Sunteți pe pagina 1din 19

845 chemical

chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 845

DN formed. It is not sure, whether the terminology (if any) of cles, where the deformation and the motion can strongly
numbers thus produces is established and settled. be coupled. Since We stems from the Lagrangian accelera-
tion term, it also reflects the particle deformation due to the
5.4. Boundary conditions: no slip and free-slip converging/diverging streamlines. Low We means low defor-
mation, and vice versa. Other situation is walking over water
Initial and boundary conditions inseparably pertain to the (hydrophobic feet assumed), where the inertia force (dynamic
governing equations, and should also be scaled. The initial load) produced by the walker must not exceed the bear-
conditions (IC) are given functions of the spatial distribution of ing power of the ‘flexible membrane’ of the surface tension,
the variables in the initial time. There is actually nothing to whence low We is required. Note that other mechanisms of
scale. On the other hand, the boundary conditions (BC) can be water walking exist, well beyond the bearing capacity of the
more elaborate. BC are equations that the variables must fulfil surface tension. For instance, small reptiles (genus Basiliscus)
at the boundary ∂˝ of the flow domain ˝, during the whole can run over water using a supporting impulse from a liquid jet
process of solution. As such, they must also be scaled. created by a specific shape of their feet. Here, other numbers
In single-phase flow, bounded by a rigid wall, the no-slip besides We play a role too.
BC applies, which is simple and merely says that the speed is The Capillary number Ca compares the viscous and capillary
zero at the wall, v = 0. There is not much to scale, besides v/V, forces. It is relevant when viscous forces dominate, i.e. slow
which is kind of unnecessary. flows and small scales. The typical situation is drainage of a
In multi-phase flow, where deformable fluid interfaces thin liquid film between two interfaces, at least one of them
separate immiscible phases with certain interfacial (surface) is fluid (to have ˛ in play). The typical applications abound:
tension ˛, the free-slip BC applies, which is far from being interactions of fluid particles with themselves, with solids,
trivial. It is a force balance at the fluid–fluid interface. Decom- and with rigid walls (coalescence, bouncing, adhesion), fre-
posing the force into normal (normal stress, pressure) and quently encountered in bubble columns, flotation columns,
tangential (shear stress) components, BC says: jump in fluid extractors, etc.
pressure is balanced by tension ˛, and jump in fluid stress The Bond number Bo and the Eotvos number Eo mean the
is balanced by surface gradient of tension S ˛ (i.e. gradient same: the ratio of gravity (˘) or buoyancy ( ˘) forces to the
along interface). There are important DN stemming from the capillary forces. The typical situation is deformation of a stag-
free-slip BC that relate to deformation of fluid particles and nant drop sitting on a horizontal plane, where Bo compares
interfaces. the hydrostatic pressure across the drop (˘gL) with the capil-
lary pressure (˛/L) inside the drop. Naturally, the length scale
5.4.1. Normal component of free-slip BC L must be the drop size. Similar situation is when bubbles or
By scaling, the normal component gives a number that com- drops are entrapped below a horizontal wall, or attached to a
pares the fluid pressure and the Laplace pressure produced by needle (L ∼ orifice size). From the equilibrium deformation of
the interface tension, N = P/(˛/L), sometimes called the Laplace bubbles and drops, the static value of ˛ can be obtained. Other
number, La. It follows from the physical situation, that the situation is standing on the water surface, where the static
length scale L cannot be chosen arbitrarily but must relate to load of the stander must not exceed the strength of the surface
the curvature of the interface that produces the correspond- tension, whence low Bo is required. An important quantity
ing capillary pressure ˛/L [Pa]. Choosing something else would relates to Bo, the capillary length C . Consider that gravity
be physical nonsense. Consequently, for L should be taken the forces dominate on large length scales at Bo ≥ 1, and capillary
radius of curvature, the size of a bubble or drop, or perhaps forces dominated on small length scales at Bo ≤ 1. Find the crit-
even the capillary length. What can be put for the fluid pres- ical (capillary) length scale C , where Bo = 1. Quickly we see that
sure scale P? Generally, anything from (5.2.7), provided that it the capillarity prevails over gravity when L ≤ C = (˛/( )˘g)1/2 .
The typical situation is the capillary elevation of the water sur-
corresponds to the physical situation in question. Upon differ-
face near the wall of the glass, which occurs in the range up
ent scalings for pressure P, the number N takes the following
to ∼ C from the wall. Beyond this, the gravity prevails and the
forms and names:

surface is flat.
˘L2 V (unsteadiness)
Unsteady scaling of P : Un = , Besides the four DN arising from the normal component of
˛T (capillarity)
the BC in (5.4.1), another composite number is often used. The
( )˘LV 2 (inertia)
Convective scaling of P : We = , Morton number Mo is an artificial conglomerate that combines
˛ (capillarity)
Re and Fr from the governing equations, and We from the BC,
˝V (viscosity)
Viscous scaling of P : Ca = , (5.4.1) in such a way, that it contains only the material properties of
˛ (capillarity)
fluids (plus gravity):
( )˘gL2

Gravitation scaling of P : Bo = Eo =
˛
We3
(gravity or buoyancy) Mo = . (5.4.2)
. Re4 Fr

(capillarity) effects are important. Typical situations are, e.g. bouncing


at a wall and path instability of rising/falling fluid parti-
Here ( )˘ means that both ˘ and ˘ can be used, putting
stress either on gravitational or buoyant aspect of We and Bo.
The unsteady number Un turns into the Weber number upon
the convective scaling for time T = L/V.
The Weber number We compares the inertia and capillary
forces. It applies to deformation of bubbles and drops at
free rise or on collisions with an obstacle where dynamic
846 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 846

The buoyant version reads Mo = ( ˘)3 ˝4 g/˘4 ˛ 3 = g( ˘)3 4 /˛ 3 ,


and gravity version reads Mo = g˝4 /˘˛ 3 = g ˝3 /˛ 3 . It can also be
obtained by DA, by forming a dimensionless group of (g, ˝, ˘,
˛).

5.4.2. Tangential component of free-slip BC


By scaling, the tangential component gives a number that
compares the surface gradient of the interface tension S ˛
and the fluid shear stress ˇ, N = S ˛/ˇ. Since the quantity S ˛
847 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 847

is called the Marangoni stress ˇ Ma , the number N is called the may suspect that effects of the latter are more frequent and
Marangoni number Ma: more important. The most of the common liquids (either
pure or in mixtures or with surfactants) have ˛ in the range
ˇMa

