CH06CH16-Joshi ARI 15 July 2015 7:42

ANNUAL
REVIEWS Further Computational Modeling of
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

Click here for quick links to

Annual Reviews content online,
including:
Multiphase Reactors
• Other articles in this volume
• Top cited articles
• Top downloaded articles
J.B. Joshi1 and K. Nandakumar2
• Our comprehensive search 1
DAE, Homi Bhabha National Institute, Mumbai 400 094, India; email: [email protected]
2
Cain Department of Chemical Engineering, Louisiana State University, Baton Rouge,
Louisiana 70803

Annu. Rev. Chem. Biomol. Eng. 2015. 6:347–78 Keywords

First published online as a Review in Advance on computational fluid dynamics, computational chemistry, gas-liquid,
July 2, 2015
gas-solid, liquid-solid, gas-liquid-solid reactors
The Annual Review of Chemical and Biomolecular
Engineering is online at chembioeng.annualreviews.org Abstract
This article’s doi: Multiphase reactors are very common in chemical industry, and numerous
10.1146/annurev-chembioeng-061114-123229
review articles exist that are focused on types of reactors, such as bubble
Copyright  c 2015 by Annual Reviews. columns, trickle beds, fluid catalytic beds, etc. Currently, there is a high
All rights reserved
degree of empiricism in the design process of such reactors owing to the
complexity of coupled flow and reaction mechanisms. Hence, we focus on
synthesizing recent advances in computational and experimental techniques
that will enable future designs of such reactors in a more rational manner
by exploring a large design space with high-fidelity models (computational
fluid dynamics and computational chemistry models) that are validated with
high-fidelity measurements (tomography and other detailed spatial mea-
surements) to provide a high degree of rigor. Understanding the spatial dis-
tributions of dispersed phases and their interaction during scale up are key
challenges that were traditionally addressed through pilot scale experiments,
but now can be addressed through advanced modeling.

347
 CH06CH16-Joshi ARI 15 July 2015 7:42

INTRODUCTION
Numerous textbooks (1, 2) describe the theory of reaction kinetics, reaction engineering, and
reactor design of the so-called ideal reactors. Kinetic and equilibrium models must necessarily
focus on the molecular processes (3) involving the interaction of electron clouds in bond breaking
and formation mechanisms (4), whereas reaction engineering and reactor design must also take
into account the impact of flow environment (5) that brings the reactant molecules together in
their natural state under reaction conditions (solid, liquid, or gas) for a chemical reaction to oc-
cur. Although pedagogically sound and simple, the introductory textbooks avoid dealing with the
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

consequences of complexities of reactor configurations or the complex and often heterogeneous,

multiphase nature of the flow environment. Homogeneous flow regimes in any multiphase reactor
(bubble columns and fluidized beds being typical examples) are seldom maintained owing to flow
instabilities, and one is forced to contend with complex, heterogeneous flow regimes. The scale of
heterogeneity can vary from the dispersed phase scale to the equipment scale; thus, during scale-up,
newer, larger-scale structures can emerge that were filtered out in smaller pilot-scale equipment.
Hence, in the practice of multiphase reactor design, a large element of uncertainty still remains,
making the reactor design process an art. These are still subjects of intense research in multiphase
flow reactor dynamics. But significant progress has been made in the areas of computational chem-
istry (4) to understand the reaction processes at the molecular level and computational transport
phenomena (6) to understand the flow, temperature, and concentration fields at the continuum
level, and a synthesis of the two fields should enable process innovation in the area of multiphase
reactor design. Although a new framework is emerging, challenges remain in upscaling of infor-
mation from high-fidelity simulations and experiments that provide very detailed information to
performance models for specific large-scale reactors that require only traditional effective kinetic
parameters, effective dispersion coefficients, residence time distribution (RTD), and effective heat
and mass transfer coefficients. It is the performance models backed up by high-fidelity data from
simulations and experiments that will allow reliable exploration of a larger design space for optimal
novel design of such multiphase reactors to achieve better performance.
There are hundreds of reactor configurations in use, including simple continuous stirred tank
reactors (CSTR) (7, 8), plug flow reactors (PFR), trickle bed reactors (9), moving bed reactors
(10), slurry loop reactors (11), fluidized bed reactors (12–14), jet reactors, air-lift reactors (15),
bubble column reactors (16, 17), and bubble-slurry column reactors (18). It is not our intention to
provide a comprehensive review of the existing design practices of every type of multiphase reactor
in use, or the application domains of such reactors, because several traditional textbooks, recent
research monographs, and review articles cited above fulfill that need adequately (see also relevant
literature, e.g., References 19, 20). Historical design procedures of such multiphase reactors are
fraught with a high degree of empiricism. Hence, we focus on reviewing the emerging, unifying
trends in the design process of such process vessels by understanding the mechanisms on various
scales, from the molecular scale, where intrinsic reaction rates must be understood using com-
putational chemistry tools [density functional theories (DFT) (21, 22) and molecular dynamics
(MD) (23, 24)], to micro- and mesoscales, where micro-macro-mixing must be understood using
computational fluid dynamics (CFD) tools, including discrete particle models (DPM) (25) and
direct numerical simulations (DNS) (26), to equipment-scale performance models. However, the
current equipment-scale design procedures are mostly the property of developers of such reactors
owing to the expensive pilot tests needed to develop the effective parameters for each class of re-
actors, whereas ASPEN/HYSYS (27) provides a generic framework for plantwide process models.
Each of these areas has a vast literature including focused review articles specific to each area, such
as DFT, MD, DNS, DPM, or CFD. Our goal is to focus on synthesizing these developments in a

348 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

Multiscale, multiphase, multiphysics

Nature of process operations offers the main challenge

Pilot-scale Lumped, homogeneous, input/output models Process

experiments performance
Equipment scale (ASPEN/HYSYS)

Σm =0
i

Lab experiments Tray efficiency overall heat/mass transfer

Volume-averaged equations: Eulerian-Eulerian models
Time-averaged equations: turbulence models
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

Eulerian-Lagrangian models: discrete particle models

Interpenetrating continua scale (FLUENT/MFiX)

Lab experiments Drag/lift forces interface heat/mass transfer

Fully resolved direct numerical simulation of
turbulence and phase interactions Dυ
ρ = –Vp + ρg + μ Vv2;
Navier-Stokes equation Dt
Continuum scale (FLUENT/OpenFOAM/MFiX)

Lab experiments Material properties viscosity, density, surface tension

MD simulation
Molecular scale (GROMACS) m
du
dt
= ΣF
Lab experiments Intermolecular interaction potential

Figure 1
Interrelations between models on various scales from the top, performance level models to the fundamental
molecular dynamics models, illustrates dependence among the parameterization at various scales and
pathways to get at these parameters.

coherent manner to enable a paradigm shift in the design of future multiphase chemical reactors by
combining advances in experimental and computational techniques and integrating or upscaling
information from the molecular scale to the equipment scale. Figure 1 shows the interrelations
between the models at various scales and the parameters that are needed as input at any chosen
scales, as well as how they can be obtained from a higher-resolution model.
The proposed paradigm shift in the design process that envisions exploring novel designs using
high-fidelity computational models (instead of extensive and expensive pilot-scale experiments)
may yet be wishful thinking, but rapid advances in computational, measurement, and manufac-
turing technologies at all scales will make this a reality sooner rather than later, and perhaps the
newer reactor designs (in chemical, energy, biotechnology, minerals, and materials processing
industries) will embrace these developments more readily.

EVOLVING TRENDS IN PROCESS DESIGN FROM ANCIENT

TO MODERN TIMES
Using techniques of separation, mixing, and reaction of chemical species, chemical engineers
produce a large range of useful products that enrich our lives. In that process, any unwanted
materials that are rejected must also be handled to mitigate environmental problems by taking a
holistic view of the design and manufacturing processes. The useful products range from fuels,
fertilizers, petrochemicals, polymers, fibers and textiles, soaps and detergents, processed foods,

www.annualreviews.org • Multiphase Reactors 349

 CH06CH16-Joshi ARI 15 July 2015 7:42

Table 1 Chronological stages of development of process equipment design

Stages Achievements Tools and techniques
5000 BC to Extraction of metals, edible and nonedible oils, natural Apprenticeship-based incremental evolutionary
1900 AD materials, and their medicinal formulations; making of design; design by analogy (if it works in one place, it
paper and cloth, pigments, fermentation, food may work in similar circumstances)
processing, and a variety of construction materials
1901 to 1950 Evolution of the concepts of unit operations, e.g., Algebraic equations as models; graphical methods for
distillation, absorption, extraction, crystallization, and nonlinear problems, nomograms; data from pilot
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

adsorption; started with George Davis (29) plants used in design

1901 to 1950 Evolution of the concepts of unit processes [Groggins Procedures were established for predicting
(30)], e.g., oxidation, hydrogenation, alkylation, physicochemical properties, phase equilibria, and
acylation, nitration, sulfonation, esterification, hydrolysis, reaction equilibria; measurement of kinetics of
carbonylation, ammonolysis, and ozonolysis single-phase reaction and lumped kinetics of
multiple reactions
1920 to 1950 Large-scale processing of petroleum crude and Innovations of reactors and processes happened on
petrochemicals led to the development of a variety of tray laboratory scale followed by validation on pilot
designs; catalytic cracking was introduced together with scale; several multiphase reactor designs were first
the large-scale use of fluidized and circulating fluidized commercialized, and subsequent attempts were
beds; gasoline desulfurization and other catalytic made to understand their hydrodynamics
processes saw the development of fixed bed catalytic
reactors, pressure swing absorption, simulated moving
bed chromatography, and bubble columns for
fermentation to give antibiotics
1930 to 2000 Development of theories of heat and mass and momentum Mathematical models (differential equations) as
transfer, including their analogies, eddy viscosity models, unifying force became dominant; empirical
and surface renewal; theories of Higbie and Danckwerts correlations on the equipment scale with data from
(31); theories of similitude, modeling, and dimensional pilot plants provided the parameters in the models;
analysis with a large number of correlations involving Re, development of tools such as ASPEN and HYSIS
Pr, Sc, Nu, and Sh were developed became possible with advent of computer
1980 to Fundamental understanding of the following: High-performance computing and high-fidelity
current (a) chemical kinetics on the basis of molecular mechanisms; measurements provided deep insight into transport
(b) detailed flow profiles, k −  profile, energy spectra, and and reaction processes; density functional theory;
other turbulence parameters could be determined; molecular dynamics; experimental fluid dynamics;
(c) bubble/drop size distributions, cluster size distribution in-situ kinetic measurements; computational fluid
and dynamics, and flow structures and their impact on dynamics; discrete particle modeling;
transparent phenomena could be understood; and population balance models
(d ) interface transport mechanisms could be determined

construction materials, and advanced materials to specialty chemicals, such as pharmaceuticals,

agrochemicals, surface coatings, surface active agents, catalysts, perfumes, flavors, and cosmetics.
More than 50 million chemicals are known to mankind, and over the past 12 months, a new
chemical has been either synthesized or found every 2.6 s (28). These products touch every aspect
of our lives. A historical perspective on how the design and operation of such processes have
evolved from apprentice-based training more than 5,000 years ago to modern computer-based
design is shown in Table 1.
Table 1 shows a large number of innovations in the development of a variety of chemical
processes, as well as the analytical/computational tools used for understanding their performance.
Most of these designs were enabled through extensive pilot-scale experimentation and develop-
ment of specific empirical correlations for such performance measures as height equivalent of

350 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

theoretical plate, plate efficiency, and mass and heat transfer coefficients. Such empiricism has
several limitations: (a) It assumes some sort of lumped/homogeneous behavior inside process ves-
sels and ignores the macro- and microscale heterogeneities, such as phase fractions, mean and
turbulent velocity fields, turbulent kinetic energy, and energy dissipation rates; (b) the flow pat-
tern in the vicinity of bubble (drop or particle) may get modified by small amounts of adventitious
surface active impurities, the concentration of which cannot be measured with the present art of
instrumentation; (c) a separate empirical correlation is typically needed for each design parameter
and piece of equipment; and (d ) model fluids, such as air, water, sand, and glass beads, are used
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

for the data collection on the pilot scale, thus making it difficult to preserve kinematic, dynamic,
and geometric similarities between the pilot and field scale processes.
In modern times, such empiricism is giving way to systematic innovations in process design.
These innovations typically begin with the discovery of a chemical in a bench-scale experiment
by a chemist. The transition from a bench-scale batch process to a plant-scale continuous process
occurs through the ingenuity of an experienced chemical process design engineer via several pilot-
scale experiments to ensure that process efficiency and safety are maintained. Often multiphase
flow is involved at every step of the way, and it is currently the least understood phenomena.
So process design has remained an art, and hefty design margins are introduced to combat the
uncertainties in the process design methodology. Walas’s (32) classic 1970 book devoted to the
subject of Chemical Process Equipment: Selection & Design has gone through three editions, the latest
of which was published in 2012. It is clear from this work that the design process is still guided by a
combination of rules of thumb (evolving from experience), design by analogy (a philosophy that if
it worked in one scenario, it might work in a similar scenario), established principles (i.e., ASPEN
type of lumped parameter models that result in algebraic equations) and data (encapsulating
information and experience from pilot-scale experiments), and experience of the designer. Clearly
at the design stage, a large number of degrees of freedom are available, exercised at present in an
ad hoc manner, but which could be used to explore a large design space in a systematic manner to
develop future plant designs that are energy efficient and environmentally benign, if only the right
tools and design objectives are developed. A substantially resolved flow field on the equipment
scale as well as the dispersed phase scale (in the vicinity of individual bubbles, drops, or particles)
will avoid the heuristic interdependencies and empiricisms of design correlations developed based
only on input/output and gross geometrical parameters.

