.

ot,l

Succeed at Scale Up

Scale-up

problems in

industrial reactor

mixing can be

costly, but are

all too common.

Here's a proven

procedure that

avoids them.

Douglas E. Leng,
Jal
haif
on

'er
lac
~ion

eeds,

i'

The Dow Chemical Co.

ailure to properly scale up mixing in

batch and continuously stirred ves
sels remains a persistent problem in
the process industries. Numerous
causes of difficulty exist, but a pattern is
evident: A lack of understanding of the pro
cess undermines many efforts. Mixing prob
lems are seldom recognized, in part because
they are not well understood, but also
because they are not examined as quantita
tively as other unit operations.
Correcting errors in scaling mixing
operations is costly and sometimes impos
sible. Errors can cause losses of productivi
ty, quality, and profit, and they also can
lead to safety problems such as reactive
chemical incidents. Although technical
problems cause most failures, nontechnical
reasons also contribute to difficulties.
Scale up of mixing can be easy or com
plex. Most difficulties arise when potential
problems have not been well thought out.
Hardest of all are multiphase processes in
which the chemistry depends on condition
of the phases.
Good quantitative tools and measures of
performance are required to solve such
problems. (See the accompanying article
by Tatterson for an examination of these
needs.) My 25 years of experience with
industrial mixing and scale-up problems
has led me to identify some general trou
blespots and to develop a procedure for
successful scale up.
First of all, the engineer" must clearly
understand the role of the mixing. Is rapid
attainment of uniformity critical to process
success? The following are typical chemi
cal processes that depend on rapid attain
ment of uniformity:
chemical reactors/polymerizers in
which reaction kinetics are equal to or
faster than the ratc of mixing:
competing chemical reactions where
poor mixing affects yields;
crystallizers thai depend on uniform

mixing to promote the growth of large uni

form crystals; and
reactions dependent on mass transport.
such as coalescing and redispersing of liq
uid-liquid and gas-liquid mixtures.
For such processes. desired results can
be achieved more easily in small equip
ment than in large equipment.
Contrast those with applications that are
less sensitive to the needs of uniform mix
ing. These include:
heat transfer:
blending of miscible fluids:
reactors inyolved with slow chemical
reactions;
suspension of solids.
These foul can be considered noncritical
applications, i.e.. they can usually be scaled
up with few difficulties.
Many engineers are most familiar with
the latter, noncritical applications.
Unfortunately. they get the impression that
mixing is simple and can be treated casual
ly. Thus. it is not surprising that so many
scale-up problems occur.

Avoiding problems
For successful scaleup of mixing in
industrial processes, a designer should fol
low six distinct steps:
1. define the process need;
2. identify all of the operational
parameters;
3. review the process history:
4. select the imponant process
parameters;
S. choose an initial equipment design
vessel design, impellers. impeller location.
bal'fles. and points of feed and exit
~treams: and
6. test the design relative to the process
needs and assumptions and then fine tUlle it
to meet the needs of the most important
variables.
Many scale-up failures can he traced
directly to the omission of one or more of

CHEMICAL ENGINEERING PROGRESS

JUNE 1991

23

MIXING

I!1

thesl! six steps. We will consider

each of the steps in turn and provide
ex:ampJes of problems.

Defining the .process need

the impeller 10 distrihute the S,.uper

heated hydrocarhon vapor. The
impeller was supposed. to disperse
the vapor into buhhles to prtwide
mass transfer for the liberation of
ammonia. driving the reaction (0 the
right to form the dimer (2 Al.On
start-up, however, solids accumulat
ed at the base of the vessel. causing
the plant to shut down. The scale up
had been based on invalid criteria,
which led to inadequate mixing for
suspension of the solid.