Ma = . (5.4.3) ∼0.02–0.08 N/m, i.e. differing by factor of 4. The effects pro-


ˇ
duced by surfactants in gas–liquid systems are enormous,
In case of a flat surface, where a tension difference ˛ within orders of magnitude. Moreover, ˛ itself influence
develops over a length L, the surface gradient can be estimated directly only relatively few hydromechanical processes. Con-
as S ˛ ∼ ˛/L. The fluid stress can be estimated by the viscous sider for instance, the bubble formation process. The resulting
scaling as ˇ ∼ ˝V/L, which gives bubble size depends directly on ˛ only at the very low-gas flow
where the equilibrium between buoyancy and surface force is
˛/L ˛ reached during the quasi-steady inflation. This is not the case
Ma = = . (5.4.4)
˝V/L ˝V of real gas–liquid contactors, where the bubble size depends
on many other effects, involved in violent breakup of irregular
It may be difficult to estimate the velocity scale V for com-
gas jets produced by the gas distributor. On the other hand, the
plex motions near interfaces on the convection basis, and the
magnitude of S ˛ is difficult to estimate. Taking it ∼ ˛/L, we
diffusive scaling is therefore applied. V is expressed with help
have ˛ within the factor of 4 (i.e. almost ‘constant’ for all liq-
of the diffusivity of the agent that causes the interface tension
uids), but the relevant length scale L can cover very vast range
variance (heat, surfactant), V ∼ (diffusivity)/L. It is consistent of values. Consider a bubble of size L. The surface gradient
with using the viscous scaling for the fluid stress: in both can develop over any distance, from 0 to ∼L, making ˛/L very
cases, the molecular transport is reflected. much varying quantity. Moreover, the gradient formation and
The specific formulation depends on the physical process evolution is a dynamic process, with fast temporal changes,
that causes the variation of ˛ along the interface. Two situa- which depends on the flow situation (unlike ˛, which is mate-
tions usually occur: variation of ˛ is caused by a difference in rial property of static equilibrium). Also, all real processes
temperature or in surfactant concentration c. In the former take place in systems that contain ‘impurities’, which read-
case, it is ˛ = (d˛/d )·(∂ /∂ϕ) and V = ˚/L. In the latter case, it ily act as surface active agents (most of chemical compound,
is ˛ = (d˛/dc)·(∂c/∂ϕ) and V = D/L. The Marangoni number then e.g. reactants). Therefore, no surprise, that bubbling into two
reads ‘extremely different’ liquids with ˛ = 0.02 and 0.08 N/m gives
⎧ almost identical results, while re-distilled water and tap water
L d˛ ∂
. . (thermocapillarity), may differ by tenth of percents in gas holdup. For these rea-
˝˚ d ∂ϕ
Ma = (5.4.5) sons, the gas holdup correlations involving ˛ are extremely
L d˛ ∂c
. . (surfactants),
˝D dc ∂ϕ unreliable, with respect to this particular variable.
Better understanding the surface processes and their
where ϕ is the angle along the interface of a bubble or drop, Surface tension effects are very complex and very tricky.
and L corresponds to their size. To close the problem, we must Surfactants cause large effects even at trace amounts. It is
find the dependence of the interface tension on the agent con- difficult to separate the effect of ˛ and that of S ˛. One
tent, ˛( ) and ˛(c), as well as the spatial distribution along the
interface of the agent itself, (x)S and c(x)S . The help comes
from the thermodynamics and physical chemistry of surfaces,
where the relation between ˛ and ( , c)S at the interface is
determined, and the relation between ( , c)S at the interface
and that in the bulk ( , c)bulk are established too. Often, the
equilibrium between the bulk and interface is assumed (for-
mulas are then called ‘isotherms’). The problem of finding the
interfacial distribution of (x)S and c(x)S is much more compli-
cated. Under severe assumptions on the flow and transport of
and c we can make simplified theories. Otherwise, we are on
mercy of numerical experiments by CFD, where all processes
are strongly coupled (flow, transport, adsoption/desorption
kinetics).
The significance of the Marangoni number is that its large
value indicates the presence of strong Marangoni stresses.
They are important in several situations. They delay the rise
of drops and bubbles in ‘contaminated’ water. They delay
the drainage of the film between interfaces, which suppress
the coalescence (bubble columns, extractors) and adhesion
(flotation). Besides, it generates many other interesting phe-
nomena. For instance, at the thermocapillary migration, fine
bubbles can ‘rise down’ in a liquid with inverse temperature
gradient (hot-bottom cold-top arrangement). Similar is the
electrocapillary motion, where the surface variation of ˛ is
caused by the electric charges.
848 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 848

proper scaling is needed. A qualitative sketch of the surfac-


tant action is shown in Fig. 3. Based on their tendency to
gather at the interface, two classes of surfactants can be dis-
tinguished: positive and negative. The positive surfactants are
attracted to the interface, their surface concentration is larger
than the bulk concentration (cs > cb ). Typically, they are organic
substances (emulsifiers, detergents, tensides, wetting agents,
etc.) that decrease the surface tension significantly. The neg-
ative surfactants are repelled from the interface, their surface
concentration is smaller than the bulk concentration (cs < cb ).
Typically, they are inorganic substances (salts of mineral acids,
electrolytes, ionic solutions, etc.) that increase the surface ten-
sion only slightly. During the bubble rise, its nose feels the bulk
concentration (cb ). Its tail feels a generally different concen-
tration, resulting from the adsorption/desorption transport
processes occurring between the bulk and the interface, as
the liquid passed around the bubble. When an equilibrium is
reached, the rear concentration can be denoted as ‘equilibrium
concentration’ (ce ). Obviously, the positive surfactants have
ce > cb , while the negative ce < cb . However, despite the opposite
concentration profiles along the interface of the positive and
negative surfactants, their surface tension profile is the same.
In both cases, the rear ˛ is low and the front ˛ is large. This
interfacial surface tension gradient ( S ˛) is a tensile force that
generates the Marangoni stresses. These stresses move the
liquid material elements along the interface, in the direction
opposite to the main flow over the bubble. The increase in the
resistance force is the natural result (retardation of bubble rise
in contaminated media). At the first guess, the students (and
not only them!) would say, that if the positive surfactant delays
the bubble rise, so the negative surfactant will accelerate it,
which is, however, not so.
849 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 849

Fig. 3 – Positive and negative surfactants in air–water system. (a) Positive surfactants are attracted to surface: nonpolar parts
point to nonpolar air, polar parts ∗ remain in polar water. Negative surfactants are repelled from surface: polar ions ∗ are
repelled by nonpolar air and attracted by polar water. (b) Concentration profile perpendicular to surface: positive surfactants
have cs > cb , negative surfactants cs < cb . (c) Positive surfactants decrease surface tension (much). Negative surfactants
increase surface tension (little). (d) Concentration profile along surface of rising bubble: positive surfactants have ce > cb ,
negative surfactants ce < cb . (e) Surface tension profile along bubble surface: both positive and negative surfactants have
same profile. Surface tension gradient ( S ) corresponds to rubber sheet with variable thickness: thicker regions shrink,
making thinner regions expand (surface gradient is seen in (a)).
5.5. Multi-phase flow solids—no flow inside), and around them in the one con-
tinuous carrying fluid. We should write (N + 1) Navier–Stokes
5.5.1. Microscale description (DNS) equations, and solve them simultaneously. They are coupled
On the ‘microscopic’ level, the flow is fully resolved, in both by sharing the same boundary condition at the particle–fluid
phases: inside N discrete dispersed particles (bubbles, drops; interface. We obtain the flow field in every point of the disper-
850 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 850

sion. The interface forces move the particles and deform the Note that L must keep the meaning of the particle size d.
interfaces. We have the full information about the problem. Dividing mercenarily by ˘ 2 = ˝ , we get dimensionlessly
This fine microscale approach is in the CFD jargon called the

‘direct numerical simulation’ (DNS). ˘ gL3 1 L2 V 2


= C, (5.5.5)
There are no extra forces, as compared with the single- 6 ˘ 2 24 2

phase flow. However, there is a difference: the surface tension

force is moved from the boundary condition (Section 5.4) which is nothing but
directly into the momentum equation (Section 5.2, Eq. (5.2.1)).
3
Usually, this new term contains the following ingredients: Ar = C Re2 . (5.5.6)
surface force [N/m3 ] ∼ interface tension ˛ [N/m], interface cur- 4
vature [m−1 ], interface area [m2 /m3 ]. As for the scaling, this