KEY DESIGN PARAMETERS FOR REACTION AND FLOW PROCESSES

IN MULTIPHASE REACTORS
Modern chemical reactors come in various forms, from simple CSTR to complex three-phase
slurry-bubble column reactors or fermenters. Even in simple CSTR the flow is never simple, as
the impeller design has a profound influence on the mixing patterns. The reaction engineering
community has developed the concepts of macromixing and micromixing (33) to address the flow-
related challenges separately and intrinsic kinetics models (3, 34) to address complex reaction
mechanisms separately. Understanding the development in both areas and coupling them will
provide new, rigorous tools for the design of modern multiphase reactors.

Understanding and Modeling Intrinsic Reaction Rate Mechanisms

Modeling the kinetics of reaction rates is a highly complex art that is slowly yielding to the power
of computational chemistry tools. But at present such tools are limited to a simple set of reaction
sequences. The intrinsic reaction rate models should focus, in principle, only on electron cloud

www.annualreviews.org • Multiphase Reactors 351

 CH06CH16-Joshi ARI 15 July 2015 7:42

interactions at the molecular level but describe and capture their impact at the continuum level
as a function of concentration and temperature. Such intrinsic rate expressions can be considered
as phenomenological models of reaction rates with kinetic parameters that are specific to each
reaction system that must be determined from carefully designed experiments in the absence of
any flow influences, or via high-resolution DFT simulations as probes. They are the most suitable
for reactor performance analysis models when integrated appropriately with the flow models. The
DFT (4, 35) is a high-fidelity computational tool that allows more detailed study of electron cloud
interactions and offers much more insight into the mechanisms as well as ways to extract the kinetic
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

parameters in the intrinsic kinetic models that can be used in performance-level reactor models,
providing a route for upscaling of data from the molecular scale to the reactor scale. Modular
development of simulation tools like CHEMKIN (36) and ANSYS (37) and their integration
with each other (CHEMKIN has recently been offered as part of the simulation software suite by
ANSYS!) opens up new possibilities for rigorous reactor designs taking the influence of flow and
reaction mechanisms with high fidelity.
At the continuum scale, a typical rate expression can be represented by (38)

r = k(T )φ(Ci ), 1.

where the temperature-dependent rate constant is often written in an Arrhenius form,

k(T ) = Ae −E/RT , 2.

where E is the activation energy, A is the preexponential constant, and R is the gas constant.
The expression for φ(Ci ) comes from the mechanistic reaction pathway hypothesis, and numerous
models have been proposed. Here, it is hypothesized that the rate dependence on temperature
and concentrations could be separated. Often simple models for concentration take the following
form:
 n
φ(Ci ) = Ci i , 3.
i

where ni is the order of the reaction with respect to the species i. This is a phenomenological
model that attempts to capture fundamental molecular processes based on the hypothesis that
the probability of two molecular species encountering each other to react depends on the molar
concentration. Clearly, effectiveness of mixing of two or more reactant species will influence the
effectiveness of the reaction, which is not taken into account in these models, but fluid dynamic
models provide that framework. In some cases, if the concentrations of some species are in excess,
they do not influence the rate expressions. In other cases, concentration of reaction intermediates
and catalyzing species that do not appear in the stoichiometric balance equation of the reaction
can influence the rates profoundly. Hence, Equation 3 is not the only form possible. For example,
in the case of a gas phase reaction of H 2 and Br2 to form HBr, the following rate expression, based
on a certain mechanistic model, has been proposed (38):
1
k(T )C H 2 (C Br2 ) 2
r(Ci , T ) = , 4.
1 + k (T ) CCHBr
Br2

where (k, k ) are constants in the rate expressions at a given temperature. Experimentally deter-
mined concentration versus time data at various reactor conditions are then used to estimate the
parameters, such as {k, A, E, ni }. It must be emphasized that such experiments must be conducted
in conditions that are not influenced by transport limitations. Otherwise, the parameter estimates
get corrupted by flow influence and lose their intrinsic nature and scale invariance with flow or
reactor geometry.

352 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

CHEMKIN (36, 39–41) has emerged as a useful tool to explore thermochemistry and reaction
mechanisms of complex reaction networks in terms of simple elementary reaction steps. Once a
list of atomic or molecular species (including any intermediates) that are likely to be involved in a
reaction system is identified as input to the computer simulator, it can aid both in thermochemical
calculations and in identifying potential reaction pathways. It is a collection of FORTRAN sub-
routines supported by a vast database of thermochemical and transport properties that the user
can call to automate routine calculations of thermochemical properties (such as heats of reaction
and free energies) to explore complex reaction network pathways. But it is still a macroscopic
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

model relying on phenomenological models (such as heat capacities and diffusivities) for captur-
ing molecular processes. The power comes from the automation of routine calculations and the
ability to explore a large number of species that can be simultaneously present in a reaction system.
A more fundamental set of computational chemistry tools based on the DFT (4, 42) has
emerged. Since the mid-1980s, every decade has seen an explosion of studies on the kinetic
mechanism and other material properties that are revealed by DFT. These are truly high-fidelity
models that track the electron cloud interactions from a quantum mechanical perspective at a
scale where no encumbrances of fluid mechanics and other bulk transport phenomena can be felt
on the kinetic property predictions; hence, they yield true intrinsic kinetic information for use in
coarser-grain reactor models. Scholl & Steckel (4) provide an accessible introduction to DFT for
engineers with a compelling example of the success of DFT computations, revealing a 12-step
reaction sequence for the ammonia synthesis reaction, which was in stunning agreement with
the experimental measurements. Because chemical reactions involving bond making/breaking are
closely tied to electron cloud interaction, DFT is the right tool to provide insight as well as
quantitative data on such mechanisms.
Advances in in situ spectroscopy measurements (43–45) and quantum mechanical calculations
(21, 42) are enabling such detailed kinetic studies of simple reaction systems without the influence
of flow effects. If one looks at industrial-scale reactors, such as fluid catalytic crackers, the feed
is a complex mixture of hydrocarbons, and the reaction pathways to products are very complex
indeed. If one looks at the biological world, where complex, highly functional, self-organizing, and
interacting structures are synthesized spontaneously in a highly selective and often elegant manner
(such as beautiful colors and patterns in flowers, complex oils in seeds, or chemicals of medicinal
value in plants) under normal atmospheric conditions, one wonders about the sustainability of
the brute force methods that are used in chemical manufacturing, often at high temperatures and
pressures and with significant environmentally hazardous by-products. Clearly understanding and
regulating complex reaction pathways and nature-inspired reactor designs are the ways of future
sustainable reactor designs.
Seemingly simple cracking or bond-breaking reactions that take place in an important class of
reactors, e.g., fluid catalytic cracking processes (which is a workhorse of refineries to produce value-
added lower–molecular weight hydrocarbon fuels from higher–molecular weight oils), can involve
thousands of reactions steps. When such is the case, piecing together a simple reaction kinetics
model becomes a challenging task. The chemistry of the cracking process, for example, includes
an initiation (CH3 CH3 → 2CH∗3 ), hydrogen abstraction (CH∗3 + CH3 CH3 → CH4 + CH3 CH∗2 ),
radical decomposition (CH3 CH∗2 → CH2 = CH2 + H ∗ ), radical addition (CH3 CH∗2 + CH2 =
CH2 → CH3 CH2 CH2 CH∗2 ), and termination steps (CH∗3 + CH3 CH∗2 → CH3 CH2 CH3 ). Even if
one takes a simple molecule, such as (CH3 − CH2 − CH2 − CH3 ), unless there is some specific
signaling to break a chosen bond, the local energetics will determine which bonds are broken
under a certain set of conditions. For example, experimental data on product yield suggest that
48% of the time the CH3 − CH2 bond is broken, whereas 38% of the time, the CH2 − CH2
bond is broken, and 14% of the time, the terminal C−H bond may be broken. Whereas

www.annualreviews.org • Multiphase Reactors 353

 CH06CH16-Joshi ARI 15 July 2015 7:42

catalyzing certain pathways provides a certain measure of control over both selectivity and speed
of reaction, the mechanisms get more complicated, with surface adsorption and desorption
playing a crucial role in the overall dynamics. When the feed to the fluid catalytic crackers is gas
oil, which is a mixture of hydrocarbons, detailed kinetic models based on molecular processes
become impossible. Lumped kinetic models are then developed (10, 46, 47). Ancheyta-Juárez
et al. (48) discuss a hierarchically embedded parameter estimation technique for such lumped
models.
With the present state of challenges in computational chemistry, it is difficult to get these
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

kinetic parameters directly from numerical simulations of such complex reaction systems using
DFT. One path that might be feasible will be to study a comprehensive list of elementary chemical
reactions using DFT and then use an intermediate-fidelity modeling framework, such as that
offered by CHEMKIN, that integrates the elementary reaction pathways to construct an effective
mechanism (49, 50) and a rate expression for use in reactor design, which then is linked to a
reactor-scale coupled transport and reaction model to investigate the impact of advection and
diffusion (macro-, meso-, micro-, and molecular-scale mixing) on the reactor performance.
Another challenge in kinetic model building is in model discrimination. For example, in syn-
thesis gas kinetics, numerous overall kinetic models have been proposed, as documented in Van
der Laan & Beenackers (51), all purporting to represent the same overall reaction, perhaps in
different parts of the parameter space. Once again, DFT can potentially offer insights to resolve
such conflicts. Such rate expressions appear as source terms in the species transport equations to
be discussed in the section on the Generalized High-Fidelity Modeling Framework.

Classification of Multiphase Reactors

Reactors must often be designed to consider both chemical reactions and physical operations, such
as micromixing, macromixing, heat transfer, mass transfer, gas-liquid dispersing, and liquid-liquid
dispersing. For all these actions, the energy required is supplied in various forms (such as pressure,
kinetic, and potential energy), which become the basis for the classification of multiphase reactors.
Class 1 equipment, which uses pressure energy, is also called column-type equipment. Various
multiphase contactors fall under this category. Gas-liquid contactors include bubble columns,
packed bubble columns, internal-loop and external-loop air-lift reactors, sectionalized bubble
columns, plate columns, and others. Solid-fluid (liquid or gas) contactors include static mixers,
fixed beds, expanded beds, fluidized beds, and transport reactors or contactors. Liquid-liquid
contactors include spray columns, packed extraction columns, plate extraction columns, and static
mixers. Similar equipment is widely used in gas-liquid-solid, gas-liquid-liquid, and gas-liquid-
liquid-solid contactors.
In multiphase contactors, the dispersed phase is either bubbles, drops, particles, or combina-
tions thereof. The size and concentration variations (clustering) of these dispersed particles have a
major impact on the performance of this equipment. The continuous phase is either liquid or gas.
The other governing parameters include column diameter, column height, sparger or distributor
design, and the design of internals. Furthermore, in class 1 equipment, the energy is supplied
through the introduction of phases. For instance, in the case of bubble columns, the gas is sup-
plied against the static pressure of the multiphase dispersion. The gas leaves the column at HD and
possesses potential energy. For gas-liquid-solid multiphase dispersion, the overall energy balance
is given by

π 2
Ei − E o = D [(±V G ± V L ± V P )(G ρG +  L ρ L +  S ρ S ) − (∓V L ρ L ∓ VG ρG ∓ V P ρ P )]H D g, 5.
4

354 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

where, in the first parentheses, velocity VG is positive in an upward direction and VP is positive
in a downward direction if densities ρ P > ρ L . The upward liquid velocity is considered positive
when ρ L < (G ρG +  L ρ L +  S ρ S ) whereas downward liquid velocity is positive when ρG <
ρ L | ρ L > (G ρG +  L ρ L +  S ρ S ). The reverse signs hold in the second parentheses. Similar
equations can be written for L-L and G-S dispersions.
Out of the total energy (Ei − Eo ), part of the energy is dissipated at the interface of the dispersed
and continuous phase and is equivalent to the total drag force F multiplied by the slip velocity of
the dispersed particle:
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