. The design/~cale-up. process

should start with a thorough exami
nation of all process needs.
Requirements are often straightfor
ward. Sometimes. however. they
may he complex and even contlict
ing: For example. it might be neces
sary to exceed minimum shear to sat- .
isfy dem;)nds for flow and
uniformity.
Vapor to
Typical process needs
Condenser
include:
supplying a uniform slurry
385hp Drive
to feed another processing
step;
facilitating a controlled
chemical reaction;
Feed from
providing for mass or heat
Second
Stage
transfer:
Reactor
dispersing liquid drops for
a suspension polymerization
or extraction;
assuring continuous mixing
of added reactants or
Isopar Vapor
from
monomers.
Petrochem
Avoid the temptation to
Superheater
quickly accept the obvious
without evaluating Jess obvious needs. For instance,
solids must be kept in sus
Identifying all of the
pension. but sometimes that Figure 1. Sparged, lOO,OOO-gallon cofltinuously-stirred
operational parameters
may not be enough, as tank reactor containing 45% solids.
List and then prioritize the
shown in the following
reasons for mixing. These
example.
The major mistake was thinking
may include requirements for mixing
A complex, three-phase, continu
time (blending). micromixing, heat
that adequate suspension of solids
was sufficient to produce desired
transfer, shear. solids suspension.
ous-stirred-tank-reactor system was
being scaled up from test data
production rates. The suspension
mass transfer (including liquid-liq
derived in 3,SOO-gal vessels. The
problem was solved by doubling the
uid, liquid-solid. and gas-liquid 1.
reaction was an equilibrium dimer
power per unit volume, as suggested
introduction of gas into liquid, reac
in tests, but that necessitated trans
tants, chain terminators, initiators.
ization, described by:
ferring the original drive to a stand
2 A (solid) ~ (2 A) (solid) + 2 NH.
and reflux return.
(gas)
.'
by SO,OOO-gal reactor. As the
Textbooks on mixing note that
revised plant got back in production,
The reactor was the last stage of a
these operational parameters do not
tests revealed good solids suspen
scale up equally. Oldshue (I) illus
three-stage train. see Figure I, and
sion was provided in all regions of
contained reacting solids suspended
trates this for scale up from a :::O-gal
the smaller SO.OOO-gal reaclor. The
in a light hydrocarbon. Melt from the
to a 2,SOO-gal \essel: If the power
second stage entered the 100,000-gal
productivity, however, was only
per unit volume (PlY) is held con
about 2S% of the projected rate for
stant, circulation lime increases
stirred reactor and snliditled immedi
that size of vessel. The reaction rate
ately to form a fasl-:-;ettling slurry
threefold, tip speed by 70%, and th\.'
was an important problem that had
containing 4S% solids. Agitation was
impeller Reynolds number by a fac
not heen solved, Further tests also
provided by a single. four-blade, 4So
tor of 8.S. I f processes reqUire both
constant PlY and shear to be main
-pi lched-blade turhine 16 ft in diam
revealed that sparging the superheat
ed hydrm.:arhon was not effective in
lained, condition;. of geomelric simi
eler operated at 26 rpm and driven
removing the ammonia, although
by a 38S-hp motor. A large, bell
larity need to he n:laxeJ.
shaped distributor \\as located under
~H.'ceptahlc reaction rates had heen
In miniplanh. nonsill1ilar geol11\.'

I,~\
I

'

I
1i
1111

achieved by sparging gas into a

3,SOO-gal pilot-plant reactor.
Finally. it was shown 'that condi
tions had been different in the pilot
plant vessel. Sparging was accom
plished by pas~ing nitrogen into the
slurry of suspended solids in miner
ai oil. Nitrogen bubbles had provid
ed the mass transfer needed to
remove the ammonia in the small
reactor, but, in the plant reactor, cal
culations ot heat transfer from the
buobles to the surrounding fluid
indicated tpat all bubbles had
collapsed due to condensa
tion. The liberated heat from
condensation was transported
to the surface by convection.
where it caused the top layer
to boil. Mixing, therefore,
also had to rapidly circulate
reacting solids through the
boiling surface layer to
remove ammonia. Many
mixing designs were pro
posed. several were tested
45' psr Impeller
and pro\'ided improvements,
but none was sufficiently
effective to permit continu
ing to operate the plant.
Fortunately, problems of
Vapor Distributor
such magnitude occur
infrequently.

24

JUNE 1991

CHEMICAL ENGINEERING PROGRESS

II

try is used to simulate mix

2.0()()-gal semiworb plant.
Input Intensity=1.0
if'!!. in a large-scale reaclor:
howe\cr. agglomerates and
), example, when blend
small lumps of polymer
tIme i, critical. miniplant
formed. which necessitated
.;tudies can be done in a
'in-line product filtration.
reactor having a smaller
Oth~r problems al~() appeared
impeller diameterltank diam
but were controllable.
eter (OIT) ratio.
The full-scale plant was an
Identifying all of the key
g,OOO-gal multi-pitch-blade
Intensity=20

parameters is, admittedly, not

turbine installation. which
always easy, but the prob
included an elaborate recycle
lems that can be caused by
system through heaters and
Intensity=5

skimping on the analysis can

coolers arranged in parallel.
be tougher. For instance,
Problems greatly increased at
Intensity=1

incomplete parameter identi

this scale and were unman
fication led to the following
ageable even with the instal
lation of large in-line filters.
problem:
Intensity=O.2