This is an important criterial equation for problems with


term could be inertially scaled by (˘V2 /L), like any other term in
falling/rising particles, where gravity, buoyancy, and drag play
(5.2.1). This would yield (˛/L)/(˘V2 ), which is nothing but We−1 .
the main role. It indicates a close relation between Re and Ar
The specific way how this surface force is converted into a
in this type of problems, where an equilibrium is established
volumetric force and implemented depends on the modelling
between the driving forces (Ar, l.h.s.) and the resistance force
approach to the DNS. Currently there are several numeri-
(Re, r.h.s.). The former are produced by an external field (e.g.
cal strategies, how to describe interfaces and their motion
gravity), while the latter involves speed as the main ingredient.
(e.g. front tracking method, level set method, volume of fluid
Since the speed in Re is often the part of the solution, the grav-
method).
itational scaling is employed V ∼ (g/L)1/2 , which transforms Re
into Ga = Re2 /Fr. Here, the concept of buoyancy reflected by Ar
5.5.2. Mesoscale description (Euler/Lagrange) has much clearer interpretation than in the single-phase flow:
On the ‘mesoscopic’ level, the flow field is resolved only in the effect of particle–fuid density difference is obvious. Note
the continuous phase, using the single-phase equation—one that both particle size L and speed V are involved in Re(L, V) and
Navier–Stokes (Eulerian view). The dispersed particles (bub- only size in Ar(L3 ). It leads to an iteration procedure in solving
bles, drops, solids) are considered being pointwise, as seen by the sedimentation problem, where the speed depends on the
the fluid. Their motion is described by set of N equations of particle size. A usual trick is to separate these two with help
of a new suitably defined number. The Lyjascenko number,
motion of system of bodies in space (Largangian view). This
Ly(V3 ) = Re3 /Ar = (V3 /g )(˘/ ˘), converts (5.5.6) into
intermediate-resolution approach is in the CFD jargon called
the ‘Euler/Lagrange simulation’ (E/L).
3
The many kinds of hydrodynamic forces acting on the par-
Ar 1/3 = C Ly2/3 , (5.5.7)
ticles are given by various closure formulas, obtained by other 4
means (e.g. experiment, DNS). The forces depend on the local

where L stands on l.h.s. and V2 on r.h.s., so that the size


flow field near the particles. The particles can also affect the
fluid motion, as a feedback (coupling). As compared with the and speed are decoupled. Another problem of course is with
single-phase flow, there are many new equations with new C = C(Re(L, V)).
forces. All of them can be scaled, and new DN will appear. Second, consider an unsteady motion, when the resistance
Consider only the simplest case of a sedimenting particle in a force consumes the initial momentum of a particle:
stagnant unbounded fluid, without (particle → fluid) coupling,
which is the paradigm of sedimentation: du 1 d2 2
m =− ˘f u C (IC : t = 0, u = u0 ) [Hyb/s]. (5.5.8)
dt 2 4

du In the simplest case of the Stokes drag, we have a linear


m = gravity − buoyancy − drag [Hyb/s]. (5.5.1)
dt relaxation process and (5.5.8) becomes

Substituting the typical closures for the forces, it reads du


m = −3 ˝du [Hyb/s] (unsteadiness) = (drag). (5.5.9)
dt

du 1 d2 This is a rare occasion in two-phase flow when we can


m = d3 ˘p g − d3 ˘f g − ˘f u2 C [Hyb/s]. (5.5.2)

dt 6 6 2 4 find the relaxation time (original disturbance is reduced by


1/e ≈ 64%; a stable equilibrium presumed, a node). Choosing
First, consider a steady particle motion, under the force the variables (u, t), their scales (V, T), and the parameters (m,
equilibrium: d, ˝), we get by scaling:

mV
1 d2 = ˝dV [ML/T2 ]. (5.5.10)
gd3 ˘ = ˘u2 C ( ˘ = ˘p − ˘f ) [N] T
6 2 4
(gravity) − (buoyancy) = (drag). (5.5.3)
851 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 851

Choosing the variables (d, u), their scales (L, V), and the By turn, we have the relaxation time:
parameters (˘, ˘, g, C), we get by scaling:
m
Tr = [T]. (5.5.11)
˝d

Substituting for the particle mass m = ( /6)d3 ˘p , the time is

1 L2 ˘ p d2 ˘ p d2
gL3 ˘ = ˘V 2 C [N]. (5.5.4) Tr = ∼ [T]. (5.5.12)

6 2 4 6 ˝ ˝
852 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 852

Another way is to solve the linear problem (5.5.9) directly: stress in the particulate phase, which by the kinetic theory
is ∼˘p V2 , where V ∼ d, where is the velocity gradient and d
3 ˝d the particle size, and the viscous fluid stress ∼˝ . With ∼ V/L,
u(t) = u0 exp − t [m/s]. (5.5.13)
m
and L ∼ d, Ba = ˘p dV/˝, which actually is St. Equivalently, it

can be recasted in terms of the friction forces (particle col-


The exponent should be of the form (t/Tr ), in order, at t = Tr ,
lisional)/(fluid viscous). The presence of the interstitial fluid
the ratio u/u0 be exp(−1) = 1/e. Accordingly, the relaxation time
can be neglected at large Ba, to reach the limit of so-called
is
‘dry’ granular flows.

1 ˘ p d2 ˘ p d2
Tr = ∼ [T]. (5.5.14)
18 ˝ ˝ 5.5.3. Macroscale description (Euler/Euler)
On the ‘macroscopic’ level, the dispersed particles (bubbles,
The variance of (3 ) between (5.5.12) and (5.5.14) is the mod- drops, solids) are considered as a phase smoothly distributed
est price for bypassing the exact solution by the scaling, which in space, forming a ‘pseudo-continuum’. The particles are
is acceptable. then assigned continuous concentration and velocity fields.
The relaxation time just found is useful for a variety of situ- This approximation is acceptable when we describe the sys-
ations, where time scales of different processes are compared, tem on length scales much larger than the discrete scales. The
to distinguish between the ‘fast’ and the ‘slow’, which in turn particles and their spacing must be much smaller than the
leads to the time decoupling, making the problem easier to system size, and than the smallest scales we want to resolve.
handle. One important case is the concept of the Stokes num- In the single-phase flow, estimate the discrete (atomic) scales
ber St. It compares two time scales: particle relaxation time Tr by 10−9 m, and the beginning of the continuum by 10−6 m,
and time of flow change Tf : say. We have three orders of magnitude to bridge the gap.
Accordingly, with 1 cm bubbles, the reactor should be of
Tr 10-m size, to consider the bubbles as the continuous phase.
St = . (5.5.15)
Tf Further, presence of many bubbles is anticipated, for their
spacing be small, e.g. comparable with bubble size. Now, the
The former says how fast the particle relaxes back to the
governing equations should be twice Navier–Stokes: one for
steady state, when accelerated with respect to the surround-
the continuous phase and one for the dispersed phase (two
ing (viscous) fluid. The latter says how fast the flow field
interpenetrating continua, twice Euler’s view). These equa-
changes. Without a priori knowledge, the basic estimate of Tf
tions are coupled via the interphase momentum transfer.
can be made, Tf ∼ L/V. Since the particle must feel the changes
This coarse macroscale approach is in the CFD jargon called
in the fluid, we couple them by the common length scale,
the ‘Euler/Euler simulation’ (E/E).
clearly given by the particle size, L = d. Then, the Stokes num-
The governing equations for single-phase flow are often
ber is
said to be derived from the ‘first principles’. These mechanical
˘p dV principles are known for a single continuum, namely in case
St = . (5.5.16)
˝ of simple fluids, but are only in the process of development
for the multi-phase systems. Here, we lack a universally valid
In other words, St is the ratio (particle inertia)/(fluid iner- equation, which would be of practical use. There are many
tia). More correctly, the particle mass used above should also general equations suggested, but they are too monstrous,
contain the added mass Ca , m = (( /6)d3 ) (˘p + Ca ˘), especially and the many closures needed for them are still missing.
here, when unsteady effects are considered. Then the particle On the other hand, there also are many simple equations,
density becomes: ˘ p → (˘p + Ca ˘). We appreciate it namely in suggested for specific systems and particular flow situations,
case of bubbles in liquids, where ˘p /˘ ≈ 10−3 , so that the bub- which are practical, yet of limited use. As a compromise, here
ble inertia is represented by the added mass, i.e. by the liquid we write the two-phase flow equations purely formally, as two
inertia, ˘p,effective ≈ ˘, since Ca is O(1). Then, the Stokes num- Navier–Stokes-like equations:
ber becomes the Reynolds number, St = ˘p dV/˝ ≈ ˘dV/˝ = Re,
meaning ∼ (particle–joint fluid inertia)/(‘viscous fluid inertia’).
• Continuous phase:
The Stokes number measures the willingness of the car

to get off the road when you turn the stirring wheel sud- ∂ ∗ ∗
denly. As such, it is used in many situations when we want ε ˘ + ∇ (ε∗ ˘∗ v) = 0 (mass) [kg/m3 s], (5.5.17)
∂t
to know how much the dispersed particles tend to follow the
streamlines of the carrying fluid. The total flow-follower has ∂