π 2
E D = FD · V s = (FG − FB )VS = D  D C |ρ D − ρC |gVS , 6.
4
where the subscripts D and C represent the dispersed and continuous phase, respectively. The
sign of Vs is positive in the direction of terminal settling. The energy ED dissipates in the vicinity
of the particle (or bubble or drop). In the majority of cases, it is turbulent energy dissipation
where the primary scale of turbulence is of the order of particle diameter. The remaining energy
(Ei − Eo − E D ) dissipates in the bulk of the continuous phase, where the primary length scale is the
column diameter. The bulk turbulence may affect the particle motion [represented by values of
drag (CD ), lift (CL ), and virtual mass (CV ) coefficients]. In turn, these values decide ED and hence
(Ei − Eo − E D ). The phenomenon of interaction between the particle motion (and also the values
of heat and mass transfer coefficients) and the bulk turbulence are discussed later. Note that the
energy may be supplied by the dispersed particles (or bubbles or drops) and creates flows at the
equipment scale. It will be useful to understand the energy spectrum under these conditions.
Class 2–type equipment includes mainly stirred tanks, in which the impellers supply kinetic
energy to the multiphase systems for performing a variety of objectives, such as those mentioned
above. Stirred tanks offer unmatched flexibility and control over the transport processes in the
reactor. The performance of a stirred reactor can be optimized by appropriate adjustments of the
reactor hardware and the operating parameters (e.g., reactor and impeller shapes; number, type,
location, and size of impellers; degree of baffling; and impeller speed).
The impellers are classified as close clearance (e.g., anchors, paddles, gates, rakes, helical rib-
bon, and helical screw ribbon) and open (e.g., disc turbines, curved- and straight-blade turbines,
pitched-blade turbines, axial and radial propellers, and hydrofoils). The energy balance for stirred
tanks is explained in detail by Joshi et al. (7). The flow leaving the impeller possesses mean and
turbulent kinetic energies. Eventually, the mean kinetic energy is also converted into turbulence.
However, the performance of the impeller mainly depends upon the relative proportions of mean
and turbulent kinetic energy as it leaves the impeller. Generally, as the power number increases,
the impeller increasingly generates turbulence. The flow-controlled operations are blending, heat
transfer, and solid suspension, whereas the turbulence-controlled operations are gas-liquid and
liquid-liquid dispersions and homogenization. As in class 1–type reactors, the turbulence has two
scales: the dispersed phase size and the impeller/tank size. The interaction between these two
scales is discussed in the next section.
In class 3 equipment, the energy supplied is in the form of potential energy associated with the
liquid. For instance, in the conventional packed column (used for either distillation or absorption),
the liquid is pumped to the top of the column, and it attains potential energy. Then it is distributed
over the packing. If the gas is introduced at the bottom, the gas phase has the pressure energy;
however, it is usually negligible as compared with the potential energy of the liquid. Even if
the gas and liquid phases flow concurrently downward, the major contribution to the energy
is by the liquid phase. Because the liquid flows as films, this equipment (class 3) may also be
termed as film contactors. Other equipment in this category includes trickle-bed reactors and

www.annualreviews.org • Multiphase Reactors 355

 CH06CH16-Joshi ARI 15 July 2015 7:42

falling-film reactors/evaporators. In all the forms of class 3 reactors, the design of packings, liquid
distributors, and other internals is crucial to controlling the performance of this equipment.
Joshi & Doraiswamy (52) provide a table classifying the various multiphase reactors in the above-
mentioned three categories. Of course, some equipment uses two or all the forms of energy. These
may be classified depending upon which energy is dominant.

Fluid Mechanical Parameters and Their Relationship with Design Objectives

Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

The previous section described the modes of energy supply in different types of reactors, which
creates mean as well as turbulent motion in all phases. The mean flow is characterized by three
components of mean velocity and nine components of shear rates at all locations in the reactor.
Sometimes, the knowledge of mean kinetic energy and mean energy dissipation rates is also useful.
The turbulence is characterized in many ways. The first way is to know the field (the local
values in the reactor) of turbulent kinetic energy (k) and the turbulent energy dissipation rate.
The second way is to know the time series of three components of fluctuating velocity. The values
of second (standard deviation), third (skewness), and fourth (kurtosis) moments of the fluctuating
velocity are useful characteristic features. The third method of processing the fluctuating velocity
is to construct energy spectra in all three directions and hence the resultant energy spectrum.
The spectrum gives the wave number ranges of inertial and Kolmogorov regions. The spec-
tral information can be processed to get the length and velocity scales at three different levels:
(a) integral length and velocity, (b) Taylor length and velocity, and (c) Kolmogorov length and
velocity. The fourth method tries to characterize the phenomena behind the turbulent fluctuations
around the mean flow. These fluctuations are the result of the passage of the deterministic orga-
nized flow structures and the random disorganized irrational motions, which together constitute
the turbulent flows. The organized deterministic patterns are known as eddies or turbulent flow
structures. This subject has been reviewed by Joshi et al. (53) and Pope (54), among others.
The quantitative estimation of above-mentioned fluid mechanical parameters are achieved
through (a) experimental fluid dynamics (EFD), (b) CFD, and (c) mathematical tools. The EFD
tools measure 1D, 2D, and 3D velocity; phase fractions; and size distribution of bubbles, drops,
and particles, and excellent reviews are available (see, for instance, References 5 and 55). Re-
cent developments in CFD are discussed in the next section. The mathematical tools include fast
Fourier transforms, quadrant techniques, variable integral time average techniques, spectral anal-
ysis, discrete and continuous wavelet transforms, proper orthogonal decompositions, and hybrids
of these methods. Joshi et al. (53) have reviewed identification of flow structures from EFD, CFD,
and mathematical tools.
We also need to know the bubble (or particle or drop) size distributions and their 3D concen-
tration and velocity profiles. This also enables the calculation of the mean free path and turbulent
mean motion on the particle scale. For this purpose, the techniques of DPM and population balance
modeling (PBM) are useful (see section on the Generalized High-Fidelity Modeling Framework).
Knowledge of mean flow, the dynamics of turbulent structures, and the size and velocity distri-
bution of the dispersed phase is useful for understanding the mechanism of transport phenomena
on the equipment scale. With this knowledge, we can make reliable estimates of the extents of
mixing and axial mixing in all the phases. This knowledge is also useful for the calculation of heat
and mass transfer coefficients at the large fluid-solid interfaces, such as reactor walls, and a variety
of reactor internals, such as coils and baffles.
The real challenge has been in elucidating fluid mechanics in the vicinity of individual bubbles,
drops, and particles. This includes the mean and turbulence characterization in the variety of
interfaces and their neighborhood in the continuous phase and also inside the bubbles and drops.

356 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

This knowledge can give the values of drag, lift, and virtual mass coefficients at all the locations
in the reactor. Estimation of these fluid mechanical parameters is possible with the CFD tool of
DNS, which is described below.

GENERALIZED HIGH-FIDELITY MODELING FRAMEWORK THAT

WILL SHAPE FUTURE REACTOR DESIGNS
The previous section reviewed the fundamental mechanisms of reaction and transport and the
key macroscopic design parameters, such as rate constants, and flow parameters, such as residence
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

time distributions; micro- and macromixing parameters; drag, lift, and virtual mass forces; heat and
mass transfer coefficients; and turbulent flow parameters. Knowledge of many of these parameters
in dispersed phase reactors is often obtained on the macroscale from pilot plant input/output
data, and they are used in performance models of equipment (often in proprietary models for
proprietary reactor configurations) for both scale up and design or to predict the performance of
a reactor.
In this section, we review the framework of high-fidelity models that have been developed in
recent years for coupling reaction processes to transport processes that have come to be known as
CFD and DNS. These models do away with the need for concepts of micro- and macromixing,
residence time distributions, and effectiveness factors. They may still be used if necessary in
performance models, but such factors can be predicted by detailed flow analysis from CFD tools.
The general framework, their strengths, and the remaining challenges are discussed in this section.
A homogeneous reaction between two miscible fluids in the CFD framework can be viewed as an
advection mechanism that, along with the dispersion (turbulent and phase), contributes to macro-
or distributive mixing, and molecular diffusion contributes to micromixing, which is accompanied
by the intrinsic reaction mechanisms which completes the eventual conversion. Hence, coupling
the intrinsic reaction kinetics models (36, 56) to flow models is an essential and nearly complete
step in building models for multiphase reactor design and analysis.

Insights from Molecular Dynamics

MD (57) simulations examine interactions between molecules using classical mechanics formula-
tion of forces between molecules, governed by
d vi
mi = Fi , 7.
dt
where Fi is the force of interaction between the molecules, obtained from gradient of the potential
energy, Fi = ∂U ∂ri
. Whereas the classical MD retrieves the required force fields U from pragmatic
models built from experimental data (such as spectroscopic data), ab initio MD relies on DFT to
provide the U. Because at the scale of MD electron interactions are not resolved directly, but the
focus is on a molecule as an entity, physical processes of, e.g., phase change and adsorption are well
predicted by MD, including transport properties like viscosity and diffusivity, which can be used in
the continuum-scale models of fluid dynamics and mass transfer. Thus, the hierarchy of modeling
frameworks and the interconnections between the parameters should be appreciated over a range
of scales, from electron cloud interaction to the equipment-scale heat and mass transfer coefficients
and reaction rate constants, as illustrated in Figure 1. The finer-resolution simulation tools are
able to isolate the mechanisms much more effectively than any experiment can do. Although there
is a clear scale separation between DFT, MD, and continuum scales, and although parameters
from one scale can be transferred to a coarser-scale model, at the continuum scale phenomena
such as turbulence and multiphase dispersions occur over a broad range of interacting scales, and

www.annualreviews.org • Multiphase Reactors 357

 CH06CH16-Joshi ARI 15 July 2015 7:42

they must be studied by resolving all scales from the dispersed-phase scale to the equipment scale.
CFD tools provide the framework for such explorations, which are discussed in the next section.

High-Fidelity Continuum-Scale Models

The intrinsic kinetic models developed in earlier sections appear as source/sink terms in the species
conservation equation, which, in its simplest form for single-phase flows, appears as follows,
coupling the intrinsic kinetic models to transport mechanisms:
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

∂Ci
+ v · ∇Ci = ∇ · D∇Ci + ri (Ci , T ). 8.
∂t
Here, the velocity field comes from the solutions to the nonlinear fluid mechanical equations
[Navier-Stokes equations (NSE) for incompressible flows]:
 
∂v
ρ + v · ∇v = ∇ · μ∇v + ρ g − ∇ p 9.
∂t

∇ · v = 0. 10.
If the flow remains laminar in a reactor, solutions to the above equations are adequate to predict
the conversion in a reactor, by DNS. Although the term DNS has been used most often in the
turbulence flows, we use it more generically to indicate numerical solutions of a set of equations
without further filtering (spatial or temporal) to extract the high-fidelity solution directly. Such
solutions may be further analyzed to extract upscaling information for coarser-grained or perfor-
mance models that are used in the design of such reactors. The advection term on the left-hand
side of Equation 9 governs the macro- or distributive mixing in specific reactor configurations, the
diffusive term on the right-hand side handles the micromixing, and the actual chemical conversion
is tracked by the last term on the right-hand side. In an idealized CSTR, for example, the lumped
parameter performance model is written by a species mass balance as
d Ci
V = F (Cin − Ci ) − ri , 11.
dt
where V is the volume of the reactor, F is the flow rate into and out of the reactor, Cin is the inlet
concentration of the reactant, Ci is the exit concentration, and ri is the rate of consumption of the
reactant by the reaction mechanism discussed earlier. Often the rate expression found in Equation 1
is used in Equation 11 under the assumption of a well-mixed state. But solving Equations 8–10
together with Equation 1 makes no such assumptions and permits the impact of incomplete mixing
on effective conversion to be modeled fully. This process is much more challenging for turbulent,
multiphase flows, as we discuss in the next sections, but is entirely under reach for single-phase
laminar flow reactors, for example in microreactor designs.
If the flow effects are not at all present, such as in the problems of diffusion-reaction in a
catalyst pellet, then Equation 8 without the advection term provides a framework for analysis;
entire monographs have been devoted to such studies of reaction-diffusion problems (58, 59). The
nonlinearity in the intrinsic reaction model still poses challenges, but in most cases it is amenable
to direct numerical solutions without further simplifying assumptions. Analytical solutions often
provide insight into the impact of diffusion limitation on the effective reaction rate, and the
introduction of the effectiveness factor concept captures such results for practical applications.
If the flow is present, but remains laminar, with mass transfer across the interface and is
accompanied by reaction in the bulk phase, the insights gained by solution of such problems are
useful in understanding the gas-liquid reactor performance. When properly coupled across the

358 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

interface and fully resolved, the basic equations provide the DNS framework for understanding
such mechanisms. Books by Astarita (60) and Danckwerts (31) deal with early developments in
this field for simple systems and offer insightful interpretations of the competition between mass
transfer and reaction kinetics.
However, if the flow is strong and turbulent, then the velocity field is difficult to predict by DNS
of the NSE in actual reactor configurations owing to the highly nonlinear advection term that
introduces turbulent structures over a wide range of scales in such equipment; however, several
recent attempts are being made at such calculations. Interestingly, in 1999 Spalart (61) estimated
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

that DNS of massively separated single-phase turbulent flows will require approximately 1016 grid
points and 107.7 time steps. Further, the hardware and software for undertaking such calculations
may be ready by 2080! In a review article on turbulent mixing, Dimotakis (62) estimates the
computational load to scale as Re 3 Sc 2 , which suggests that doubling Re increases the computational
load by eight times, whereas capturing the liquid-phase versus gas-phase mixing increases it by
roughly six orders of magnitude. Even in the case of single-phase miscible systems, DNS of
turbulence in realistic flow configurations remains a challenge, although as a computational probe
in small regions it can provide useful insights into the closure relations needed in the time-
averaged model equation for turbulence. Hence, the above Equations 8–10 are time averaged to
get dynamical equations that filter high-frequency fluctuation in velocity and concentration fields
but provide equations for the time-averaged velocity and concentration fields. The time-averaged
equations take the form of
 
∂ v̄
ρ + v̄ · ∇ v̄ = ∇ · μ∇ v̄ − ρv  v  + ρ g − ∇ p̄, 12.
∂t

∇ · v̄ = 0, 13.
and
∂ C̄i
+ v̄ · ∇ C̄i = ∇ · D∇ C̄i − v  Ci + ri (Ci , T ). 14.
∂t
The velocity-velocity correlations (−ρv  v  ) in Equation 12 and velocity-concentration correla-
tions in Equation 14 must be modeled by closure relations. In the turbulence field, several models
have been proposed, and detailed expositions on these can be found in References 54, 63, and
64. In Equation 14, the velocity-concentration correlations are modeled by an effective turbulent
mass dispersion model to capture the enhanced macromixing on the small eddy scale. It must also
be cautioned that the time-averaging operation on the intrinsic reaction kinetic model can be a
challenging task, except for a linear first-order kinetics, which is seldom the case. Haworth (65)
reviews the advances in turbulent-combustion process modeling, which remains one of the chal-
lenging tasks because of the complex turbulence-chemistry interactions and turbulence-radiation
interactions over a wide dynamic range of temporal and spatial scales, with nonlinearities arising
from combustion kinetics, radiation heat transfer, and advective transport processes. The complete
governing equations for the compressible, nonisothermal, reacting flow are given in Reference
65. These equations show the full range of coupled mechanisms and their interactions that must
be considered in a rigorous reactor design, spanning all scales (54, 66, 67).