Production of a fine
Calculations showed that
chemical was' being scaled
the energy per unit volume , /
up from a miniplant to a
was 120 hplLOOO gal at the'7
3,500-gal reactor. Agitation
5-L s,cale and decreased
was believed to be needed ill Figure 2. Schematic oj local relative turbulence
almost linearly to 3 hpll,OOO
intensities.
for solids movement equiva
gal in the production reactor.
lent to the just-suspended
Meanwhile. product contin
state. In reality, the reaction rates
ued
to
coarsen
as larger vessels were
Reviewing the process history
depended on several factors, per
used. Since the initiator half-life was
only a fraction of a second, the sizes
haps the most important of which
A process is often scaled up in
incremental stages to meet growing
was mass transfer between the two
of particles were established virtually
product-development needs. This
coalescing liquid phases. The pro
instantaneously when the monomers
iuction facility was sized according
provides an opportunity to observe
entered the reactor.. Intense
,~ reaction rates observed in mini
the development of problems due to
micromixing in the small vessel dis
plant vessels. Production rates
persed the monomers much more
scale up and can help minimize risk.
determined after plant start-up were
Problems can be subtle, as seen in
rapidly as polymerization proceeded.
the following example:
found to be more than an order of
A nonaqueous dispersion was
magnitude less than expected.
being developed from laboratory
At small scale, high shear and
through a mini plant, pilot plant,
rapid circulation led to a dispersion
. D. E. lENG is a senior research .
controlled environment and small
semiworks plant, and, finally, fuIl
scientist ill the Central
drops with a large interfacial area.
scale production. The product was to
Research Engineering
contain a certain fraction of solids in
At the large scale, shear rates were
laboratory of The Dow
lower and circulation was slower,
the form of micron-sized polymer
Chemical Co., Midland, MI
with both factors contributing to
particles. The procedure developed
(517/636-3387; Fax.: 517/638
was to continuously feed initiated
conditions classified as coalescence
9674). He has worked for Dow
monomers into a vigorously agitated
controlling. This led to large drops
at Midland for 35 years.
and much less interfacial area.
hot fluid to rapidly disperse the
During the last 25 years, he
Because the kinetics were controlled
monomers and provide uniformity
has been involved primarily in
and heat transfer. The rate of poly
by mass transfer, the change from
mixing and multiphase
merization was governed by the rate
dispersion dominance to coalescence
research, applications, prob- ,
of the addition of monomer, which
dominance explained the unexpected
lem solving, and engineering
behavior on scale up.
contained free-radical initiators with
administration. He obtained a
The scale-up problem might have
a half-life of less than a second at the
BSc and an MSc from Queen's
temperatures in the reactor.
been anticipated if the agitator had
Univ., Kingston, Ont., and a
1
Superior product was obtained in
been stopped during early laborato
PhD from Purdue Univ. A fel- .{
S-L glass reactors with the desired
ry work so that very rapid coales
low of the AIChE and a mem-j
viscosity, particle sizes, and percent
cence could have been seen. The
ber of the American Chemical l
problem was greatly eased by
age of solids. The results were not
Society, he was the chairman, '.
adding a second impeller, which
quite as good in the next size reac
of the Engineering Foundation
tors, but cerlainly good enough to
):ncreased the overall circulation
Conference on Mixing (1981) .. ';.i
maintain enthusiastic support from
rates and provided an additional
.'
;,',?..;,":
management and customcrs. In the
region ror drop dispersion.
,tii

"-. '1

CHEMICAL ENGINEERING PROGRESS JUNE 1991

25

,.

I
1., ;

J!
I!

11 '