St = 0 (passive scalar, tracer). On the other hand, particles with ε∗ ˘∗ v + (v.∇ )(ε∗ ˘∗ v) = ∇ ε∗ ∗
+ ε∗ ˘∗ f + S
∂t
large St easily hit the wall in bendings of a duct. It is used 3

(momentum) [Hyb/m s]. (5.5.18)


in devices (impactors) where aerosol particles are sorted out
by their value of St, being expelled from the main stream to
the wall by their inertia, in multiple progressively narrowing • Dispersed phase:
U-bends. The adhesion efficiency of the flotation process also
depends on the value of St, with which a bubble collides with Other numbers appear in specific areas of multi-phase
particles: the lower St, the better for adhesion (no bouncing). flows, at this mesoscale level of description. For instance,
853 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 853

the Bagnold number in granular flows, Ba = ˘ p d2 /˝, compares ∂


ε ∗∗ ˘∗∗ + ∇(ε∗∗ ˘∗∗ u) = 0 (mass) [kg/m 3 s], (5.5.19)
the effect of the interstitial fluid on the motion of the ∂t
granules (grains). In terms of stresses, it is a ratio of the
collisional

ε∗∗ ˘∗∗ u + (u.∇ )(ε∗∗ ˘∗∗ u) = ∇ ε∗∗ ∗∗ + ε∗∗ ˘∗∗ f − S
∂t
(momentum) [Hyb/m3 s]. (5.5.20)
854 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 854

The continuous phase has density ˘∗ , velocity v, stress ten- • Continuous phase:
sor ∗ , volume fraction ε∗ . The dispersed phase has density

˘∗∗ , velocity u, stress tensor ∗∗ , volume fraction ε∗∗ . The mass ∂˘


+ ∇ (˘v) = 0 (mass) [kg/m3 s], (5.5.22)
conservation for the two-phase mixture needs ε∗ + ε∗∗ = 1. The ∂t
main multi-phase problem is to formulate the stress tensors ∂
∗∗ and the interaction term S. As for the scaling, two kinds of ˘v + (v.∇ )(˘v) = −∇ p + ˝∇ 2 v + ˘f
∂t
new DN could appear, as compared with single-phase flow.

One comes from scaling the stress term ∗∗ , which very likely (momentum) [Hyb/m3 s]. (5.5.23)
will not reduce to simple Newtonian formulation yielding Eu
and Re. The other comes from the momentum transfer term
S. The latter would indicate how strong the phase coupling While a single dispersed particle does not affect the flow,
is. a large number of them can exert collective buoyancy effects.
These macroscopic effects consist in fluid density variations
If the dispersed phase constitutes a macroscopic dispersion, caused by distribution of the particle concentration, ˘ = ˘(c).
the particles are the mechanical individuals (small bodies) The interphase coupling is as follows. The (micro) dispersed
and have their own momentum, inertia, gravity, buoyancy, phase imports the fluid velocity v from (5.5.23) into (5.5.21),
whence the momentum equation (5.5.20) applies. If the dis- and exports the concentration c from (5.5.21) into the fluid den-
persed phase constitutes a microscopic dispersion, the particles sity in (5.5.22) and (5.5.23). This is the convection–buoyancy
lack these qualities, and (5.5.20) is needless. The individual two-way coupling.
particles passively follow the flow (St = 0); very fine particles Note that Eqs. (5.5.21)–(5.5.23) formally coincide with those
called ‘passive scalars’ and used for flow visualization (trac- for heat and mass transport considered in Sections 6 and
ers). The true chemical solutions (salt or dye in water) surely 7. Indeed, these equations represent the non-inertial micro-
belong to this category (sub-colloidal). There is a legitimate disperse limit of the governing equations (5.5.17)–(5.5.20)
question: what is between these two extremes, the macro derived for the macro-dispersed multi-phase mixtures. This
and micro. When the dispersed phase earns the right to be simple fact opens an interesting window of research: build-
awarded the full momentum equation? Likely, there are no ing analogies between the well-understood single-phase flows
strict and unequivocal criteria, to decide. Probably, it depends with heat and mass transport, and much less understood
on our choice, what effects and on which scales we wish multi-phase flows. The buoyant coupling from (5.5.21) to
to study. In the molecular dynamics, very small particles (5.5.22)–(5.5.23) can be facilitated by any buoyant agent that
(solute/solvent molecules) are moved by force laws of vari- behaves like a passive scalar (true solute, heat, fine particles,
ous degrees of resolution. Some forces are derived directly etc.). As the next step, the first-order inertial effects can be
from molecular potentials, some are modelled as random added to this base state. For instance, the buoyancy-driven
thermal noise to account for Brownian effects, there is a instabilities in sedimenting layers, fluidized, beds, and bubbly
resistance force, and the overall formulation can be within columns may shear certain common features with phenom-
the Langevine ansatz (random forcing, stochastic differential ena of thermal convection or halinoconvection. As for the
equations). In the Stokesian (hydro)dynamics, very small par- scaling, DN related to heat and mass transport is introduced
ticles (fine particles suspended in fluid) are moved by forces in Sections 6 and 7, together with the coupling to the hydrody-
of both hydrodynamic (Stokes limit) and nonhydrodynamic namics. They naturally apply also to (5.5.21)–(5.5.23). Another
(Brownian, colloidal, interparticle, etc.) origin. Another aspect issue is what is the relevant density of a multi-phase mixture,
is reflected by rheology, where the dispersed phase affects the when evaluating the buoyancy force acting on a submerged
intrinsic momentum transport substantially. body. Usually, one takes either the pure fluid density or the
For microscopic dispersions, Eq. (5.5.20) is omitted and effective mixture density. It seems however, that both can be
(5.5.19) is modified accordingly. The quantity (ε∗∗ ˘∗∗ ) is replaced relevant, depending on the relation between the body size
with the scalar concentration c. The speed u is identified with and the size and spacing of the dispersed particle in the
v. The molecular diffusion term is added to the r.h.s., since the mixture.
macroscopic particles did not have this molecular transport
mechanism.3 Eq. (5.5.19) then becomes: 5.5.4. Retention time distribution
A brief note is in place, on the retention time distribution
(RTD) in equipments. It is the very first thing one must do,
• Dispersed phase:
before starting any kind of thoughts about the processes in
∂c a given apparatus. We assume that the tracer concentration
+ (v.∇ )c = D∇2 c (mass) [kg/m3 s]. (5.5.21) obeys (5.5.21), being one-way coupled with (5.5.22) and (5.5.23).
∂t
Strictly speaking, the diffusion term (D) in (5.5.21) is inconsis-
These modifications must also be reflected by Eqs. (5.5.17) tent with the role of a ‘pure flow follower’. The proper tracer
and (5.5.18). Since the dispersion does not exist on the should stick to a fluid particle, and not to diffuse. Therefore, it
macroscale, the carrying fluid occupies the whole volume, should obey:
ε∗ = 1. Omit the apostrophe at ˘∗ and ∗ . Set S = 0, since the

fluid does not receive momentum from the dispersed par- Dc


= 0 (mass) [kg/m3 s], (5.5.24)
ticles. For Newtonian fluid equations, (5.5.17) and (5.5.18) Dt