Computational Fluid Dynamics: Single-Phase Laminar

and Turbulent Flow Models
The coupled momentum, heat, and species transport and reaction in a single-phase flow of two mis-
cible reactant species are governed by Equations 8–10. If the flow remains laminar, the parameters

www.annualreviews.org • Multiphase Reactors 359

 CH06CH16-Joshi ARI 15 July 2015 7:42

appearing in those equations can be obtained from laboratory experiments under ideal conditions
(such as kinematic, thermal and mass diffusivities, activation energies, and rate constants—these
are the closure parameters in models that enable moving from what are essentially molecular
processes to the continuum level), and we have a complete model that is valid on any scale and
reactor configuration. Although they are a complex set of nonlinear, partial differential equations,
their engineering solution by computational means is quite readily attained at present, thanks to
advances in computational hardware and algorithms (OpenFOAM R
). However, if the flow con-
ditions are pushed into the high–flow rate inertial regime, while the equations remain valid on
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

sound conservation principles through a series of flow instabilities and bifurcations, turbulence
ensues, which is inherently time dependent with flow structures that span a whole range of spatial
and temporal scales. Although it is feasible, in principle, to resolve such features by brute force
computational power (which is what DNS of turbulent and turbulent-reactive flows is about),
its use should be limited to the extent that we need high-fidelity information for the turbulence
closure models. This approach is discussed rigorously in References 54 and 63 and from a prac-
tical point of view in Reference 64. Several practical turbulence models have emerged for use in
single-phase flows, the most popular among them being the k − ε and k − ω models, the Reynolds
stress model (RSM), and the large eddy simulation (LES) model. Most commercial simulators
offer these as standard packages in their model suites. But their extension to multiphase flows is
not quite straightforward, although such use in commercial simulators is enabled.

Computational Fluid Dynamics: Two-Fluid Models

For Coarse-Grained Flow Structures
The concept of interpenetrating continua is used to propose two-fluid Eulerian-Eulerian models
that smear the information about the interaction between a continuous fluid and a dispersed
phase, whether the dispersed phase is a fluid or a solid. These models are quite popular extensions
of single-phase flow models but are not founded on rigorous, consistent derivation from the
underlying interactions of the dispersed phase with the continuous phase, particularly in the
turbulent domain; hence, there is a great deal of uncertainty in the closure models needed to capture
such lost information. Although ad hoc in nature, they are likely to remain popular for industrial-
scale problems, because of the computational expediency and the perhaps poorly conceived notion
that the necessary closure models can be tuned from pilot-scale experiments to study performance
on a different scale with the same equipment. Although it is a pragmatic approach for short-
term use, there is still a need to apply the same rigor to multiphase flows that the turbulence
community has applied to single-phase flows. An excellent example of such an approach is that
of Fox (68). Sustained research over a period of time on selected DNS simulations and high-
fidelity experiments may offer insight on the appropriate closure models needed for the rigorously
developed averaged multiphase, turbulent models, such as in the work of Fox (68). This has to
be done for each class of multiphase flows, such as gas-solid, gas-liquid, and solid-liquid systems.
Whitaker (69, 70) and Slattery (71) developed this approach initially for porous media flows in
the low–Reynolds number limit to show that Darcy’s empirical model for such flows is indeed
recoverable and consistent with NSE in the creeping flow limit, and Anderson & Jackson (14)
developed similar equations for fluidized beds. Since then, many others have studied similar spatial
and ensemble averaging procedures (68, 72). In such procedures, the identity and location of each
dispersed phase is lost, but velocities and volume fractions of each phase are treated as a continuum,
and the transport of momentum, heat, and mass between the phases are modeled by closure
relations. This is where the earlier discussions of drag, lift, virtual mass, heat, and mass transfer
coefficients at the interface are useful to provide approximate closure relations. When models

360 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

Continuity
Δ Δ
∂ϕi ρi + • (ϕi ρiUi –Γi ϕi ) = 0 i = L, G
∂t

Momentum
∂ϕi ρiUi +
Δ
[( Δ
• ϕi (ρi Ui Ui )–μi( Ui + UTi )
Δ
)] = ϕ (B –
i i
Δ
p) + Mi i = L, G
∂t

Species
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

[
∂ϕi ρiYi j + • ϕi (ρi UiYi j –Γij Yi j ) =
Δ Δ
]
∂t

Σ
k = 1,k≠j
m jik i = L, G j = 1 ...N

Figure 2
The two fluid, Eulerian-Eulerian model found in commercial computational fluid dynamics (CFD)
simulators.

of such resolution are used, they can identify heterogeneous regions within a process vessel, if
the closure models turn out to be accurate representations of the interphase transfer processes.
This is one step better than the perfectly homogeneous flow conditions normally assumed in
performance models such as CSTR and PFR. When the flow becomes turbulent, the challenges
increase manyfold as the fine structures of the turbulence must be smoothed out by time averaging,
and developing closure models for such lost information that are rigorous and scale invariant has
remained a challenge. Additionally, when dispersed phases can interact with turbulent flows,
sometimes contributing to turbulent energy production and dissipation, the extension of the two-
fluid Eulerian framework to multiphase, turbulent flows is not at all on a sound theoretical basis.
However, numerous studies are routinely using those model combinations to study real chemical
process systems.
The conservation equations in the Eulerian-Eulerian framework are summarized in Figure 2.
Although they have the ability to capture gross-scale heterogeneities, fine structures are difficult
to predict, particularly for dense solid-fluid systems. Kinetic theory–based granular temperature
approaches have been developed to address that challenge to some extent (73–75).

Computational Fluid Dynamics–Discrete Particle Model:

For Flow Structure–Solids Interactions
Whereas the two-fluid model (TFM) provides a coarse description of interacting phases and relies
extensively on the highly empirical closure models, the DPM offers a higher resolution by tracking
both fluid-particle (F-P) and particle-particle (P-P) interactions suitable for fluid-solid systems. In
the inertial regime, the F-P interactions become nonlinear and important, whereas in a dense flow
regime, the P-P interactions play a more dominant role. In fluid-fluid systems, coalescence and
breakup processes determine the large-scale flow structures, whereas in F-P systems, collisional
interactions can dominate the flow structures, introducing greater diversity of the flow behaviors in
fluid-particle systems. In addition, void fraction fluctuations introduce additional strong nonlinear
effects to the F-P interaction (76) by introducing variations in effective suspension densities of
clusters. Questions of whether the simple drag laws based on local homogeneous/isotropic flows
are adequate to capture the cascading scales of structures that form in particulate systems remain
to be answered.

www.annualreviews.org • Multiphase Reactors 361

 CH06CH16-Joshi ARI 15 July 2015 7:42

Dense particulate flows occur in a wide variety of industrial processes, such as coating, dry-
ing, granulation, gasification, and catalytic cracking, to name only a few. The efficiency of these
processes is profoundly affected by the fluid-particle dynamics, which is fundamentally prescribed
by these two interactions. Indeed, improving the particle-laden fluid flow patterns may signifi-
cantly enhance mixing, heat and mass transfer, reaction rates, surface or interface renewal, and
phase contacting or segregation. Although much of our earlier understanding of these complex
fluid-particle systems has involved experimental studies, CFD has increasingly become the tool
of choice for many of these systems because of sustained advances in the model fidelity. The
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

computational hardware and algorithms provide insight into the fundamental processes that are
much more difficult to obtain without great effort in carefully conducted experiments. The cou-
pling and interaction of various mechanisms are readily incorporated in a simulation tool. The
macroscopic effects can then be compared with pilot-scale experiments, thus gaining confidence
in the simulation tools. Significant research efforts have focused on the development of numerical
models to unearth complex flow behavior and quantify its impact on the desired chemical process
performance at all scales.
In the DPM (77, 78) approach, the individual particle motion is described in a Lagrangian
framework by modeling the interparticle interactions through either the hard-sphere or the soft-
sphere dynamics. The hard-sphere model starts from the integral form of the particle motion
equations. Impulsive and instantaneous collisions between particles are assumed. Restitution and
friction coefficients are rigorously introduced to account for the energy dissipation owing to
particle-particle and particle-wall interactions. Unlike in the hard-sphere model, contacts among
particles in the soft-sphere model are tracked. The contact forces, namely, normal, damping, and
frictional forces, are accounted for by their equivalent simple mechanical elements, such as springs,
dashpots, and sliders, respectively. The soft-sphere model is typically a time-driven approach that
requires a small simulation time step to yield the contact displacements and forces. Generally
speaking, this feature makes it less efficient than the hard-sphere model (13, 78). Nevertheless,
compared with the implementation of the hard-sphere model, where complicated collision han-
dling algorithms must be well designed to achieve efficiency coding, the soft-sphere model is easier
and more straightforward. Moreover, this approach is capable of a broad range of flow regimes,
including jammed or static granular systems, that the hard-sphere model usually fails to simulate
owing to its inability to capture elastic and multiparticle contacts. The fluid phase is treated as
a continuum in the averaged sense, and the summary of equations is shown in Figure 3. The
summary of equations for the hard-sphere and soft-sphere approaches is given in Figures 4 and 5,
respectively.

Computational Fluid Dynamics–Population Balance Models: For Flow

Structure–Bubble Interactions
Whenever characteristics of the dispersed phase, such as its size, can change dynamically owing to
its interaction with the continuous phase in a flowing environment, it becomes necessary to capture
and model the impact of such changes on the dynamics of the multiphase dispersions. For example,
a change in size can change the drag and buoyancy forces, thus affecting the velocity of a collective
gas plume or solid dispersions. PBM coupled with the CFD has provided a framework for tracking
such changes in dispersed phase characteristics. The applications for CFD-PBM framework are
quite varied indeed, from bubble breakup and coalescence in bubble column reactors to reactive
crystallization. A comprehensive book by Ramkrishna (79) and a more recent review article by
Ramkrishna & Singh (80) provide excellent expositions of the PBM approach. Another review
by Sporleder et al. (81) also contributes to a clear exposition of the basic ideas. Articles using

362 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

Continuity equation
∂ Δ
(ερg ) + • (ερg ug ) = 0
∂t

Momentum equation

∂
• (ερg ug ug ) = – ε p + • (ετg ) + Sg + ερg g
Δ Δ Δ
(ερg ug ) +
∂t

Momentum exchange source terms

Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

NP V β
Sg = – 1
V ∫VkΣ= 1 p,k
1–ε
(ug–up,k)δ(x–xp,k)dV

Figure 3
Discrete particle modeling (DPM) framework for continuous phase.

