.; Zhao, S.; Xu, C.; Feng, L. Optimization of Dual-Impeller Configurations in a Gas–Liquid
Stirred Tank Based on Computational Fluid Dynamics and Multiobjective Evolutionary Algorithm. Ind. Eng.
Chem. Res. 2016, 55, 9054–9063. [CrossRef]
19. Na, J.; Kshetrimayum, K.S.; Lee, U.; Han, C. Multi-objective optimization of microchannel reactor for
Fischer-Tropsch synthesis using computational fluid dynamics and genetic algorithm. Chem. Eng. J. 2017,
313, 1521–1534. [CrossRef]
20. De-Sheng, Z.; Jian, C.; Wei-Dong, S.H.I.; Lei, S.H.I.; Lin-Lin, G. Optimization of hydrofoil for tidal current
turbine based on particle swarm optimization and computational fluid dynamic ethod. Thermal Sci. 2016, 20,
907–912.
21. Sierra-Pallares, J.; del Valle, J.G.; Paniagua, J.M.; Garcia, J.; Mendez-Bueno, C.; Castro, F. Shape optimization
of a long-tapered R134a ejector mixing chamber. Energy 2018, 165, 422–438. [CrossRef]
22. Brar, L.S.; Elsayed, K. Analysis and optimization of cyclone separators with eccentric vortex finders using
large eddy simulation and artificial neural network. Sep. Purif. Technol. 2018, 207, 269–283. [CrossRef]
23. Jung, I.; Kshetrimayum, K.S.; Park, S.; Na, J.; Lee, Y.; An, J.; Park, S.; Lee, C.-J.; Han, C. Computational
Fluid Dynamics Based Optimal Design of Guiding Channel Geometry in U-Type Coolant Layer Manifold of
Large-Scale Microchannel Fischer–Tropsch Reactor. Ind. Eng. Chem. Res. 2016, 55, 505–515. [CrossRef]
24. Park, S.; Na, J.; Kim, M.; Lee, J.M. Multi-objective Bayesian optimization of chemical reactor design using
computational fluid dynamics. Comput. Chem. Eng. 2018, 119, 25–37. [CrossRef]
25. Rößger, P.; Richter, A. Performance of different optimization concepts for reactive flow systems based on
combined CFD and response surface methods. Comput. Chem. Eng. 2018, 108, 232–239. [CrossRef]
26. Abu-Mostafa, Y.S.; Magdon-Ismail, M.; Lin, H.-T. Learning from Dat; AMLBook: New York, NY, USA, 2012;
Volume 4.
27. Kapsoulis, D.; Tsiakas, K.; Trompoukis, X.; Asouti, V.; Giannakoglou, K. A PCA-assisted hybrid algorithm
combining EAs and adjoint methods for CFD-based optimization. Appl. Soft. Comput. 2018, 73, 520–529.
[CrossRef]
28. Muller, J.; Shoemaker, C.A.; Piche, R. SO-MI: A surrogate model algorithm for computationally expensive
nonlinear mixed-integer black-box global optimization problems. Comput. Oper. Res. 2013, 40, 1383–1400.
[CrossRef]
29. Boltersdorf, U.; Deerberg, G.; SCHLÜTER, S. Computational study of the effects of process parameters on
the product distribution for mixing sensitive reactions and on distribution of gas in stirred tank reactors.
Recent Res. Dev. Chem. Eng. 2000, 4, 15–43.
30. Guha, D.; Dudukovic, M.P.; Ramachandran, P.A.; Mehta, S.; Alvare, J. CFD-based compartmental modeling
of single phase stirred-tank reactors. AlChE J. 2006, 52, 1836–1846. [CrossRef]
31. Nørregaard, A.; Bach, C.; Krühne, U.; Borgbjerg, U.; Gernaey, K.V. Hypothesis-driven compartment model
for stirred bioreactors utilizing computational fluid dynamics and multiple pH sensors. Chem. Eng. J. 2019,
356, 161–169. [CrossRef]
Processes 2019, 7, 9 19 of 20
32. Zhao, W.; Buffo, A.; Alopaeus, V.; Han, B.; Louhi-Kultanen, M. Application of the compartmental model to
the gas-liquid precipitation of CO2 -Ca(OH)2 aqueous system in a stirred tank. AlChE J. 2017, 63, 378–386.
[CrossRef]
33. Bezzo, F.; Macchietto, S.; Pantelides, C.C. A general methodology for hybrid multizonal/CFD models.
Comput. Chem. Eng. 2004, 28, 501–511. [CrossRef]
34. Vrabel, P.; van der Lans, R.; Cui, Y.Q.; Luyben, K. Compartment model approach: Mixing in large scale
aerated reactors with multiple impellers. Chem. Eng. Res. Des. 1999, 77, 291–302. [CrossRef]
35. Du, J.; Johansen, T.A. Integrated Multilinear Model Predictive Control of Nonlinear Systems Based on Gap
Metric. Ind. Eng. Chem. Res. 2015, 54, 6002–6011. [CrossRef]
36. Srilatha, C.; Morab, V.V.; Mundada, T.P.; Patwardhan, A.W. Relation between hydrodynamics and drop size
distributions in pump–mix mixer. Chem. Eng. Sci. 2010, 65, 3409–3426. [CrossRef]
37. Bezzo, F.; Macchietto, S. A general methodology for hybrid multizonal/CFD models—Part II. Automatic
zoning. Comput. Chem. Eng. 2004, 28, 513–525. [CrossRef]
38. Nienow, A.W.; Drain, S.M.; Boyes, A.P.; Mann, R.; El-Hamouz, A.M.; Carpenter, K.J. A new pair of reactions
to characterize imperfect macro-mixing and partial segregation in a stirred semi-batch reactor. Chem. Eng. Sci.