3 But there is a concept of ‘hydrodynamic diffusion’ of


become
macroscopic particles, due to (mostly repulsive) interaction
forces, see e.g. Davis (1996).
855 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 855

which is (5.5.21) with D ≡ 0. This can directly be


solved numerically, together with (5.5.22) and (5.5.23), to
get the full information about the flow and
concentration fields. Then, it is easy to monitor
numerically the tracer content at the exit, to create the
RTD response curve. This is not the way we wish to
856 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 856

go in practical applications of RTD. Let us treat the unwanted main flow. Contrasting all these assumptions with the real-
diffusion term in (5.5.21). One cure would be the assumption, ity of the actual flows through real technological systems, it
that the flow is much faster than the tracer diffusion. This is would rather be naive to expect that Pe, whence the single
not usually done, and even if, this is usually not true. Another, scalar quantity Dt (also called: axial dispersion Dax ), can con-
and more practical cure, is to assume that the tracer is simulta- tain the whole truth about the hydrodynamics. We are in need
neously transported by two mechanisms, molecular diffusion of something smarter.
(Dm ) and turbulent dispersion (Dt ). Eq. (5.5.21) can be treated as Actually, there has been an attempt is this direction, and a
follows. Upon the Reynolds decomposition, the actual veloc- kind of Smart RTD (SRTD) has been suggested. Imagine that a
ity field splits into the mean and fluctuating parts, v = V + v∗ . fluid particle has a watch that measures its age ˇ, being set to
The convective term in (5.5.21) becomes (V· )c + (v∗ · )c, keep-
zero at entering the equipment. Thus, ˇ is the retention time
ing the physical meaning of the convective mass flux, j = J + j∗ .
of the fluid particle. At non-relativistic motions, the time on
The mean term, J = (V· )c, is harmless. The fluctuating term is
the watch coincides with the ‘common’ time t on the labora-
modelled as the mass flux driven by the turbulent dispersion,
tory clock. The trivial and seemingly useless physical fact that
j∗ = Dt 2 c. Eq. (5.5.21) thus reads
ˇ = t can be recasted into a useful form. Because the watch is
fixed to the moving fluid particle, its age ˇ must follow the
∂c
+ (V · ∇ )c = (Dm + Dt )∇2c (mass) [kg/m3 s]. (5.5.25) convective derivative:
∂t

Suffice to require that Dm • Dt , which can even be realistic, =1 (retention time) [-], (5.5.31)
Dt
at least in some flow situations. This must, however, very care-
fully be checked in slow flows in microchannels. In case of 1D which is already dimensionless. Here, we assume that the
dominant main flow, we have V = V = Q/S, and the tracer equa- quantity ˇ can be considered to be the field quantity. It is rather
tion is uncoupled from the flow equation, and can be solved counter-intuitive, but any fluid particle located at time t in
independently: place x can be assigned the amount of time ˇ(x, t) it has spent
in the equipment. Thus solving the above equation for ‘conser-
∂c ∂c ∂2 c vation of particle age’, we obtain the real ‘distribution’ of the
+V = Dt (mass) [kg/m3 s]. (5.5.26)
∂t ∂x ∂x2 local retention time within the equipment. Having the field ˇ(x,
t), we can easily calculate the field of concentration, reaction
Fixing the coordinates to the mean flow V deletes the mean rate, conversion, etc. Eq. (5.5.31) can be treated in a similar way
convection term and yields like (5.5.24). It can directly be solved numerically, together with
(5.5.22) and (5.5.23), to get the full information about the flow
∂c ∂2 c and RTD fields. It can also be simplified, to save the computing
= Dt (mass) [kg/m3 s], (5.5.27)
∂t ∂x2
power. After the decomposition, V + v∗ , we have the following
counterpart of (5.5.25):
which is diffusion in stagnant medium. At absence of turbu-
lent fluctuations, the axial dispersion is zero, Dt = 0, and the
∂ˇ
equation + (V · ∇ )ˇ + (v∗ · ∇ )ˇ = 1 (retention time) [-]. (5.5.32)
∂t

∂c
= 0 (mass) [kg/m3 s], (5.5.28) The mean ‘flux of age’, (V· )ˇ, is physically plausible, since
∂t
the stream passing through a given location contains parti-
solves to c = c0 = const. It is the plug flow, where the cross- cles of various age. The fluctuating part, (v∗ · )ˇ, can either
section area marked with c0 is carried through the system with be modelled by a turbulent diffusion term, j∗ = Dts 2 ˇ, or left
the mean speed V. Anticipating the scaling applied to Eq. (7.1), as it is and take a suitable closure for the fluctuating veloc-
(5.5.26) becomes the RDT analogue of Eq. (7.4): ity v∗ . It can be modelled, e.g. by random functions reflecting
truly the local structure of the turbulence. There is wealth of
Fo + Pe = 1. (5.5.29) information about the scaling behaviour of turbulent velocity
fluctuations in various flow situations, in the literature. This
Here, the numbers are defined using the turbulent diffusiv- brings us naturally to the stochastic modelling of RTD. Con-
ity: sidering the 1D dominant main flow, where the mean speed
is a constant, V = V = Q/S, we have the counterpart of (5.5.26):
L2 (unsteadiness)
Fourier number, Fo = ,
Dt T (turbulent mass diffusion) ∂ˇ ∂ˇ
LV (hydrodynamic convection) + (V + v∗ ) =1 (retention time) [-]. (5.5.33)
Peclet number, Pe = . ∂t ∂x
Dt (turbulent mass diffusion)

(5.5.30) In fully 1D case, both V and v∗ are scalars. Fixing the coor-
dinates to the mean flow V deletes the mean convection term
Two ways leads to the exclusivity of the Pe number occur- and yields the following counterpart of (5.5.27):
rence in the RTD problems. First, use the basic time scaling,

T = L/V, and the Fourier number becomes the Peclet number, ∂ˇ ∗ ∂ˇ


+v =1 (retention time) [-], (5.5.34)
Fo = Pe. This makes Eq. (5.5.29): 1 + 1 = 1/Pe. Second, fixing the ∂t ∂x
coordinates to the mean flow V deletes the convection term
857 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 857

(Pe), and by the same scaling Fo becomes Pe. This makes Eq. which can be treated within the framework of the stochas-
(5.5.29): Pe + 0 = 1, which corresponds to (5.5.27). Thus, Pe is tic differential equations, with a suitable random v∗ = f (x, t).
the only important number in RTD, with the 1D dominant At absence of turbulent fluctuations, v∗ = 0, the equation
858 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 858

becomes where the convective speed is recasted into the mass


flow.
∂ˇ
=1 (retention time) [-] (5.5.35) Since we customarily divided (6.3) with the diffusion term,
∂t
both numbers are based on the diffusion rate. If we want other
combinations, we must compose them. These compositions
and solves to ˇ = t. It is the plug flow, where all fluid particles
in the moving coordinate system have age ˇ = t. This age cor- can contain thermal and hydrodynamic quantities to reflect
responds to the position x = Vt from the inlet, measured in the their coupling. One route leads to the Prandtl number:

stationary coordinate. With pipe of length L, the particles exit


(momentum diffusion)
at age ˇ = t = L/V. Prandtl number, Pr = . (6.7)
˚ (heat diffusion)

6. Transport of heat The Prandtl number is prepared as follows. Take the Peclet
number, Pe = LV/˚. Replace the hydrodynamic convection (LV)
For the transport of heat in a moving environment, three bal- with the hydrodynamic diffusion ( ) with help of Re, LV = Re,
ances must be considered together: fluid mass balance (5.1.1), to get Pe = Re/˚. Divide by Re, since it is dimensionless, to get
fluid momentum balance (5.2.1), and the heat balance, which Pe = /˚. Give the product a new name: the Prandtl number,
can be written as Pr = Pe/Re. This number is often used in thermal processes,
where the ratio of two material properties of fluid determines

+ (v · ∇ ) = ˚∇ 2 [K/s] how much the flow is affected by heat diffusion (thermal con-
∂t
vection, boiling, etc.).