PBM in specific applications are too numerous, and the following are illustrative of the range of
applications in multiphase flow equipment (82–85).
The PBM uses a number density function f defined in a phase space composed of spatial
coordinates r = (x1 , x2 , x3 ) and one or more properties of the dispersed phase, such as size, mass,
and temperature. The latter are referred to commonly as internal coordinates, ζ = (ξ1 , ξ2 . . . ξn ).
Then f (r, ζ , t)drd ζ is the ensemble of entities having coordinates in the range of (d r, d ζ ) around
(r, ζ ) at a given time t. In general, the Population Balance Model (PBM) is an integro-partial
differential equation with an accumulation term, convection terms for each dimension of the
phase space, and source terms that account for the death and birth of the entities. The origin of
the PBM is attributed to Smoluchowski (86), who focused on studies of coagulation of colloidal
particles. Following Sporleder et al. (81), the conservation statement reads

Accumulation = input − output,

Relative velocity
v12 = G12 –(r1ω + r2ω) × n, G12 = v1–v2

Normal and tangential unit vectors J

x1–x2 v012–(G012 • n)n ω1

n= t= r1, m1 n
|x1–x2| |v012–(G012 • n)n| r2, m2 ω2

t
Collision impulse equation
G012
m1(v1–v01) = J v012 • n
Jn = m1m2(1 + en)
(m1 + m2)
m2(v2–v02) = –J
2m1m2(1 + et)v012 • t
I1 –μf Jn (Sliding) μf <
(ω –ω0) = J × n en μf et 7(m1 + m2)Jn
r1 1 1
I2 Jt =
(ω –ω0) = J × n 2m1m2v012•t 2m1m2(1 + et)v012 • t
r2 2 2 μf ≥
–(1 + et) (Sticking)
7(m1 + m2) 7(m1 + m2)Jn

Figure 4
Discrete particle modeling (DPM) framework for particulate phase using hard-sphere models.

www.annualreviews.org • Multiphase Reactors 363

 CH06CH16-Joshi ARI 15 July 2015 7:42

Spring-dashpot system

Contact force and torque i

Fci = Σ
j≠i
(Fcn, ij + Fct, ij )

Dashpoint Normal
Spring
Σ
dωi
Tci = (Rinij × Fct, ij ) Ii = Ti
dt
j≠i

Hertzian nonlinear model (viscoelastic)

j
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

Fcn, ij = (–knδn – ηn, vij • nij)nij

–ktδt – ηn, vt, ij , if sticking, |Fct, ij|<μf |Fcn, ij|

Fct, ij =
–μf |Fcn, ij|δt /|δt |, if sliding, reset δt = –Fct, ij /kt

i Tangent
δn = Ri + Rj – |xj–xi|
Spring Slider
δt = –nij × (nij × δt–Δt) + vt, ijdt
j
vt, ij = vij –(vij • nij)nij + (Riωi + Rjωj) × nij

Dashpoint

Es, σs, e, μf
Es 2R 2 2RGs 2|1ne| mkn
kn = δ½n kt = δ½n ηn = ηt =
3(1–σ2s ) 2–σ2s π2–1n2e

Figure 5
Discrete particle modeling (DPM) framework for particulate phase using a soft-sphere approach.

written symbolically over all entities as

  
D
f |dR| = J B − J D |dR| = J|dR|,
Dt R R R

where J(r, ζ , t)drdζ represents the net rate of production of entities with coordinates between (r, ζ )
and (r + d r, ζ + d ζ ) consisting of both birth events, J B , and death events, J D . Using Transport’s
and Gauss’s theorems, it can be written as
∂ f (r, ζ , t)
+ ∇r · (v r (r, ζ , t) f (r, ζ , t) + ∇ζ · (v ζ (r, ζ , t) f (r, ζ , t) = J(r, ζ , t).
∂t
The above PBM model framework can be found implemented in commercial CFD codes,
such as Fluent. Further theoretical developments have extended the above equations to include
diffusion terms such as d = Dr ∇r f + Dζ ∇ξ f , where Dr and Dξ are the diffusivities in the physical
and property space, respectively. These terms arise to account for random fluctuations in the actual
flow and growth velocities. The PBM framework has general applicability to study the evolution
of any set of a population in the biological or the physical world, as discussed by Ramkrishna (79).
In the case of bubble column reactors, for example, specific kernels must be developed for the
coalescence and breakup process of bubbles that are governed by fluid and interfacial dynamical
processes. If size is used as one of the internal coordinates, then a breakup of a bubble gives
rise to the death of one size and the birth of two other sizes. A coalescence event has similar
consequences on both J B and J D . A large literature has developed on such models based on
experimental observations. Reviews are beginning to appear (see, for example, Reference 87), and
several alternate model options are being built into the simulation codes.

364 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

Solid
Fluid
dUi
ρL du = ρLg + • σ Mi
Δ = Mig + Fi
dt in Ω\P(t) dt
BC: u = uΓ(t) on Γ in P(t)
Δ
•u=0 dωi
Ii + ωi × Iiωi = Ti
dt
Γ
IC: u|t = 0 = u0 in Ω\P(0)
dXi
= Ui
dt

Ω dθi
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

= ωi
Coupling ∂Pi(t) dt
u = Ui + ωi × ri on ∂Pi(t), i = 1..N
IC: Xi|t = 0 = Xi,o; θi|t = 0 = θi,o
Pi(t)
∫ ∂P (t) σ • ndS;
Fi =
i

T =∫ r × σ • ndS
i ∂P (t)i
i

Figure 6
Modeling framework for direct numerical simulation (DNS) of fluid-particle systems.

Computatational Fluid Dynamics–Direct Numerical Simulation:

The Ultimate High-Fidelity Modeling Tool
CFD–DNS hydrodynamics of fluid-particle systems. DNS of the coupled model equations
based on conservation laws by fully resolving the interactions between the continuous phase and
the dispersed phase is emerging as an important probe to understand the multiphase flow physics
and possibly extract closure models needed in the coarser-grained TFM and DPM discussed
previously. A recent review article by Tenneti & Subramaniam (88) presents an excellent overview
of the challenges in conducting such simulations and in extracting closure information needed
for the coarse-grained models from particle resolved DNS (PR-DNS) of multiphase gas-solid
flows. Emergence of several computational algorithms, such as the fictitious domain method (89,
90), immersed boundary method (91), lattice Boltzmann method (92, 93), and advanced parallel
computational architecture with GPUs, and the algorithms that exploit them are enabling DNS
probes to be used effectively to understand the transport mechanisms in both fluid-solid systems
and fluid-fluid systems with deformable interfaces. The governing equations and their couplings
for the fluid-solid systems are shown in Figure 6, whereas those for the fluid-fluid systems are
shown in Figure 7. The present state of the algorithms permits PR-DNS simulations to be carried
out for perhaps up to a particle Reynolds number of 500–800, and dispersed phase numbers
on the order of several thousands. Even the single-particle dynamics gets very complicated and
time dependent, with symmetry breaking and vortex shedding occurring at Re of above 220—see
Natarajan & Acrivos (94). The interactions in the turbulent regimes will be extremely difficult
to resolve from the PR-DNS approach at very high Reynolds numbers. The turbulent-particle
interaction in dense dispersed phases will remain a challenge for the foreseeable future. A recent
study by Fox (68) presents a rigorous analysis of developing mesoscale models for gas-solid systems.
Kuipers and his group (91, 95–100) have explored DNS as a tool in studying many chemical
process systems, whereas reactive systems using DNS have been focused primarily on turbulent
combustion (101–105).

CFD–DNS: particle-turbulent interactions/modulations. As one of the most difficult re-

search problems, particle-turbulent interactions have interested researchers for more than 50 years.

www.annualreviews.org • Multiphase Reactors 365

 CH06CH16-Joshi ARI 15 July 2015 7:42

Fluid 1 σi = piδ + 2/Rei D[ui] Fluid 2

Du1 Δ Δ
D[ui] = 0.5[ ui + ( ui)T] Du2
= f1 + • σ1
Δ
λ = f2 + • σ2
Δ
Dt in Ω1 Dt in Ω2
BC: u1 = u1,Γ(t) on Γ
•u=0
Δ Δ
• u2 = 0

Ω2 Ω2 URρ1
Coupling
Re1 =
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

Ω1 μ1
(σ1– σ2) • n = 1 κn, u1 = u2 on Σ
We URρ1
Ω2 Re2 = = ηRe1
μ2
∂Σ Δ Ω2
+ (u • )Σ = 0
∂1 We = URρ1/σ

λ = ρ1/ρ2

Figure 7
Modeling framework for direct numerical simulation (DNS) of fluid-fluid systems.

In his classical review paper, Hetsroni (106) concluded that the presence of particles with a low
particle Reynolds number tends to suppress the turbulence of the carrier fluid. Particles with high
particle Reynolds number (>400) tend to enhance turbulence, most likely owing to vortex shed-
ding. Joshi (107) and Pandit & Joshi (108) have shown that, in a C D − Re plot, the range of creeping
flow increases to higher Re with an increase in s . Further, the fully turbulent regime advances to
lower Re with an increase in s . Authors have also shown that, similar to the effect of particles on
turbulence modulation, the turbulence also modifies the settling velocity from the terminal value.
In some cases, it results in hindered settling, and in other cases enhancement has been observed,
particularly when the flow behavior becomes heterogeneous. These are early macroscopic obser-
vations on the phenomenology of fluid-particle interactions in the turbulent regime. Since the
late 1990s, the DNS probe is being used increasingly to understand the mechanisms of particle-
turbulence interaction (109–112). However, the quantitative understanding is still not clear. This
is because different researchers have focused on different parameter spaces in their studies, such
as use of (a) experiments with varying quality (isotropy, homogeneity, intensity) of carrier fluid
flows, (b) varying physical properties (including size and shape) of particles and fluids, (c) ways of
particle-fluid coupling (varies from one way to six ways), and (d ) lack of quantification of errors in
DNS numerical solvers (see section on CFD–DNS Hydrodynamics of Fluid-Particle Systems). In
fact, comprehensive and well-planned research must be undertaken to understand the sensitivity
of the four above-mentioned parameters on the particle turbulence interaction.
In addition to the modulation of turbulence, the effect of turbulence on particle aggregation has
been analyzed (113, 114). Accordingly, the core-annulus phenomenon in fluid catalytic crackers has
been explained. Researchers (115–117) have made useful contributions to the prediction of onset of
transition to heterogeneity from homogeneity. The subject of the transition from homogeneous to
heterogeneous dispersions has been addressed by Jackson and coworkers (14, 118, 119), Batchelor
(120), Didwania & Homsy (121), Joshi et al. (122), Thorat et al. (123), Thorat & Joshi (124),
Sundaresan (125), and Ghatage et al. (126), among others. All these cases begin from some kind of
disturbance, which either decays to the homogeneous regime or propagates to the heterogeneous
regime. Further, such a transition has been shown to follow enhancement in particle setting
velocity owing to possible particle aggregation/generation of strong convection currents in the
continuous phase. However, like turbulence modulation, the phenomena of aggregation as well

366 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

as the onset of transition have not been quantitatively understood, and substantial additional work
is needed to analyze the parametric sensitivity of the four above-mentioned parameters.
Practically all the research work has been devoted to the interaction of turbulence with solid
particles where the interface is rigid. Future work must include the analysis of interactions with
bubbles and drops where the interface can be rigid, partially mobile, or completely mobile. Further,
bubbles and drops may change shapes and sizes through breakage and coalescence in the multiphase
reactor environment.
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

CFD–DNS: drag, lift, and virtual mass. The knowledge of drag, lift, and virtual mass forces is
needed in the framework of TFM for understanding the flow fields of continuous and dispersed
phases from a coarse-grained model. A large number of empirical/semiempirical correlations have
been reported in the past 70 years; however, the estimated values vary over a wide range. During
the past 20 years, DNS has been employed increasingly for quantitative determination of these
forces. However, the quantitative understanding is not complete, particularly in the high–Reynolds
number range. This is mainly because of the four points raised in the section on Particle-Turbulent
Interactions/Modulations. In addition to these, the following points need attention:
1. Though substantial DNS simulations have been performed for drag, lift has received scant
attention, and virtual mass has remained practically unattended.
2. Substantial DNS simulations have been performed for solid particles, whereas bubbles and
drops have received scant attention.
3. Many studies have examined the DNS simulation of low Reynolds number flows around a
single particle and a collection of particles of varying concentration, but higher Re results
from DNS remain a challenge.
Additional problems or issues to be addressed include
4. Substantial DNS simulation of flow around a particle, bubble, and drop in the Re ∞ range of
1–103 to obtain 3D instantaneous velocities at temporal resolutions up to the Kolmogorov
scale. The impact of boundary layer separation and also the size, shape, and stability of wakes
behind a particle on variability in drag and lift must be explored over a range of practical
Re ∞ .
5. DNS simulation of flow around a single particle in the Re ∞ range of 103 –106 , capturing
the sudden drop in drag-coefficient, CD , at Re ∞ of approximately 2 × 105 , owing to the
transition from laminar to turbulent boundary layer separation.
6. DNS simulation of flow around a pair of particles and a large but finite number of neigh-
boring particles.
7. DNS simulation of setting/rising of a particle/bubble in small diameter tubes and to under-
stand the wall effect.
8. DNS simulation of gas-solid, gas-liquid, and solid-liquid multiphase reactors in a homoge-
neous and heterogeneous regime in the Re ∞ range of 1–106 , to obtain 3D component instan-
taneous velocities at temporal resolutions up to a small scale permissible by the computational
power. Estimation of skin, form, and total friction at all locations on particle/bubbles/drops
within the multiphase reactor. Simulation of boundary layer separation in multiparticle sys-
tems and also understanding of size, shape, and stability of wakes with respect to Re and
dispersed-phase volume fraction.
9. Quantification of homogeneity, isotropy, and energy spectrum in multiphase (solid-liquid,
gas-liquid, liquid-liquid) dispersions as a function of Re ∞ ,ε L , and diameter and shape of
dispersed particles/bubbles/drops.
10. Development of scaling laws for the inertial, dissipation, and large-scale subrange in solid-
liquid, gas-liquid, and solid-gas dispersions.

www.annualreviews.org • Multiphase Reactors 367

 CH06CH16-Joshi ARI 15 July 2015 7:42

11. DNS simulation for estimation of lift, virtual mass, and Basset forces for a single and multi-
particle systems (and bubbles and drops) over a wide range of Re and flow fields in multiphase
reactors.
12. Measurement of instantaneous velocity within the multiphase reactors using techniques
such as high-speed particle image velocimetry (PIV) at sampling rates of up to 10 kHz,
in combination with refractive index matching of the liquid and solid phases, along with
application of mathematical techniques, such as discrete and continuous wavelet transforms
and proper orthogonal decomposition, to identify and characterize the flow structures in
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

terms of shape, size, velocity, and energy distributions.

13. Understanding of the mechanism of vorticity generation and hence the origin of turbulence
in the presence of a second phase.