1992, 47, 2825–2830. [CrossRef]
39. Conn, A.R.; Scheinberg, K.; Vicente, L.N. Introduction to Derivative-Free Optimization; Siam: Philadelphia, PA,
USA, 2009.
40. Boukouvala, F.; Misener, R.; Floudas, C.A. Global optimization advances in Mixed-Integer Nonlinear
Programming, MINLP, and Constrained Derivative-Free Optimization, CDFO. Eur. J. Oper. Res. 2016, 252,
701–727. [CrossRef]
41. Bhosekar, A.; Ierapetritou, M. Advances in surrogate based modeling, feasibility analysis, and optimization:
A review. Comput. Chem. Eng. 2018, 108, 250–267. [CrossRef]
42. Regis, R.G.; Shoemaker, C.A. Improved strategies for radial basis function methods for global optimization.
J. Glob. Optim. 2007, 37, 113–135. [CrossRef]
43. Oeuvray, R.; Bierlaire, M. BOOSTERS: A derivative-free algorithm based on radial basis functions. Int. J.
Model. Simul. 2009, 29, 26–36. [CrossRef]
44. Wild, S.M.; Shoemaker, C.A. Global Convergence of Radial Basis Function Trust-Region Algorithms for
Derivative-Free Optimization. SIAM Rev. 2013, 55, 349–371. [CrossRef]
45. Regis, R.G.; Wild, S.M. CONORBIT: constrained optimization by radial basis function interpolation in trust
regions. Optim. Meth. Softw. 2017, 32, 552–580. [CrossRef]
46. Wang, Z.L.; Ierapetritou, M. A novel feasibility analysis method for black-box processes using a radial basis
function adaptive sampling approach. AlChE J. 2017, 63, 532–550. [CrossRef]
47. Jones, D.R.; Schonlau, M.; Welch, W.J. Efficient global optimization of expensive black-box functions.
J. Glob. Optim. 1998, 13, 455–492. [CrossRef]
48. Boukouvala, F.; Ierapetritou, M.G. Derivative-free optimization for expensive constrained problems using a
novel expected improvement objective function. AlChE J. 2014, 60, 2462–2474. [CrossRef]
49. Regis, R.G. Trust regions in Kriging-based optimization with expected improvement. Eng. Optim. 2016, 48,
1037–1059. [CrossRef]
50. Beykal, B.; Boukouvala, F.; Floudas, C.A.; Sorek, N.; Zalavadia, H.; Gildin, E. Global optimization of grey-box
computational systems using surrogate functions and application to highly constrained oil-field operations.
Comput. Chem. Eng. 2018, 114, 99–110. [CrossRef]
51. Wang, Z.L.; Ierapetritou, M. Constrained optimization of black-box stochastic systems using a novel
feasibility enhanced Kriging-based method. Comput. Chem. Eng. 2018, 118, 210–223. [CrossRef]
52. Fernandes, F.A.N. Optimization of Fischer-Tropsch synthesis using neural networks. Chem. Eng. Technol.
2006, 29, 449–453. [CrossRef]
53. Henao, C.A.; Maravelias, C.T. Surrogate-Based Superstructure Optimization Framework. AlChE J. 2011, 57,
1216–1232. [CrossRef]
54. Sen, O.; Gaul, N.J.; Choi, K.K.; Jacobs, G.; Udaykumar, H.S. Evaluation of multifidelity surrogate modeling
techniques to construct closure laws for drag in shock-particle interactions. J. Comput. Phys. 2018, 371,
434–451. [CrossRef]
55. Müller, J. MISO: mixed-integer surrogate optimization framework. Optim. Eng. 2015, 17, 177–203. [CrossRef]
Processes 2019, 7, 9 20 of 20
56. Zhang, W.P.; Wang, X.; Feng, X.; Yang, C.; Mao, Z.S. Investigation of Mixing Performance in Passive
Micromixers. Ind. Eng. Chem. Res. 2016, 55, 10036–10043. [CrossRef]
57. Lin, X.Y.; Wang, K.; Zhang, J.S.; Luo, G.S. Liquid-liquid mixing enhancement rules by microbubbles in three
typical micro-mixers. Chem. Eng. Sci. 2015, 127, 60–71. [CrossRef]
58. Fluent, A. Ansys Fluent Theory Guide; ANSYS Inc.: Canonsburg, PA, USA, 2011; Volume 15317, pp. 724–746.
© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).