MIXING

In larg('r s..:alc vessels. the

kl.' rthe mass transfer coeffi
mixing intensity was two
cient). it is the interfacial
orders of magnitude less. so
area. a"that is atTected in k1u
the dispcr;.ion of the
measurements, Hen(;c kl.a is
a function of the superfi(;i:l
monomer \\'a~ 'slow relative
to the rat(' of reaction. This
gas velocity and PIV. the
r('sulted in a big difference in
power per unit volume,
InterfaciaI
For example. consider the
the Damkohler number.
Vortex
Clearly. increasing the
scaleup of a microbial fer
total power to the level pro
mentation. The role of mix
vided in laboratory-scale
ing was to provide gas dis ,
work \\'as not a dable solu
persion, supply nutrient
tion. Instead. attention was
blending, keep solids in
Baffles
turned to assessing what
suspension. and maintai'l
could be done to improve
reI ati ve \y low shear. Th:
the mixing intensity at the
broth became rheologically
monomer inlet without
more non-Newtonian with
increasing the overall mix
time. The scale up was to
ing energy. The location of
provide correlations for 80
the monomer inlet was at a
m 3 vessels in which the
Figure 3. Improper geometry for the dispersion of
region of low turbulence,
desired dissolved oxygen
,
monomer ill water.
and local energy-dissIpation
could be maintained by a
rates in agitated vessels are
combination of impeller
attaining uniformity generally do not
known to be highly nonuniform.
speed and gas rate. The criticn
Recently published studies (2,3)
scale up well.
scale-up variable was selected to be
Mixing time was used as a crite
have shown that regions close to
the kLa, so that equivalence in kLa
rion for the scale up of a polymer
the impeller gave dissipation rates
could provide similar dissolved
20 times greater than the average in
ization process in which a block
oxygen levels at corresponding
the reactor as a whole. Figure 2
copolymer was being produced.
stages in the fermentation.
shows schematically that regions
Polymerization rates were fairly
Studies were conducted in 30-L
fast, and locally poor mixing gave
near the impeller provide more
and 2S0-L laboratory vessels. The
intense energy dissipation. Adding
inferior results, leading to a product
kLa measurements were made by the
the monomer through a cooled inlet
with a broad distribution of side
steady-state method: oxygen con
pipe directly to the impeller dis
chain lengths. Calculations indicat
sumption was determined from the
charge stream worked so well at the
ed what mixing times could be
difference between the inlet and out
production scale that filtration was
achieved reasonably in the planned
let oxygen concentrations and the
obviated.
production reactors, and miniplant
airflow rate. Probes measured the
Thus, careful observation and
conditions were chosen that would
dissolved oxygen in the liquid phase.
analysis gave warning that a prob
duplicate these conditions. Tests
The relationship between the two
lem, could occur, Fortunately, the
were then conducted in the mini
sizes of fennenters suggested a cor
solution was fairly simple to imple
plant to determine whether the
relation of the fonn:
k aa.(PAl: 14(S )oAfi(lI)
product quality was sensitive to
ment and worked well.
L
r~'
agitation rates above and below the
where S~ is the superficial gas veloci
ty. It was necessary to correlate kL
calculated rate based on equal mix
Selecting the important pro
ing time. In this case. scale up was
from a determination of the zero
cess parameters
Textbooks generally treat mixing
successful. due in part to the recog
shear viscosity (11) of broth sampled
as a series of topics that have little in
nition that mixing time was a criti
from the fermenter. The correlation
common. These topics can, however.
cal fac.tor for product quality.
proved to be accurate, provided
Ga.~-liquid mixing usually
impeIler flo\\' was sufficient to
lead to an orderly checklist for
relates to phenomena su(;h as gas
selecting parameters that are impor
entrain gas buhbles to the regions
below the impeller,
tant to the mixing process.
holdup, interfacial area (kiP, the
Blending of lIliscihlt! liquids deals
mass-transfer coefficient and inter
Many published studies on
facial area). and the ratio of power
gas-liquid mixing have been done ;11
with topics related to design and
while gassing to power for the
impeller and vessel dimensions, and
low gassing rates. conditions under
which mixing is fairly straightfor
generally ends up with a discussion
ungassed state (P /PIf~)' The latter is
expressed as a ru'netion of the aera
ward. Gas rates. however. arc usually
about mixing time. 8 m , A process is
seldom scaled up oy holding mixing
tion number QIND\ where Q is the
much highcr in industrial mixing
gassing rate, N is the impeller
time constant because this generally
often I to 3 \'olumcs of gas per vol
requires high power commitments.
speed. and D is the impeller diame
ume of liquid pcr minute - and can
ter. Sin(;e mixing has lillie effect on
lead to impclkr cavitation. Wh~'n
Processes that d.::pend on rapidly

II

II

~ !

26

JUNE 1991

CHEMICAL ENGINEERING PROGRESS

that happens. impeller discharge

)lWS diminish and the distribution
,)j" ga~ in the vessel can change dra
matically from fairly uniform to
poor. .-\5 discharge rates decrease.
impeller flow can no longer entrain
gas and carry it to all regions. The
region below the impeller near the
tank bottom is usually the first to
become void of bubbles. Under
flooded conditions. the impeller can
often only distribute gas radially to
regions above it. The ratio PIPIt~
falls to less than 0.5 under these
conditions.
Liqllid-liquid dispersiolls are
commonly associated with drop dis
persion and the generation of inter
facial area. Various mechanisms
have been proposed for drop disper
sion. usually relating fluid forces
(laminar or turbulent shear) to drop
surface forces. The fluid forces can
arise from impact at the walls. or
rotational, dilational or turbulent
shear. Resistance forces are com
monly those due to interfacial ten
sion or surface viscosity. The drop
breaks
when fluid forces exceed
,
cohesi\e forces.
The process of forming the dis
persion is also important. but it has
recehed little attention. Consider
two separated liquid phases. A and B.
of diiferent density with A on top
and B below. To create a dispersion
of B in A, an impeller must be placed
in A. For the reverse. the impeller
must be placed in B. This sounds
simple and logical. but its impor
tance was learned the hard way:
A suspension polymerization was
being scaled up from 3,500-gal to
lO.OnO-gal size. All factors about the
vessel geometry were similar, The
process technology had been estab
lished for many years in the smaller
reactor. Therefore. a prototype
10.OOO-gal vessel was to be tested
first: if it was successful, the remain
ing pl)lymerizers would be installed.
Agitation at both scales was provid
ed by a single. three-blade. retreat