CFD–DNS heat and mass transport in fluid-particle systems. Like the concept of drag
coefficients, a large number of empirical equations are reported in the published literature for
k, h, and a for a variety of multiphase systems and reactors. The limitations of empiricism were
discussed in the section on Evolving Trends in Process Design. To increase the reliability of
estimations, DNS of heat and mass transport processes can come to the rescue. Recently, Deen
et al. (127) wrote an excellent review of DNS of fluid-particle mass, momentum, and heat transfer
in dense gas-solid flows. The review lucidly describes the current status and gives suggestions for
future work.
The governing equations for mass and heat transfer in the DPM framework are as follows
(127):
d C Ap DA
Vp = SP Sh (C AF − C AP ) 15.
dt dp
and
d T Ap kF
mp Cpp = SP Nu (T F − T P ). 16.
dt dp
Note that in the DPM framework, heat and mass transfer coefficients are needed as inputs to
the model, whereas in the PR-DNS framework, only the flux of heat and mass transfer is needed on
the right-hand side of the above equations (see References 127 and 128 for details), which can be
computed from the fully resolved solution. Still, the above equations assume that the particle has a
lumped parameter representation with a uniform temperature inside. If the thermal conductivity is
small (Bi number is large), the conduction equation inside the solid must also be solved. Upscaling
of heat-transfer information from high-fidelity models to coarse-grained models is addressed in
Reference 129. In addition, future work must include the rate of four points given in the section
on Particle-Turbulent Interactions/Modulations and thirteen points in the section on Drag, Lift,
and Virtual Mass on heat and mass transfer.

Application of Direct Numerical Simulations in Determining

Low-Order Models
With the present status of computational power, it is impossible to undertake full-scale DNS
simulations of commercial-size reactors. Therefore, low-order models are needed. A rigorous
development of turbulent, multiphase flow field equations from basic continuum-level descriptions
using spatial/time/ensemble averaging procedures to identify the closure models that are needed
is an ongoing research effort (68). In the meantime, pragmatic/ad hoc extension of single-phase
turbulent models to multiphase flow situations is being attempted owing to widespread availability
and use of commercial CFD simulation software. For single-phase turbulence, several models exist,

368 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

such as k − ε and k − ω models, RSM, and LES models. These models are listed in increasing
order of accuracy and in their need for computational time. At the present time, the k − ε model
has indeed been a workhorse in industrial practice. The principal limitations of the k − ε model
are the assumptions of isotropy and the methodology by which the modeling of terms in the k and
ε is executed. First of all, Reynolds averaging of the equations of continuity and motion results in
the Reynolds stress term (ui u j ) and is modeled as

∂ui
(ui u j ) = νt , 17.
∂xj
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

where νt is the eddy diffusivity and is modeled as

k2
νt = Cμ . 18.

To get k and ε in Equation 18, the governing equations for k and ε must be solved. The equa-
tions for both k and ε contain six terms: (a) accumulation, (b) convective transport, (c) diffusive
(molecular) transport, (d ) turbulent transport, (e) production, and ( f ) dissipation (see Supple-
mental Text for details; follow the Supplemental Material link from the Annual Reviews home Supplemental Material
page at http://www.annualreviews.org). The two-fluid versions of these equations have been
studied using the Reynolds averaging; these versions involve double and triple correlations of
velocity and hold-up terms. The terms needing closure contain complex correlations, and their
modeled forms include constants such as Cμ , Cε1 , Cε2 , σk , and σε for the k − ε model. A limita-
tion of the k − ε model is the assumption of isotropy and model flow cases (such as a decaying
turbulence region behind a grid and a turbulent boundary layer) for which the model constants
(turbulence parameters) are estimated. To reduce the severity of the assumptions, the k − ε model
parameters can be estimated using DNS simulations, which gives temporal and special variation
of all the fluctuating velocity components over a wide range of wave numbers. This information
can be employed for the estimation of correlations cited above and for more reliable estimates of
turbulence parameters in k − ε and RSM models. The DNS simulation data can also be used for
the subgrid-scale modeling needed in LES.
However, the selection and estimation of model parameters (as described earlier) has been far
from satisfactory based on the knowledge of fluid mechanics, including the physics of turbulence.
The use of DNS for the estimation of drag, lift, and virtual mass coefficients was described
in the section on Drag, Lift, and Virtual Mass, and its use for the estimation of mass (k) heat
transfer (h) coefficients was described in the section on Heat and Mass Transport in Fluid-Particle
Systems. Figure 8 shows the interrelationships between the various parameters in such turbulent-
multiphase flow models at various degrees of resolution.
Because DNS needs large computational resources, these simulations may be performed for a
few typical representative regions in the equipment. For instance, here we discuss one speculative
procedure. Consider a stirred tank reactor of 25,000 L with an average power consumption (P/V)
of 2 kW/m3 and an average gas hold-up (∈G ) of 25%. The local values of P/V and ∈G are markedly
different and may lie in the range of 0.1 to 40 kW/m3 and 0.05 to 0.5, respectively. We may select
approximately 10 locations for DNS simulations so that 10 representative values of P/V (e.g., 0.1,
0.25, 0.5, 1, 2, . . . , 40) are selected (fortunately, the order of P/V and ∈G is practically the same).
Then we undertake only 10 DNS simulations over the above-mentioned values of P/V; however,
the volume of simulation at each location may be 1 L. This methodology considerably reduces the
need for computational power but, more importantly, gives reliable estimates of parameter values
in the k − ε model and RSM over a range of operating conditions. It also gives accurate values
of drag, lift, virtual mass, and heat and mass transfer coefficients and also the kernels needed in

www.annualreviews.org • Multiphase Reactors 369

 CH06CH16-Joshi ARI 15 July 2015 7:42

Multiphase flow models

Break-up and coalescence Direct numerical simulations Discrete particle model

Attachment/detachment Microscale modeling Mesoscale modeling
Aggregation/break up
Mesoscale modeling

Dispersed phase (bubbles, particles, drops)

Population balance model CD, CL,Cn,Ck, h, k i. Velocity field
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

Kernals ii. Turbulence field

iii. Cluster size distribution and dynamics

C1e, C2e,Cm SGS

RSM constants
sk, se, C3e model
Dispersion
Pr Energy
Wall

κ-ε RSM LES

Pragmatic turbulent flow models

Figure 8
Parametric relationships needed in a set of pragmatic multiphase turbulent flow models. Note that ad hoc
extension of k − ε Reynolds stress models (RSM) or large eddy simulation (LES) models into multiphase
domains is not yet well supported by theoretical development (68). Abbreviation: SGS, sub grid scale.

PBM. It is envisaged that, as the computational power increases with time, the role of DNS will
increase. The resulting understanding of fluid mechanics will not only help in the reliable designs
but also result in creative platforms for innovations.
Because the DNS is powerful for understanding the transport phenomena, including interfacial
dynamics, there is a possibility of understanding the fundamental mechanism for the breakup and
coalescence of bubbles and drops, aggregation of particles, and breakup of aggregates. It should
also be possible to understand the mechanism of bubble particle attachment and detachment. The
knowledge of all these phenomena is crucial for the sound application of coarse-grained PBM, in
conjunction with the TFM.
Although DNS simulations enable fully resolved analysis of fluid-particle interaction (without
the need for drag-lift coefficients), the fluid-particle and particle-particle interactions can also be
analyzed in the context of DPM. This technique (with the input of CD , CL , and CV ) gives the
concentration and velocity fields of the particle/bubble/drop phase. It also gives the knowledge of
cluster size distribution and the cluster dynamics.

COMPUTATIONAL FLUID DYNAMICS FRAMEWORK TO ENABLE

INNOVATIONS IN MULTIPHASE REACTOR DESIGN
Sustainable development of chemical process technologies requires increasing reliance on renew-
able resources, concern for the environment, and optimal use of energy resources. These factors
must be considered at the early stage of the design process so that individual process units, such as

370 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

multiphase reactors, separators, or mixers, can be designed by exploring a large design space using
advanced, validated computational tools, when a large degree of freedom is still available, that
are traditionally wasted with ad hoc design decisions. Managing the internal flow structures by
ingenious design assisted by CFD simulations can bring forth innovations. Some early examples
are illustrated below.
The mixing process occurs owing to transport at three levels: molecular, eddy, and bulk con-
vection. Usually the bulk motion is superimposed on either eddy diffusion (macromixing) or
molecular diffusion (micromixing) or both. In a majority of cases of practical interest, it is suffi-
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

cient to estimate the extent of macromixing. Using CFD, Patwardhan & Joshi (130), Nere et al.
(131), Kumaresan & Joshi (132), and Lorenz et al. (133) have shown the occurrence of an optimum
field of convective and dispersive (turbulent) motion with practically the same extent of blending
at substantially reduced power consumption by a factor of 10. Similarly, the CFD can enable the
design of internals (in, e.g., bubble columns or fluidized beds), which gives improved axial mixing
(getting toward plug flow).
For an external loop airlift reactor (EL-ALR), the CFD provides a framework to optimize
mixing, heat transfer coefficient, interfacial area, solid suspension, and mass transfer coefficient,
depending upon the rate-controlling step [see Joshi et al. (134), Lele & Joshi (135, 136), and Roy
et al. (137)]. The optimum geometrical and operating parameters of EL-ALR are widely different
for each of the above design objectives, and significant benefits can be accrued by CFD simulations
in terms of either capital or operating (pressure drop) costs or both.
The design of distributors/spargers governs the performance of a variety of multiphase reactors
(e.g., bubble columns, fluidized beds, three-phase slurry reactors, stirred tanks, packed columns,
fixed beds, and chromatographic separation). CFD enables the design of fractal-type distributors,
which give substantially enhanced performance even at much lower pressure drops across them.
The fractal concept can be extended to impeller designs in stirred tanks.
In froth flotation, particle-bubble attachment is an important step for which a certain level of
turbulence is needed. However, if the turbulence level exceeds a limit, detachment may occur.
Thus, for a given flotation application, there exists an optimum level of turbulence intensity that
should preferably be homogeneous and isotropic throughout the equipment. CFD can enable
optimal designs, resulting in substantially improved recovery at much lower power consumption
and also in reduced costs in crushing and grinding for the preparation of raw material.
Gas-inducing reactors recycle the unreacted gas (or liquid) from the gas space to the impeller
zone through a channel (138). For this purpose, a rotating impeller must create a low-pressure
zone. CFD enables an optimum design of the gas-inducing reactors (139), which can be even ten
times smaller than the classical stirred tanks for a variety of applications, such as hydrogenation
or oxidation using oxygen. Similarly, annular centrifugal extractors can miniaturize extractor and
separator volumes (as compared with classical extractors) by a factor up to 100 when optimized
using CFD simulations.

CONCLUSIONS AND SUGGESTIONS FOR FUTURE WORK

1. In this review, we have mapped out the evolution of ideas and tools used in the design
practice of multiphase reactors, emphasizing the increasing role of computational tools for
rational design procedures and innovative design outcomes.
2. Computational chemistry tools involving DFT and MD are providing insights into events on
the molecular scale pertaining to chemical reactions, intrinsic kinetics, and physical processes
and transport parameters.

www.annualreviews.org • Multiphase Reactors 371

 CH06CH16-Joshi ARI 15 July 2015 7:42

3. Figure 8 illustrates the importance of DNS. In addition to high-fidelity experiments,

DNS probes are emerging as important tools in the estimation of parameters appearing in
coarse-grained models, such as (a) interface force parameters (C D , C L , C V ), (b) kernels for
population balance, and (c) turbulence parameters for k − ε models and RSM. The overall
role of DNS is expected to increase with time.
4. Since 1990, approximately 250 papers have been published on DNS simulation of multi-
phase systems. Significant progress has been reported on a variety of algorithms. However,
the current status permits simulations with a limited particle number and range of Reynolds
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

number (140, 141). Substantial additional work is needed. Specific suggestions have been
made on (a) particle-turbulence interactions, (b) interface force coefficients, and (c) estima-
tion of parameters of lower-order turbulence models.
5. There has long been a need for understanding the origin of clustering and heterogeneity in
multiphase dispersions. In this context, Derksen & Sundaresan (117) have performed DNS
simulations and established the reasons for initiation of clustering. This pioneering work
must be taken forward for understanding transitions in a variety of multiphase dispersions.
6. Practically all research work has been devoted to the interaction of turbulence with solid
particles where the interface is rigid (142). Future work must include DNS analysis of in-
teractions with bubbles and drops where the interface can be rigid, partially mobile, or
completely mobile. Further, bubbles and drops may change shapes and sizes through break-
age and coalescence in multiphase reactor environments, which must be included in the
DNS simulations.
7. The approach for innovative industrial reactor design must include the following: (a) re-
search work on high-fidelity experiments and simulations; (b) development of coarse-grained
models with validated parameters from high-fidelity studies; (c) stepwise procedures for in-
novation, design, and optimization; and (d ) equipment selection. There are almost 400
different types of equipment used in the chemical process industries. A flow chart as well
as detailed procedures for selection, design, and optimization of process equipment and
the possible methods of execution (which principally includes synergistic partnership of
University and Industry) have been described by Mathpati et al. (143).

DISCLOSURE STATEMENT
The authors are not aware of any affiliations, memberships, funding, or financial holdings that
might be perceived as affecting the objectivity of this review.

ACKNOWLEDGMENTS
Funding from Gordon A and Mary Cain Chair program at the Louisiana State University is
gratefully acknowledged.