(unsteadiness) + (convection) = (diffusion). (6.1) Another route leads to the heat Grashof number, which is
typical for heat-driven buoyancy effects:
Choosing the variables ( , v, x, t), their scales ( , V, L, T),

and the parameters (˚), we get by scaling: ˛ gL 3 (buoyancy force)


Grashof number, Gr =
2 . (6.8)
V ˚ (viscous force)

+ = [ /T]. (6.2) Take the hydrodynamic Archimedes number, Ar = ( ˘/˘)Ga.


T L L2
Replace the general expression for the density difference ˘
Dividing by the temperature scale disappears, and we get with a specific expression for the thermal expansivity of fluid,
˘/˘ = ˛ , where ˛ is the coefficient of thermal expansion.
1 V ˚ Plug it into Ar to get Ar = (˛ )Ga. Give it name the heat
+ = 2 [1/T]. (6.3)
T L L Grashof number, alias the thermal Archimedes number. Gr
is encountered in thermofluid mechanics. It is essentially a
Dividing by the diffusion term (˚/L2 ), because of tradition, hydrodynamic number, where the heat enters as the ‘buoyant
we get agent’, to produce density gradients.
A close derivative of Gr is the thermal Rayleigh number,
L2 LV Ra = Gr Pr:
+ = 1. (6.4)
˚T ˚
˛ gL3 (buoyancy force)
Rayleigh number, Ra = . (6.9)
We can divide by any term, but then the DN would not ˚ (viscous force)
have their usual names. The diffusion term is the most typi-
This number is decisive for the onset of thermal convection
cal one for heat and mass transport phenomena, so it is the
and its evolution via series of bifurcation. It compares the driv-
natural scaling basis. Assigning the DN their proper names
ing thermal disturbance acting on a parcel of fluid, with
we get
rates of transport processes that tend to smear it out (diffusion
of momentum and heat ˚).
Fo + Pe = 1. (6.5)
The way the above composite numbers are ‘derived’ by
making various combinations and replacements seems to be
The following DN arise, having a clear physical meaning in
neither transparent nor free from ambiguity and subjectivity.
terms of process rates:
Actually, they can be obtained correctly, from the correspond-
ing governing equations. For instance, Pr, Gr and Ra appear
L2 (unsteadiness)
Fourier number, Fo = , naturally by scaling the coupled equations for flow and heat
˚T (heat diffusion)
transfer, under a useful approximation (Boussinesq), where
LV (hydrodynamic convection)
Peclet number, Pe = (heat diffusion)
. the only buoyancy-affected density is that at the external
˚ force field term (˘f). Often, the linearization near a uniform
(6.6) base state is considered (stability studies), or the uniform part
of the density field is absorbed in the pressure term, which
When using the basic time scaling, T = L/V, the Fourier the transport by hydrodynamic convection (flow of medium, V)
number becomes the Peclet number, Fo = Pe. The unsteady and the molecular diffusion of heat (heat diffusivity ˚). A mod-
and convective effects are then comparable, as a direct ified version of Pe is the Graetz number, Gr = (mass flow)cp / L,
consequence of our choice of scaling. This may not always cor-
respond to reality. The Peclet number facilitates the coupling
between the hydrodynamics and heat transfer: it compares
859 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 859

gives the buoyancy term proportional to the density difference


˘ = ˘ − ˘0 . Depending on the scales employed, the numbers
appear in different places, as either Pr and Ra, or Pr and Gr.
Note that, based on the physical analogy and scaling argu-
ments, it is possible to introduce Ra also for dispersed layers,
e.g. for bubbly layers in bubbly columns, Ra = g∗ eL3 / mix Dhydro
(g∗ – is the reduced gravity, e –volume fraction of bubbles
(voidage, gas holdup), mix –effective viscosity of bubbly mix-
ture, Dhydro –hydrodynamic diffusivity of bubbles).
860 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 860

Other numbers are also related to buoyancy effects. A Choosing the variables (c, v, r, x, t), their scales (C, V, R, L,
basic quantity is the buoyant (Brunt–Vaisala) frequency ω T), and the parameters (D), we get by scaling:
of the density difference driven oscillator, (d2 z/dt2 ) = ω2 z,

where ω = ((g/˘)(d˘0 /dz))1/2 . It is an elementary prototype for C VC DC


R [C/T]. (7.2)
+ = +
internal gravity waves in stratified environments. Here we T L L2
can encounter the Richardson number, in several variations.
Dividing by the diffusion term (DC/L2 ), because of tradition,
The gradient Richardson number Ri = ω2 /(∂u/∂z)2 . The global
we get
Richardson number is Ri = g∗ L/V2 . Note that it relates to the

Froude number Fr’ = V2 /g∗ L = 1/Ri, when Fr is corrected for buoy- L2 LV RL2

ancy (apostrophe’). Other kinds of Ri also exist. When both + = 1+ . (7.3)


DT D DC
buoyancy (stratification) and rotation are present, there are

numbers indicating their relative effects (stratification param- We can divide by any term, but then the DN would not have
eter, Burger number). The buoyant frequency also follows from their usual names. The diffusion term is the most typical for
the gravitational time scaling T = V/g in (5.2.6). Taking V = L/T, heat and mass transport phenomena, so it is natural to take it
the time scale becomes T = L/gT, which in terms of frequency as the scaling basis. Assigning the DN their proper names we
ω ∼ 1/T reads ω2 = g/L. Employing the reduced gravity g → g∗ to get
account for the density variance, and designating ˘/L the
scale estimate for (d˘0 /dz), we get what was due. Fo + Pe = 1 + Da2 . (7.4)
Boundary conditions in heat transfer are of two kinds. Either
the temperature w is given at the boundary ∂˝ or the heat The following DN arise, having a clear physical meaning in
flux jw through it. The former case is simple to scale, w / . terms of process rates:
The latter is given by
L2 (unsteadiness)
Fourier number, Fo = ,
DT (mass diffusion)
jw = − (∇ )w 2
[J/m s], (6.10)
LV (hydrodynamic convection)
Peclet number, Pe = (mass diffusion)
,
where is the heat conductivity. To calculate the flux, the tem- D
1/2
perature field must be known, which is not always the case. RL2 (reaction)
Damkohler number, Da = .
Therefore, another expression for the flux is introduced, with DC (mass diffusion)
help of the empirical heat transfer coefficient kh :
(7.5)

j w = kh ( )w [J/m2 s], (6.11)

When using the basic time scaling, T = L/V, the Fourier


number becomes the Peclet number, Fo = Pe. The unsteady
where ( )w is the bulk–boundary temperature difference.
and convective effects are then comparable, as a direct con-
Equating (6.10) and (6.11) yields the formula for the
sequence of our choice of scaling. This may not always
coefficient:

correspond to reality. The Peclet number facilitates the cou-


kh = −
(∇ )w [J/m2 s K]. (6.12) pling between the hydrodynamics and mass transfer: it
( )w
compares the transport by hydrodynamic convection (flow of
A simple scaling of (6.12) by ( /L) leads to the Nusselt num- the medium, V) and the molecular diffusion of mass (mass
ber Nu (also Biot number, Bi): diffusivity D).
Since we customarily divided (7.2) by the diffusion term,
kh the numbers are based on the diffusion rate. If we want other
Nu = , (6.13)
/L combinations, we must compose them. These compositions
can contain diffusion and hydrodynamic quantities to reflect
which is nothing but the dimensionless heat transfer coef-
their coupling. One route leads to the Schmidt number:
ficient. The length scale L comes from the near-interface

temperature gradient in (6.10), ( )w ∼ 1/L, and should relate, (momentum diffusion)


e.g. to the thickness of the thermal boundary layer. Although Schmidt number, Sc = . (7.6)
D (mass diffusion)
Nu does not present any intellectual challenge on the grounds

of scaling, it is the most desired quantity in the heat transfer, For the transport of mass of a solute in a moving environment,
since everyone wants to know how much heat passes through three balances must be considered together: fluid mass bal-
the interface, without computing the temperature and flow ance (5.1.1), fluid momentum balance (5.2.1), and the solute
fields. Numerous correlations do exist for Nu in the engineer- mass balance, which can be written as
ing literature.