cuneo glussed-steel impeller located

at the bottom of the reactor. The
polymerizers had two finger baffles
placed 1800 apart.
)
The polymerizer had to sequen
tially mix the phm;es (waler with sus
pl'nding agl'nt. and monomers with

catalYSt). form a suspension of

monomer drops in water. heat to
polymerize. polymerize. and finally
cool, The monomers were more
dense and initially formed the bot
tom layer. while water and suspend
ing agent formed the top layer.
Agitation begun, and heat-transfer
coefficients were calculated online
by the latest instrumentation. As the
polymerizer began its first run. the
heat-transfer coefficient feli from 50
to less than 2 Btu/h/ftC/oF. reactor
temperatures went out of control. and
emergency venting to a condenser
was activated to provide evaporative
cooling and avoid serious reactive
chemicals problems.
The monomer layer, instead of the
water layer, had become the continu
ous phase. and heating had initiated
polymerization - resulting in a
mess. Similar results were produced
in two further runs. The problem was
solved by placing a second impeller
in the upper. water layer, which
caused monomer to be thrown up
into that layer. Laboratory mixing
tests showed that the single lower
impeller created a large interfacial
vortex that allowed water from the
upper layer to reach the impeller and
become dispersed, as shown in
Figure 3 , The problem was not
apparent in the 3,500-gal vessel, pos
sibly because relatively lower tan
gential velocities existed, resulting in
a shallower interfacial vortex that did
not reach the impeller.
Coalescence of drops occurs upon
collisions with other drops, or
between drops and solid surfaces

such' as baffles. vessel \\alls. and

impellers. Conditions of shear lead to
drop collisions, High local shear
rutes also leud to drop di'persion. In
a coalescing and dispersing system.
coalescence occurs principally in
regions of low shear. v.hile disper
sion occurs in regions of high shear
near the impeller. An agitated disper
sion is both dynamic and periodic,
Large-scale eddies appear. disappear,
and reappear from time to time. sug
gesting highly transient behavior.
Colliding pairs of drops coalesce if
the forces pressing them together
endure long enough for drainage and
thinning of the continuous phase to
occur between them. If insufficient
drainage occurs. the drop pair will
separate as fluid forces diminish.
Other factors. such as the mobility of
the liquid-liquid interface. the defor
mation of the drops upon impact, and
foreign matter collected at the sur
face. also playa role in the drainage.
Our current ability to scale up liq
uid-liquid dispersions depends on
whether the system is coalescing or
noncoalescing. Coalescing systems
are extremely difficult to scale up if
equal interfacial area or drop-size
distribution must be maintained.
NOllcoalescillg systems often
involve suspension-polymerization
processes employing a water-soluble
protective polymer that collects at
the drop surfaces and prevents coa
lescence. Mixin'g for these systems ~)
need only provide for dispersion and !
suspension because coalescence is I"~
virtually absent. Figure -+ shows that
mean drop sizes become smaller and

suspension
line, drops
are suspended
Log
Drop
Size

.....f - - - - - - _ l _ Dispersion:
drops above

the dispersion
line are
dispersed.
Log Impeller Speed

Fi~llre

-I. Dispersioll-suspensio/l relations/ri/ls ill slispensioll poIYllleri:.atio/l.

CHEMICAL ENGINEERING PROGRESS JUNE 1991

27

MIXING

------.---~~--------------------------~~~---------------------------

drops arc hetter slispenued as

impeller speed" increase. Decreasing
the agilalor spe\.:d uo\.:s the opposite,
The drop sile eorn.:sponciing. to the
inter.scctipn Ill' the dispersion and
suspension line" is the largest mean
drop size that can he produced under
those specific conditions. The aver
age size of drops depends on either
turbulent or laminar shear. while
suspension depends on drop sizes.
density differences. and agitation
parameters (-J.).
Another factor to consider is that
the time to reach a completely dis
persed condition is short in small
equipment but long in large equip
ment. A light-transmission apparatus
for measuring interfacial area was
placed on a I.OOO-gal suspension
polymerizer to measure the time
required for complete dispersion. The
noneoalescing drops were still being
dispersed after 24 h of agitation.