LITERATURE CITED
1. Levenspiel O. 1998. Chemical Reaction Engineering. Hoboken, NJ: Wiley
2. Fogler HS. 2010. Essentials of Chemical Reaction Engineering. Upper Saddle River, NJ: Prentice Hall
3. Boudart M. 1968. Kinetics of Chemical Processes. Upper Saddle River, NJ: Prentice Hall
4. Sholl D, Steckel JA. 2009. Density Functional Theory: A Practical Introduction. Hoboken, NJ: Wiley
5. Joshi JB, Patil TA, Ranade VV, Shah YT. 1990. Measurement of hydrodynamic parameters in multiphase
sparged reactors. Rev. Chem. Eng. 6:73–227

372 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

6. Marchisio DL, Fox RO. 2007. Multiphase Reacting Flows: Modeling and Simulation. New York: Springer
7. Joshi JB, Nere NK, Rane CV, Murthy BN, Mathpati CS, et al. 2011. CFD simulation of stirred tanks:
comparison of turbulence models. Part II: axial flow impellers, multiple impellers and multiphase dis-
persions. Can. J. Chem. Eng. 89:754–816
8. Bourne JR, Yu SY. 1994. Investigation of micromixing in stirred-tank reactors using parallel reactions.
Ind. Eng. Chem. Res. 33:41–55
9. Ranade VV, Chaudhari R, Gunjal PR. 2011. Trickle Bed Reactors: Reactor Engineering & Applications.
Amsterdam: Elsevier
10. Weekman VW Jr. 1968. A model of catalytic cracking conversion in fixed, moving, and fluid-bed reactors.
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

Ind. Eng. Chem. Process Des. Dev. 7:90–95

11. Jesus NJC, Melo PA, Nele M, Pinto JC. 2011. Oscillatory behaviour of an industrial slurry polyethylene
reactor. Can. J. Chem. Eng. 89:582–92
12. Saccone G, Salatino P, Lettieri P. 2010. CFD Modeling: of Fluidized Bed Reactors. Saarbrücken, Ger.:
VDM Verlag
13. Deen NG, Annaland MVS, Van der Hoef MA, Kuipers JAM. 2007. Review of discrete particle modeling
of fluidized beds. Chem. Eng. Sci. 62:28–44
14. Anderson TB, Jackson R. 1967. A fluid mechanical description of fluidized beds. Ind. Eng. Chem. Fundam.
6:527–39
15. Vial C, Poncin S, Wild G, Midoux N. 2002. Experimental and theoretical analysis of the hydrodynamics
in the riser of an external loop airlift reactor. Chem. Eng. Sci. 57:4745–62
16. Deckwer W-D. 1991. Bubble Column Reactors. Hoboken, NJ: Wiley
17. Sommerfeld M. 2004. Bubbly Flows: Analysis, Modeling and Calculation (Heat and Mass Transfer). New
York: Springer
18. Wang TF, Wang JF, Jin Y. 2007. Slurry reactors for gas-to-liquid processes: a review. Ind. Eng. Chem.
Res. 46:5824–47
19. Ranade VV. 2001. Computational Flow Modeling for Chemical Reactor Engineering. Waltham, MA:
Academic
20. Jakobsen HA. 2014. Chemical Reactor Modeling: Multiphase Reactive Flows. New York: Springer
21. Norskov JK, Abild-Pedersen F, Studt F, Bligaard T. 2011. Density functional theory in surface chemistry
and catalysis. Proc. Natl. Acad. Sci. USA 108:937–43
22. Bell AT, Head-Gordon M. 2011. Quantum mechanical modeling of catalytic processes. Annu. Rev. Chem.
Biomol. Eng. 2:453–77
23. Gubbins KE, Liu Y-C, Moore JD, Palmer JC. 2011. The role of molecular modeling in confined systems:
impact and prospects. Phys. Chem. Chem. Phys. 13:58–85
24. Miller WH. 2001. The semiclassical initial value representation: a potentially practical way for adding
quantum effects to classical molecular dynamics simulations. J. Phys. Chem. A 105:2942–55
25. Hoomans BPB, Kuipers JAM, Briels WJ, van Swaaij WPM. 1996. Discrete particle simulation of bubble
and slug formation in a two-dimensional gas-fluidised bed: a hard-sphere approach. Chem. Eng. Sci.
51:99–118
26. Tryggvason G, Scardovelli R, Zaleski S. 2011. Direct Numerical Simulations of Gas-Liquid Multiphase
Flows. Cambridge: Cambridge Univ. Press
27. Couper JR, Penney WR, Fair JR, Walas SM. 2012. Chemical Process Equipment: Selection and Design.
Oxford, UK: Butterworth-Heinemann. 3rd ed.
28. Madrigal A. 2009. Humans have made, found or used over 50 million unique chemicals. Wired.com, Sept.
9. http://www.wired.com/2009/09/humans-have-made-found-or-used-over-50-million-unique-
chemicals/
29. Davis GE. 1904. A Handbook Of Chemical Engineering, Volumes I and II. Manchester: Davis Bros. 2nd ed.
30. Groggins PH. 1947. Unit Processes in Organic Synthesis. New York: McGraw Hill. 3rd ed.
31. Danckwerts PV. 1970. Gas-Liquid Reactions. New York: McGraw-Hill Chem. Eng. Ser.
32. Walas SM. 1990. Chemical Process Equipment: Selection and Design. Oxford, UK: Butterworth-Heinemann
33. Bałdyga J, Bourne JR. 1999. Turbulent Mixing and Chemical Reactions. Hoboken, NJ: Wiley
34. Coker AK. 2001. Modeling of Chemical Kinetics and Reactor Design. Houston: Gulf Prof. Publ.

www.annualreviews.org • Multiphase Reactors 373

 CH06CH16-Joshi ARI 15 July 2015 7:42

35. van Mourik T, Bühl M, Gaigeot M-P. 2014. Density functional theory across chemistry, physics and
biology. Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 372:20120488
36. Kee RJ, Rupley FM, Miller JA. 1989. Chemkin-II: A Fortran Chemical Kinetics Package for the Analysis of
Gas-Phase Chemical Kinetics. Livermore, CA: Sandia Natl. Labs
37. Drennan SA. 2009. Enabling Detailed Chemistry: Reaction Design Provides CHEMKIN-CFD for Free to
ANSYS FLUENT Users Who Are Looking to Improve the Accuracty of Their Flow Simulations. ANSYS Ad-
vant. http://www.ansys.com/staticassets/ANSYS/staticassets/resourcelibrary/article/AA-V3-I2-
Enabling-Detailed-Chemistry.pdf
38. Hill CG. 1977. An Introduction to Chemical Engineering Kinetics and Reactor Design. Hoboken, NJ: John
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

Wiley & Sons

39. Miller JA, Kee RJ, Westbrook CK. 1990. Chemical-kinetics and combustion modeling. Annu. Rev. Phys.
Chem. 41:345–87
40. Coltrin ME, Kee RJ, Miller JA. 1984. A mathematical-model of the coupled fluid-mechanics and
chemical-kinetics in a chemical vapor-deposition reactor. J. Electrochem. Soc. 131:425–34
41. Miller JA, Branch MC, Kee RJ. 1981. A chemical kinetic-model for the selective reduction of nitric-oxide
by ammonia. Combust. Flame 43:81–98
42. Castro A, Appel H, Oliveira M, Rozzi CA, Andrade X, et al. 2006. octopus: a tool for the application of
time-dependent density functional theory. Phys. Status Solidi B Basic Solid State Phys. 243:2465–88
43. Mozharov S, Nordon A, Littlejohn D, Wiles C, Watts P, et al. 2011. Improved method for kinetic
studies in microreactors using flow manipulation and noninvasive Raman spectrometry. J. Am. Chem.
Soc. 133:3601–8
44. Zinoveva S, De Silva R, Louis RD, Datta P, Kumar C, et al. 2007. The wet chemical synthesis of Co
nanoparticles in a microreactor system: A time-resolved investigation by X-ray absorption spectroscopy.
Nucl. Instrum. Methods Phys. Res. Sect. A Accel. Spectrom. Detect. Assoc. Equip. 582:239–41
45. Kinne M, Fuhrmann T, Whelan CM, Zhu JF, Pantförder J, et al. 2002. Kinetic parameters of CO
adsorbed on Pt(111) studied by in situ high resolution x-ray photoelectron spectroscopy. J. Chem. Phys.
117:10852–59
46. Chen C, Yang BL, Yuan J, Wang ZW, Wang LY. 2007. Establishment and solution of eight-lump
kinetic model for FCC gasoline secondary reaction using particle swarm optimization. Fuel 86:2325–
32
47. Gupta RK, Kumar V, Srivastava VK. 2010. Modeling of fluid catalytic cracking riser reactor: a review.
Int. J. Chem. React. Eng. 8:1542–6580
48. Ancheyta-Juárez J, López-Isunza F, Aguilar-Rodriguez E, Moreno-Mayorga JC. 1997. A strategy for
kinetic parameter estimation in the fluid catalytic cracking process. Ind. Eng. Chem. Res. 36:5170–74
49. Hillewaert LP, Dierickx JL, Froment GF. 1988. Computer generation of reaction schemes and rate-
equations for thermal cracking. Aiche J. 34:17–24
50. Clymans PJ, Froment GF. 1984. Computer-generation of reaction paths and rate-equations in the
thermal-cracking of normal and branched paraffins. Comput. Chem. Eng. 8:137–42
51. Van der Laan GP, Beenackers A. 1999. Kinetics and selectivity of the Fischer-Tropsch synthesis: a
literature review. Catal. Rev. Sci. Eng. 41:255–318
52. Joshi JB, Doraiswamy LK. 2008. Chemical reaction engineering. In Albright’s Chemical Engineering
Handbook, ed. LF Albright, pp. 737–968. Boca Raton, FL: CRC Press. 1909 pp.
53. Joshi JB, Tabib MV, Deshpande SS, Mathpati CS. 2009. Dynamics of flow structures and transport
phenomena, 1. Experimental and numerical techniques for identification and energy content of flow
structures. Ind. Eng. Chem. Res. 48:8244–84
54. Pope SB. 2000. Turbulent Flows. Cambridge, UK: Cambridge Univ. Press
55. Goldstein RJ. 1996. Fluid Mechanics Measurements. Washington, DC: Taylor & Francis
56. Babkovskaia N, Haugen NEL, Brandenburg A. 2011. A high-order public domain code for direct nu-
merical simulations of turbulent combustion. J. Comput. Phys. 230:1–12
57. Haile JM. 1997. Molecular Dynamics Simulation: Elementary Methods. Hoboken, NJ: Wiley Intersci.
58. ben-Avraham D, Havlin S. 2000. Diffusion and Reactions in Fractals and Disordered Systems. Cambridge,
UK: Cambridge Univ. Press

374 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

59. Aris R. 1975. The Mathematical Theory of Diffusion and Reaction in Permeable Catalysts: Vol. 1: The Theory
of the Steady State. Oxford, UK: Oxford Univ. Press
60. Astarita G. 1967. Mass Transfer with Chemical Reaction. Amsterdam: Elsevier
61. Spalart PR, ed. 1999. Strategies for Turbulence Modeling and Simulations. Amsterdam: Elsevier
62. Dimotakis PE. 2005. Turbulent mixing. Annu. Rev. Fluid Mech. 37:329–56
63. Pope SB. 2010. Self-conditioned fields for large-eddy simulations of turbulent flows. J. Fluid Mech.
652:139–69
64. Wilcox DC. 1993. Turbulence Modeling for CFD. La Cañada Flintridge, CA: DCW Ind.
65. Haworth DC. 2010. Progress in probability density function methods for turbulent reacting flows. Prog.
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