7. Transport of mass
861 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 861

The Schmidt number is prepared as follows. Take the Peclet enough to affect the flow (e.g. fast-phase changes, boiling, dis-
number, Pe = LV/D. Replace the hydrodynamic convection (LV) tillation, etc.). In turbulence, the viscosity and diffusivity can
with the hydrodynamic diffusion ( ) with help of Re, LV = Re, be not molecular but ‘turbulent’ (eddy viscosity; coefficient of
to get Pe = Re/D. Divide by Re, since it is dimensionless, to get dispersion), hence turbulent Schmidt number.
Pe = /D. Give the product a new name: the Schmidt number, Another route leads to the mass Grashof number, which is
Sc = Pe/Re (also: mass or diffusion Pr). This number is often used typical for mass-driven buoyancy effects:
in mass transfer processes, where the transport rate is large
∂c
+ (v · ∇ )c = D∇2 c + r [kg/m 3 s]
∂t ˇ (buoyancy force)
Grashof number, Gr = . (7.7)
cgL3
2
(unsteadiness) + (convection) = (diffusion) + (reaction). (7.1) (viscous force)
862 chemical
chemicalengineering
engineeringresearch
researchand
anddesign
design 8 86 6 ( (2 20 00 08 8) ) 835–868
835–868 862

Take the hydrodynamic Archimedes number, Ar = ( ˘/˘)Ga. A simple scaling of (7.11) by (D/L) leads to the Sherwood
Replace the general expression for the density difference ˘ number Sh (also Sherman number, Sm):
with a specific expression for the concentration expansivity of
fluid, ˘/˘ = ˇ c, where ˇ is the coefficient of concentrational Sh =
km
, (7.12)
expansion. Plug it into Ar to get Ar = (ˇ c)Ga. Give it name the D/L
mass Grashof number, alias the diffusion Archimedes number.
Gr is encountered in flows with ineligible concentration gradi- which is nothing but the dimensionless mass transfer coef-
ents, which may interfere with the flow (e.g. halinoconvection, ficient. The length scale L comes from the near-interface
concentration gradient in (7.9), ( )w ∼ 1/L, and should relate,
double diffusive convection, thermosolutal convection). It is
e.g. to the thickness of the concentration boundary layer.
essentially a hydrodynamic number, where the mass enters
Although Sh does not present any intellectual challenge on
as the ‘buoyant agent’, to produce density gradients.
the grounds of scaling, it is the most desired quantity in the
A close derivative of Gr is the concentration (salinity, mass)
mass transfer, since everyone wants to know how much mass
Rayleigh number, Ra = Gr Pr:
passes through the interface, without computing the concen-

ˇ (buoyancy force) tration and flow fields. Numerous correlations do exist for Sh
Rayleigh number, Ra = cgL3 . (7.8)
(viscous force) in the engineering literature.
D

The physical picture is similar like with the thermal Ra: 8. Correlations
concentrationally buoyant fluid parcel driven by c moves,
and diffusion of momentum ( ) and mass (D) tend to oppose In the preceding sections, many important DN were intro-
the motion and to weaken the driving force. Instead of ˇ c,
duced and commented, and most of them are listed in Table 2.
more simple choice c/c0 can also be used. Mass Ra is decisive
They can be divided into two classes. First, the basic (primary)
for halinoconvection. Both thermal and mass buoyancy effects
DN that follows directly from scaling the balance equations
are present in double diffusive convection, which, besides oth-
and their BC. Second, the other (secondary) DN that are
ers, has application in oceanology (hot/cold, more/less salty
derived from the basic, or formed by their combinations, or
water) and geology (layering in magna chambers).
created ‘artificially’. The basic DN are the following: momen-
Besides the above coupling between diffusion and hydro-
tum transport (Eu, Fr, Re, Sr) and BC (Bo, Ca, We, Ma); heat
dynamics via Pe, there is also coupling between the diffusion
and reaction via Da. A typical situation occurs in heteroge- transport (Fo, Pe) and BC (Nu); mass transport (Da, Fo, Pe) and BC
neous catalysis, where the diffusion and reaction interplay. (Sh). The basic numbers and their link to their closest relatives
Solving the corresponding equation in case of a model situ- is shown in Fig. 4.
ation (a cylindrical pore in a catalyst, a spherical pellet), we All the numbers obtained by the equation scaling (Section
obtain the concentration profile, whose mean value cm nor- 4) can be reproduced by the dimensional analysis (Section
malized by the bulk concentration c0 is the effectiveness factor 3). However, we must know beforehand the relevant physi-
F = cm /c0 , which also is ∼(mean reaction rate)/(maximum rate). cal quantities that should be grouped. It seems that most of
The model solution for a pore gives F ∼ tanh(Th)/Th, where DN in engineering were first obtained by DA. Here, we prefer
the Thiele number (modulus) is Th = L(k/D)1/2 . Here, L is the to relate them to the equations, to give them better physical
pore length and k the rate constant. Note that for r = kc, the interpretation.
Damkohler number coincides with the Thiele number, Da = Th. Regardless of their origin, DN are used for making corre-
Often, a ‘generalized’ Th is introduced, to retain the last equal- lations. There are two main problems encountered. First, to
ity also for reactions of higher orders. Reaction and mass choose suitable DN. We need a complete list of independent
transfer is combined in the Hatta number, Ha. There also are numbers. Second, to choose suitable characteristic scales to
numbers typical for the reaction kinetics itself. For instance, evaluate the DN. Both the numbers and the scales must be
the Arrhenius number Ah compares the activation energy and relevant for the problem. It is very difficult to choose them
kinetic energy of molecules. correctly without the sound knowledge of the underlying
Boundary conditions in mass transfer are of two kinds. Either processes and the physical meaning of the numbers. Here,
the concentration cw is given at the boundary ∂˝ or the mass the numbers generated by scaling of equations have a great
flux jw through it. The former case is simple to scale, cw /C. The advantage over those produced by dimensional analysis. They
latter is given by contain the correct quantities and have a clear meaning—ratio
of different effects in terms of common physical quantities
jw = − D(∇ c)w [kg/m2 s], (7.9) (force, rate, time, speed, etc.). However, the problem with the
scales still remains. The procedure of equation scaling can give
where D is the (mass) diffusivity. To calculate the flux, the con- certain hints what should the proper scales be, but this is not
centration field must be known, which is not always the case. always sufficient.
Therefore, another expression for the flux is introduced, with Some choices of scales are apparently wrong. For instance,
help of the empirical mass transfer coefficient km : consider the Bond number, Bo = ˘ gL2 /˛, which comes from
the normal component of the free-slip boundary condition.
jw = km ( c)w [kg/m2 s], (7.10) This number is highly relevant for behaviour of bubbles in
liquids. Accordingly, it enters numerous correlations for bub-
where ( c)w is the bulk–boundary concentration difference. ble size and speed, interfacial area, mass transfer coefficient,
Equating (7.9) and (7.10) yields the formula for the coefficient: which are designed for bubble column reactors. Which length
scale L is appropriate? The physics strongly recommends the
D(∇ c)w bubble size. Despite this, many authors have been using the
km = − [m/s]. (7.11) bubble column size, which is apparently wrong. When we look
( c) w at books and review papers on bubble columns, it is easy to find
chemical engineering research and design 8 6 ( 2 0 0 8 ) 835–868 855

can think that the smallest dimension is the most important.