Becaus.e proUllCliol1 schedules

usually do no! allow for allain111ent
of "complete" dispersion. scale up
becomes more complicated. One
ends up comparing completely dis
perscd drops made in sillall equip
ment with partly dispersed drops
produced in large polymerizers.
Therefore. agilation speed has to be
increased in the large vessel to pro
duce drop size:.. equivalent to those
made in the miniplant polymerizer.
Coalescing systems are, by com
parison. complicated to scale up
for a number of reasons. Rates of
chemical reactions often depend on
the interfacial area, which in turn
hinges on two dynamic factors:
coalescence and dispersion. Other
important points for these systems
are as follows:
1. Dispersion is localized near
the impeller in regions of high
shear; practically no dispersion

Table 1. Rules for scaling-up similar vessels under

turbulent conditions
NDX =Constant
Value of X

Rules

Processes

Constant tip speed,

constant torque/volume

Same maximum shear,

simple blending

0.85

Solids suspension

Used in Zweitering
equation for Nis, for
easily suspended solids

0.75

Solids suspension

Scale-up of average solids

suspensions

0.S7

Power/volume

Suspension of fast-settling
slurries, turbulent
dispersion, gas-liquid
operations where kLa's
must be scaled,
refctions requiring
micromixing

0.5

Constant Reynolds
number

Similar heat transfer,

equal viscous/inertial
forces

0.0

Constant speed

Equal mixing time, fast

reactions

..

Note: Using these rules for scale-up requires a Reynolds number greater than 10',
and geometry of similar proportions.

28

JUNE 1991

CHEMICAL ENGINEERING PROGRESS

occurs elsewhere. except initiallv.

2. Coalescence occurs e,ery
where drop colli~ions can OCCllr.
3. Circulation times increase
greatly in large 'e,sels.
The net effect of these three f~\Ctr)r\
is that dispersion dominates in sma! l
scale equipment and coalescem:c
dominates in large-scale. commer
cial equipment.
.
Solid-liquid dispersiolls com
monly involve the suspension of
solids' that are more dense than the
suspending phase. A reliable way to
predict how much agitation is need
ed to suspend solids is provided b:
the Zweitering equation (5;:
i.;N = [5 VO I d ~.: (g/).nlpp4~ X<),I3] I
.

.-DO.X5

"Iiu

where: N., represents revolutions per

second; isa constant that depends
on the selected impeller, its diameter
ratio. and bottom clearance: v is the
~ kinematic viscosit,. m 2/sec; d is the
~' mean particle size, m; D "is tht
impeller diameter. m; X is the per
centage of solids: (gt::.plp) is the
gravitational constant times the den
sity difference di\ided by the liquid
density. mlsec:::'
The just-suspended criterion is
good for dissolving solids. but it is
inadequate if homogeneity is
required; see Table 1 for recommen
dations for homogeneous suspension.
A more difficult. and less studied.
problem is the one of predicting the
N for solids less dense than the sus
pgnding fluid. In this case, the free
surface does not provide efficient
geometry for flow reversal compared
to the curved bottom of a tank. An
effective solution for the case of
floating solids is to place a second
impeller close to the free surface to
help disperse tloating solids. This ini
tiates suspension. and the second
impeller provides needed unifonnity.
An example for !loating solids is
a process to produce foaming-in
place beads. The process involved
adding cylindri,'ai thermoplastic
polymer granules ll1 hOI water con
taining an antil'oagulanl. A foaming
or blowing agent was added. and the
contents were heated to above the
glass-transition temperature. The
granules changed from cylindrical to
spherical particlc~ during the heating

:'1

:I
en'le. Batches were cooled to ambi
tcmperatures. vented to remove
C " , ; flammable blowing agen!.
;lIll. ,11en inspected prior to being
!ran~ferred out of the rolymerizeL
In~pcction~ provided a rough quality
l'heck. Suspension conditions in the
reactor were never ideal. as evi
denced by c~usting on the walls. baf
tle. impeller. and shaft.
On one occasion. conditions at the
end of the cycle were different. and
the beads were stickier than usual.
As the operator began his customary
inspection. he was greeted with a
surge of water, beads. and blowing
agent coming out of the manway. A
spark of unknown origin ignited the
tlammable mixture. and the entire
plant was engulfed in flames.
Fortunately. no one was hun. but the
process was never operated again.
A suspension failure had
occurred. causing massive agglomer
ation to take place. This agglomerat
ed material formed a solid layer
bridging the top surface of the liquid.
as shown in Figure 5. Venting had
only reduced pressure in the head
<-'tee above the crust layer, not
bw it. As the crust layer broke,
pressure from below forced the con
tents out through the opening.
Agitation had failed to provide ade
quate surface movement to assist in
stabilization of the panicles. The
impeller and baffle designs were
inadequate for this application.
Research showed that beads near
the top surface needed to be kept in
motion. to constantly wet and renew
the suspending agen!. When stag
nant conditions existed. water
drainage and agent suspension
occurred and led to interbead adhe
sion. In this prolonged plasticized
condition. the integrity of the beads
was lost. and a continuous polymer
bridge was formed across the sur
face of the contents. The level of liq
uid above the impeller is very criti
cal in marginal cases of floating
suspensions. and it may have been
the cause of the failure. although this
is not certain.
Agitation requirements in crystal
lization processes involve suspend
'g solids in liquids, providing low
lear (so as not to fral..'llIre the crys
tals). and supplying sufficient blend-

the entry point for diluent addition.