Energy Combust. Sci. 36:168–259

66. Fox RO. 2003. Computational Models for Turbulent Reacting Flows. Cambridge: Cambridge Univ. Press
67. Pope SB. 1981. Transport equation for the joint probability density function of velocity and scalars in
turbulent flow. Phys. Fluids 24:588–96
68. Fox RO. 2014. On multiphase turbulence models for collisional fluid-particle flows. J. Fluid Mech.
742:368–424
69. Whitaker S. 2010. The Method of Volume Averaging (Theory and Applications of Transport in Porous Media).
New York: Springer
70. Whitaker S. 1986. Flow in porous media I: a theoretical derivation of Darcy’s law. Transp. Porous Media
1:3–25
71. Slattery JC. 1969. Single-phase flow through porous media. AIChE J. 15:866–72
72. Capriz G, Ghionna VN, Giovine P. 2002. Modeling and Mechanics of Granular and Porous Material. Basel,
Switz.: Birkhäuser
73. van Sint Annaland M, Bokkers GA, Goldschmidt MJV, Olaofe OO, van der Hoef MA, Kuipers JAM.
2009. Development of a multi-fluid model for poly-disperse dense gas–solid fluidised beds, part I: model
derivation and numerical implementation. Chem. Eng. Sci. 64:4222–36
74. Gidaspow D, Jung J, Singh RK. 2004. Hydrodynamics of fluidization using kinetic theory: an emerging
paradigm. Powder Technol. 148:123–41
75. Gidaspow D. 1994. Multiphase Flow and Fluidization: Continuum and Kinetic Theory Descriptions. Waltham,
MA: Academic
76. Li J, Kuipers JAM. 2007. Effect of competition between particle-particle and gas-particle interactions
on flow patterns in dense gas-fluidized beds. Chem. Eng. Sci. 62:3429–42
77. Wu CL, Berrouk AS, Nandakumar K. 2009. Three-dimensional discrete particle model for gas–solid
fluidized beds on unstructured mesh. Chem. Eng. J. 152:514–29
78. Hoomans BPB, Kuipers JAM, Briels WJ, van Swaaij WPM. 1996. Discrete particle simulation of bubble
and slug formation in a two-dimensional gas-fluidised bed: a hard-sphere approach. Chem. Eng. Sci.
51:99–118
79. Ramkrishna D. 2000. Population Balances Theory and Applications to Particulate Systems in Engineering.
Waltham, MA: Academic
80. Ramkrishna D, Singh MR. 2014. Population balance modeling: current status and future prospects.
Annu. Rev. Chem. Biomol. Eng. 5:123–46
81. Sporleder F, Borka Z, Solsvik J, Jakobsen HA. 2012. On the population balance equation. Rev. Chem.
Eng. 28:149–69
82. Yeoh GH, Cheung CP, Tu J. 2014. Multiphase Flow Analysis Using Population Balance Modeling: Bubbles,
Drops and Particles. Amsterdam: Butterworth-Heinemann
83. Becker PJ, Puel F, Dubbelboer A, Janssen J, Sheibat-Othman N. 2014. Coupled population balance–
CFD simulation of droplet breakup in a high pressure homogenizer. Comput. Chem. Eng. 68:140–50
84. Cheng JC, Fox RO. 2010. Kinetic modeling of nanoprecipitation using CFD coupled with a population
balance. Ind. Eng. Chem. Res. 49:10651–62
85. Chen P, Sanyal J, Duduković MP. 2005. Numerical simulation of bubble columns flows: effect of different
breakup and coalescence closures. Chem. Eng. Sci. 60:1085–101
86. Smoluchowski MV. 1916. Drei Vorträge über Diffusion, Brownsche Molekularbewegung und
Koagulation von Kolloidteilchen. Z. Angew Phys 17:557–71

www.annualreviews.org • Multiphase Reactors 375

 CH06CH16-Joshi ARI 15 July 2015 7:42

87. Sajjadi B, Raman AAA, Shah RSSRE, Ibrahim S. 2013. Review on applicable breakup/coalescence models
in turbulent liquid-liquid flows. Rev. Chem. Eng. 29:131–58
88. Tenneti S, Subramaniam S. 2014. Particle-resolved direct numerical simulation for gas-solid flow model
development. Annu. Rev. Fluid Mech. 46: 199–230
89. Veeramani C, Minev PD, Nandakumar K. 2007. A fictitious domain formulation for flows with rigid
particles: a non-Lagrange multiplier version. J. Comput. Phys. 224:867–79
90. Glowinski R, Pan TW, Hesla TI, Joseph DD, Periaux J. 2001. A fictitious domain approach to the direct
numerical simulation of incompressible viscous flow past moving rigid bodies: application to particulate
flow. J. Comput. Phys. 169:363–426
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

91. de Jong JF, van Sint Annaland M, Kuipers JAM. 2012. Membrane-assisted fluidized beds—part 1: de-
velopment of an immersed boundary discrete particle model. Chem. Eng. Sci. 84:814–21
92. Aidun CK, Clausen JR. 2010. Lattice-Boltzmann method for complex flows. Annu. Rev. Fluid Mech.
42:439–72
93. Ladd AJC, Verberg R. 2001. Lattice-Boltzmann simulations of particle-fluid suspensions. J. Stat. Phys.
104:1191–251
94. Natarajan R, Acrivos A. 1993. The instability of the steady flow past spheres and disks. J. Fluid Mech.
254:323–44
95. Kriebitzsch SHL, van der Hoef MA, Kuipers JAM. 2013. Fully resolved simulation of a gas-fluidized
bed: a critical test of DEM models. Chem. Eng. Sci. 91:1–4
96. Baltussen MW, Seelen LJH, Kuipers JAM, Deen NG. 2013. Direct numerical simulations of gas–liquid–
solid three phase flows. Chem. Eng. Sci. 100:293–99
97. Deen NG, Kriebitzsch SHL, van der Hoef MA, Kuipers JAM. 2012. Direct numerical simulation of flow
and heat transfer in dense fluid-particle systems. Chem. Eng. Sci. 81:329–44
98. Jain D, Deen NG, Kuipers JAM, Antonyuk S, Heinrich S. 2012. Direct numerical simulation of particle
impact on thin liquid films using a combined volume of fluid and immersed boundary method. Chem.
Eng. Sci. 69:530–40
99. Dijkhuizen W, Roghair I, Van Sint Annaland M, Kuipers JAM. 2010. DNS of gas bubbles behaviour using
an improved 3D front tracking model—drag force on isolated bubbles and comparison with experiments.
Chem. Eng. Sci. 65:1415–26
100. Kuipers JAM, van Swaaij WPM. 1997. Application of computational fluid dynamics to chemical reaction
engineering. Rev. Chem. Eng. 13:1–118
101. Bekdemir C, Somers B, de Goey P. 2014. DNS with detailed and tabulated chemistry of engine relevant
igniting systems. Combust. Flame 161:210–21
102. Chakraborty N, Lipatnikov AN. 2013. Effects of Lewis number on conditional fluid velocity statistics
in low Damköhler number turbulent premixed combustion: a direct numerical simulation analysis. Phys.
Fluids 25:045101
103. Luong MB, Luo ZY, Lu TF, Chung SH, Yoo CS. 2013. Direct numerical simulations of the ignition of
lean primary reference fuel/air mixtures with temperature inhomogeneities. Combust. Flame 160:2038–
47
104. Minamoto Y, Dunstan TD, Swaminathan N, Cant RS. 2013. DNS of EGR-type turbulent flame in
MILD condition. Proc. Combust. Inst. 34:3231–38
105. Mousazadeh F, van Den Akker HEA, Mudde RF. 2013. Direct numerical simulation of an exothermic
gas-phase reaction in a packed bed with random particle distribution. Chem. Eng. Sci. 100:259–65
106. Hetsroni G. 1989. Particles-turbulence interaction. Int. J. Multiphase Flow 15:735–46
107. Joshi JB. 1983. Solid-liquid-fluidized beds—some design aspects. Trans. Inst. Chem. Eng. A Chem. Eng.
Res. Des. 61:143–61
108. Pandit AB, Joshi JB. 1998. Pressure drop in fixed, expanded, fluidized beds and packed columns. Rev.
Chem. Eng. 14:321–71
109. Zhao F, van Wachem BGM. 2013. Direct numerical simulation of ellipsoidal particles in turbulent
channel flow. ACTA Mech. 224:2331–58
110. Gui N, Yan J, Fan JR, Cen KF. 2013. A DNS study of the effect of particle feedback in a gas-solid three
dimensional plane jet. Fuel 106:51–60

376 Joshi · Nandakumar

 CH06CH16-Joshi ARI 15 July 2015 7:42

111. Kondaraju S, Choi M, Xu XF, Lee JS. 2012. Direct numerical simulation of modulation of isotropic
turbulence by poly-dispersed particles. Int. J. Numer. Methods Fluids 69:1237–48
112. Dritselis CD, Vlachos NS. 2008. DNS/LES study of fluid-particle interaction in a turbulent channel
flow at a low Reynolds number. Numer. Anal. Appl. Math. 1048:735–38
113. Derksen JJ. 2012. Direct numerical simulations of aggregation of monosized spherical particles in ho-
mogeneous isotropic turbulence. AIChE J. 58:2589–600
114. Xu Y, Subramaniam S. 2010. Effect of particle clusters on carrier flow turbulence: a direct numerical
simulation study. Flow Turbul. Combust. 85:735–61
115. Wei M, Wang LM, Li JH. 2013. Unified stability condition for particulate and aggregative fluidization-
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

exploring energy dissipation with direct numerical simulation. Particuology 11:232–41

116. Shnip AI, Kolhatkar RV, Dinkar S, Joshi JB. 1992. Criterion for transition from homogeneous to het-
erogeneous regime in two dimensional bubble columns. Int. J. Multiphase Flow 18:705–26
117. Derksen JJ, Sundaresan S. 2007. Direct numerical simulations of dense suspensions: wave instabilities in
liquid-fluidized beds. J. Fluid Mech. 587:303–36
118. Anderson TB, Jackson R. 1969. A fluid mechanical description of fluidized beds—comparison of theory
and experiment. Ind. Eng. Chem. Fundam. 8:137–44
119. Anderson TB, Jackson R. 1968. Fluid mechanical description of fluidized beds—stability of state of
uniform fluidization. Ind. Eng. Chem. Fundam. 7:12–21
120. Batchelor GK. 1988. A new theory of the instability of a uniform fluidized bed. J. Fluid Mech. 193:75–
110
121. Didwania AK, Homsy GM. 1982. Resonant sideband instabilities in wave-propagation in fluidized-beds.
J. Fluid Mech. 122:433–38
122. Joshi JB, Deshpande NS, Dinkar M, Phanikumar DV. 2001. Hydrodynamic stability of multiphase
reactors. Adv. Chem. Eng. 26:1–130
123. Thorat BN, Shevade AV, Bhilegaonkar KN, Aglawe RH, Veera UP, et al. 1998. Effect of sparger design
and height to diameter ratio on fractional gas hold-up in bubble columns. Chem. Eng. Res. Des. 76:823–
34
124. Thorat BN, Joshi JB. 2004. Regime transition in bubble columns: experimental and predictions. Exp.
Therm. Fluid Sci. 28:423–30
125. Sundaresan S. 2003. Instabilities in fluidized beds. Annu. Rev. Fluid Mech. 35:63–88
126. Ghatage SV, Peng Z, Sathe MJ, Doroodchi E, Padhiyar N, et al. 2014. Stability analysis in solid-liquid
fluidized beds: experimental and computational. Chem. Eng. J. 256:169–86
127. Deen NG, Peters EAJF, Padding JT, Kuipers JAM. 2014. Review of direct numerical simulation of
fluid-particle mass, momentum and heat transfer in gas-slid flows. Chem. Engg. Sci 116:710–24
128. Deen NG, Kuipers JAM. 2014. Direct numerical simulation of fluid flow accompanied by coupled mass
and heat transfer in dense fluid–particle systems. Chem. Eng. Sci. 116:645–56
129. Mathey F. 2013. Numerical up-scaling approach for the simulation of heat-transfer in randomly packed
beds. Int. J. Heat Mass Transf. 61:451–63
130. Patwardhan AW, Joshi JB. 1999. Design of gas inducing reactor. Ind. Eng. Chem. Res. 37:49–80
131. Nere NK, Patwardhan AW, Joshi JB. 2003. Liquid-phase mixing in stirred vessels: turbulent flow regime.
Ind. Eng. Chem. Res. 42:2661–98
132. Kumaresan T, Joshi JB. 2006. Effect of impeller design on the flow pattern and mixing in stirred tanks.
Chem. Eng. J. 115:173–93
133. Lorenz O, Schumpe A, Ekambara K, Joshi JB. 2005. Liquid phase axial mixing in bubble columns
operated at high pressures. Chem. Eng. Sci. 60:3573–86
134. Joshi JB, Ranade VV, Gharat SD, Lele SS. 1990. Sparged loop reactors. Can. J. Chem. Eng. 68:705–41
135. Lele SS, Joshi JB. 1994. Some design aspects of air-lift fluidized bed. Chem. Eng. J. 55:145–46
136. Lele SS, Joshi JB. 1993. Hydrodynamics and mass transfer in air-lift reactors. In Encyclopedia of Fluid
Mechanics, Supplement 2, ed. NP Cheremisinoff, pp. 409–510. Houston: Gulf Prof. Publ.
137. Roy S, Joshi JB. 2008. CFD study of mixing characteristics of bubble column and external loop airlift
reactor. Asia-Pac. J. Chem. Eng. 3:97–105
138. Joshi JB, Sharma MM. 1977. Mass-transfer and hydrodynamic characteristics of gas inducing type of
agitated contactors. Can. J. Chem. Eng. 55:683–95

www.annualreviews.org • Multiphase Reactors 377

 CH06CH16-Joshi ARI 15 July 2015 7:42

139. Murthy BN, Deshmukh NA, Patwardhan AW, Joshi JB. 2007. Hollow self-inducing impellers: flow
visualization and CFD simulation. Chem. Eng. Sci. 62:3839–48
140. Reddy RK, Joshi JB, Nandakumar K, Minev PD. 2010. Direct numerical simulations of a freely falling
sphere using fictitious domain method: breaking of axisymmetric wake. Chem. Eng. Sci. 65:2159–71
141. Reddy RK, Jin S, Nandakumar K, Minev PD, Joshi JB. 2010. Direct numerical simulation of free falling
sphere in creeping flow. Int. J. Comput. Fluid Dyn. 24:109–20
142. Reddy RK, Sathe MJ, Joshi JB, Nandakumar K, Evans GM. 2013. Recent developments in experimental
(PIV) and numerical (DNS) investigation of solid-liquid fluidized beds. Chem. Eng. Sci. 92:1–12
143. Mathpati CS, Tabib MV, Deshpande SS, Joshi JB. 2009. Dynamics of flow structures and transport
Downloaded from www.annualreviews.org. Guest (guest) IP: 217.218.30.161 On: Sat, 17 Aug 2024 02:50:41

phenomena, 2. Relationship with design objectives and design optimization. Ind. Eng. Chem. Res. 48:8285–
311

378 Joshi · Nandakumar