Archimedes number Ar = ( ˘/˘)gL3 / 2 (=( ˘/˘)Ga)
However, taking L = H3 and ignoring the lateral walls at a thin
Bagnold number Ba = ˘ p L2 /˝
layer, simply because A > H, gives a wrong result. The failure is
Biot number Bi = Nu
Bodenstein number Bd = VL/Dax especially evident in stability considerations. The presence of
Bond number* Bo = ( )˘gL2 /˛ (=We/Fr) walls reduces the spectrum of possible wavenumbers (modes)
Capillary number* Ca = ˝V/˛ substantially, and in effect, stabilizes the layer with respect
Cauchy number Ch = ˘V2 /E (=Mc2 ) to the onset of convection. Therefore, the critical value of Ra
Cavitation number Cv = P/˘V2 depends on both H and A. The stability issues are very sen-
Damkohler number* Da = (RL2 /DC)1/2 sitive to the proper choice of scales. For instance, the critical
Dean number Dn = Re(rpipe /rcurv. )1/2
value of Ra is different for rectangular and circular finite con-
Deborah number De = Tr /Tf
tainers, of the same size. Thus, the argument of ‘importance’
Eckert number Ec = V2 /cp
may work, provided that it is physically based: resolve the
Ekman number Ek = /˝L2
Eotvos number Eo = Bo processes occurring along different directions, find their inter-
Euler number* Eu = P/˘V2 relation, and assess their relevance in a given situation. In case
Fourier number* Fo = L2 /˚T (heat) of convection, the buoyant rise is vertical, and the heat and
Fourier number* Fo = L2 /DT (mass) momentum diffusion is horizontal, say, to the first approxi-
Froude number* Fr = V2 /gL mation. It is unlikely, that the same choice of scales applies
Galileo number Ga = gL3 / 2 (=Re2 /Fr) equally well to convection in infinite horizontal layer and in
Grashof number Gr = ˛ gL3 / 2 (heat)
thin vertical slots.
Grashof number Gr = ˇ C gL3 / 2 (mass)
Some choices of scales are even more ambiguous. By select-
Knudsen number Kn = Lmolec /Ldomain
ing the scales, we can select the view of the world. For
Laplace number La = P/(˛/L) (=Eu We)
Lewis number Le = ˚/D (=Sc/Pr) instance, imagine a two-phase flow in a long thin electrolysing
Ljascenko number Ly = (˘/ ˘)(V3 /g ) (=Re3 /Ar) microchannel of size A × B × C = 100 m × 10 mm × 1 m. In
Mach number Mc = V/Vsound these days of the scale-down boom, such an equipment is
Marangoni number* Ma = ˛/˝V not unusual. At the wall, there are electrodes and bubbles
Morton number Mo = g˝4 ˘/˘ 2 ˛ 3 (=We3 /Fr Re4 ) are produced by electrolysis. The spectrum of bubbles sizes is
Nusselt number* Nu = kheat /( /L) quite broad, from few microns, as they are formed, to few cen-
Peclet number* Pe = LV/˚ (heat)
timetres, as they coalesce. The point is to choose the proper
Peclet number* Pe = LV/D (mass)
Prandtl number Pr = Peheat /Re = /˚ length scale and to define Re for making correlation formulas
Rayleigh number Ra = ˛ gL3 / ˚ (heat) (=Gr Pr) designed for the operational quantities (e.g. liquid flow, bub-
Rayleigh number Ra = ˇ C gL3 / D (mass) (=Gr Sc) ble size and concentration, pressure drop, wall shear, etc.). The
Rayleigh number Ra = g∗ eL3 / mix Dhydro (dispersion) choice L = A means the side view at the channel. We see the
Reynolds number* Re = ˘LV/˝ = LV/ flow between two virtually infinite parallel horizontal planes,
Richardson number Ri = ( ˘/˘)(gL/V2 ) (=( ˘/˘)/Fr) 100 m apart. This situation is known as the Poiseuille flow.
Rossby number Ro = V/˝L
The choice L = B means the top view at a segment of the chan-
Schmidt number Sc = /D (=Pemass /Re)
nel. We see the flow between two finite and closely spaced
Sherwood number* Sh = kmass /(L/D)
Stanton number Sn = (kheat /V)(˚/ ) (heat) (=Nu/Peheat ) walls, 1 cm × 1 m in size. This situation is known as the flow
Stanton number Sn = kmass /V (mass) (=Sh/Pemass ) through the Helle–Shaw cell. The choice L = C means the global
Stokes number St = ˘p LV/˝ view at the channel. We see the flow between two finite par-
Strouhal number* Sr = L/TV

Šebestová number Še = 1/Mc allel plates, which are narrow and 1-m long. This situation
Thiele number Th = L(kreac /D)1/2
corresponds to the development of boundary layers in a rect-
Weber number* We = ( )˘LV2 /˛
angular channel. In these cases, different flow profiles along
Basic numbers are marked by an asterisk (*) ( )˘ means ˘ or ˘. different directions are relevant. What length scale should
Note large diversity both in names and notation of dimensionless then be chosen for the flow correlations? One way around
numbers in literature. Those used here are by no means the best or this severe anisometry (1:100:10,000) seems to be the hydraulic
obligatory. radius 2AB/(A + B) = 99 m, which, however, leads to nowhere.
Taking this figure actually means selecting the picture of a
flow through a 99 m dia capillary, which is far from being
that this mistake occurs in a great number of correlation for- related to any possible view at our flow situation. So far, the
mulas published over more than 30 years, some of them even L for a single-phase flow has only been considered. The case
became the classics. These correlations are used for design- with the bubbles is left as a homework for students; enough to
ing factories and plants, and they work well. Imagine how they tease them is to ask for introducing the correct Re for a bubble
would work, if the correlations would be correct. column.
Some choices of scales are ambiguous. We suffer from the In correlations, the composed numbers are often used,
presence of more that one candidate for the length scale. For obtained by combining several simple numbers with clear
instance, consider the Rayleigh number, Ra = ˛ gL3 / ˚ ∼ L3 , physical meaning. The product, however, can have no clear
for a natural convection in a horizontal fluid layer heated from meaning. What can safely be combined? In (5.2.9), we combine
below. In case of infinite layer of height H, the obvious choice two numbers, Fr and Re, generated by the same govern-
is L = H. In case of a layer confined also by two lateral walls ing equation (5.2.1), to compose Ga. It is acceptable, when
separated by distance A, both H and A may be chosen. Thus, they share the same scales. If not, then Ga would, for

there are several possibilities, L3 → H3 , H2 A, HA2 , A3 . Which instance, be Ga = Re2 /Fr = {(˘V∗ 2 /L∗ )/(˝V∗∗ /L∗∗ 2 )} /{(˘V /L )/(˘g)},
2 ∗2 ∗
one is correct? Finally, when the layer is inside a finite con- where we correctly discriminate between the inertial (∗ )
tainer A × B × H, the combinations grow. We can resort to the and viscous (∗∗ ) scales for length and speed. Even with the

S-ar putea să vă placă și