If diluenh are intniduccd near the
intake of the impeller. uniformity
can often be reachl.'d in time, ,hort
enou!!h to allow crystal !.!rowth to
domi~ate over nucl~ati(ln~ Because
one of the biggest unknowns is
shear and its effect on cry~tah. vari
able-speed drives are commonly
used for tank crystallizers.

Coagulated
layer of beads

C;ll

V
V

'\

-- ,
'
-,"
...,J

Choosing an initial equipment

design

i
Figure 5. Conditioll of suspellsion
failure caused by poor surface mixing.

ing to minimize concentration and

temperature gradients. Process
needs are specific to the system.
The usual objective in scale up of
tank crystallizers is to produce
large. uniform crystals that can be
easily separated and washed.
Conditions of uniformity are more
difficult to attain in large vessels,
which tend to produce a broader,
finer population of crystals. This
makes subsequent processing more
difficult. If nonsolvents are used {O
"salt out" the crystals. the same
mixing phenomena are encountered.
There is one important difference
the rate of approach to uniformity
can be influenced by the location of

Approximately 75% of all new

installations use contoured or hydro
foil impellers because they produce
axial tlow and develop a more order
ly circulation pattern. create low
shear. and require less power than
traditional impellers. The effects on
flow pattern produced by these
hydrofoils are ,shown in Figure 6.
Most processes require these fea-,
tures. Installations needing high~
shear mixing often couple a high
shear turbine with a hydrofoil
impeller. thereby maintaining good
circulation while providing high
shear/turbulence.
The position of the impellers is
important. in part because the total
volume of liquid may change (usual
ly increase) during many processes.
More than one impeller can be used,
even for normally shaped vessels,
i.e.. where the ratio of tank height to
tank diameter (HIT) is 1.2. In partic
ular, paired hydrofoils offer advan

45 PST

,/

\ /

t, ,j
I

j.

. I

J
FOil/"

u0

Guv

Figllre 6. Dijlerences ill ol'em/lfloll' pallerlls hetweell a ./5_

pitclted-blade turbi/le al/{I a 1t,I'dmflJiI.

CHEMICAL ENGINEERING PROGRESS JUNE 1991

29

,,

I
MIXING

~--------------------~----~~
Big Eddies

tages - their di~charge pallerns

match rather than conflict. as in other
designs. A second impeller is helpful
for solids suspension to provide mix
ing even while draining the contents
of the vessel. The upper impeller of a
dual-impeller systel11 ,hould be posi
tioned at a diswnce of about half an
impeller diameter (Dl from the liquid
surface while introducing solids, and
even for entraining gas into the sys
tem. Normally. the upper turbine is
placed about one impeller diameter
from the surface.
The selection of the ratio of the
impeller diameter to diameter of the
tank (Din is also important. In
industrial applications. ratios from
0.25 to 0.65 are used for common
impellers, while larger ratios are
used for anchors, gates. and helical
types of impellers. If the process
requires flow rather than turbulence,
DfT ratios of 0.4 to 0.6 are chosen.
For micromixing and processes
requiring turbulence. DfT ratios of
0.25 to 0.35 would be selected and
impellers would be operated at high
er speeds than when flow is required.
The trend in solids suspension has
been to use smaller. higher speed
hydrofoil impellers. These produce a'1
coherent jet of liquid that impinges (
on the bottom of the tank. resulting
in effective suspension of solids and
resuspension of settled solids.
Although the Rushton turbine is
commonly referred to in publica
tions, it is rarely used now for new
industrial gas-liquid installations.
An improved design is the concave
disk, which resembles a Rushton but
has six concave, semicylindrical
blades facing the direcliun of move
ment. This provides excellent gas
dispersion. can handle much more
gas before flooding. and lIses less
power than the Rushton turbine. A
new hydrofoil with three wide blades
is also useful. It disperses gas effec
tively and has little tendency to
flood. even at high gas tllnv rales.
The four-blade, 45 c -pilch-blade
turbine is still used for systems of
moderately high viscosity. e.g .. those
up to 40.000 cPo It pnnides more
local shear than the ncw wide-blade
hydrofoil designs: the latll..' 1'. howev
er. supply more flow and a beller
pattern and u~c less energy. At sti 11

J
30

JUNE 1991

Vortex

Figure 7. Typical surface conditiolls: (A) Turbulellt large eddies promote good
surface mixi/lg; (B) Laminar conditiolls with large ce1ltral vortex - good for addi
tion; (C) Flat surface, IlOt recommended for chemical additions.

higher viscosities, particularly where

heat transfer is a problem, special
impellers - including anchor, gate.
nelical. and double-helical types
should be used. In cases where peri
odic sweeping by blades of vessel
walls facilitates heat transfer, these
impellers can be designed to come
very close to the walls (or, in some
cases. to actually scrape the walls).
Some detail is given by Ulbrecht and
Patterson (6).
Fluids up to 3,000-4,000 cP
require some degree of baffling. The "
standard configuration calls for four ~
baffles 90 apart with a width of
Til 0 or Til 2 and located T172 from
the tank wall. This arrangement
works well unless the presence of
solids or non-Newtonian behavior
leads to regions of stagnation.
Baffles placed midway between the
impeller tip and the vessel wall help
to eliminate problems of stagnation.
particularly for thixotropic slurries. ~
Full-length baffles sometimes lead to
problems because they create a rela
tively quiet. vortex-free surface- (
making it more difficult to entrain
solids, liquids, and even gases.
Baffles placed below the surface
allow the fOf-mation of a free vortex. ~
which assists the transfer of materi- :)
als into the vessel.
Careful consideration must be
given to locations or the entries and
exits in stirred vessels. Materials
arc often added directly to the free
surface to avoid using dip pipes.
whit.:h can beconle plugged. Ir the
reaction is fast, the free surface
must be turhulent to adequately
cope with reaction kinetics. To fore

CHEMICAL ENGINEERING PROGRESS

stall such problems. it generally is

preferable to add feed streams close
to the impeller inlet. Feed can some
times be added to the bottom of the
reactor where an impeller is usually
located. In continuous processing.
avoid placing the feed and exit
points near each other because this
leads to bypassing and distorts the
residence-time distribution.
Designs often make use of steady
bearings to help a\'oid mechanical
problems associated with the use of
long, thin shafts. If steady bearings
cannot be tolerated, however. an
alternative is to U$e a shorter. thicker
shaft and place the drive at the bot
tom of the vessel. Using bottom
drives eliminates clutter at the top of
the vessel and allo\\s more options.
Motionless (or in-line) mixers
should be considered as an alterna
tive to stirred vessel designs for cer
tain types of applications. Fast chem
ical reactions often are ideal
applications for motionless mixers.
Also. such mixers excel for applica
tions in which competing reactions
take place and backmixing must be
avoided. In-line mixers are also used
to improve heat transfer to exchang
ers and pipe walls. Simple distribu
tive mixing depends on the length of
the mixer and the number of ele
ments. Dispersive mixing. such as
drop sizing. depends not only on the
number of elements but also the
pressure drop and !low rate, which
rclate to energy dissipation.
An advantage of motionless mix
ing over stirred-\essel mixing is the
elimination of bypassing. All Iluids
pass through regions of uni form

I.

radial and axial mixing. whereas

lr
" in stirred vessels is random.
l~;i,,~'I'l11nOW can lead to a problem
. the process requires the intense
,;[xing found only in regions near
.he impeller. These differences
rn)hubly account for why in-line
mhers invariably scale up success
fulh. unlike stirred vessels, which
are ~)ften fraught with problems.
),lotionless mixers come in a
\'ariety of designs and are usually
made to fit inside standard pipe
,iZe5. Systems can be designed for
~ither laminar or turbulent mixing
applications.
Designs for in-line mixers are
based on the degree of nonmixing in
the feed and the desired degree of
mixing required for the product.
The quality of mixing is usually
expressed as the ratio of the stan
dard deviation (of unmi xedness of
an added species) in the feed to the
average composition. Mixing is
generally considered to be complete
","'hen this ratio is 0.05. This is
equhalent to the "95% mixed" con
di';(m often referred to in stirred
\ )1 technology. The length need
ed for the motionless mixer depends
on the type of mixer selected, the
initial state. and the final uniformity
specified. A comparison of types
can be found in Hamby et af. (5),
along with the design procedure. It
is also important to take pressure
drop into account. Efficient mixers
that appear to require ~port lengths
might also demand an excessive
pressure drop.

Testing the design

It is important to check all of the
process requirements to see how well
each has been met by the proposed
design. The checklist would include
the power to be used. the speed, the
gear box. and the ability to achieve
the required process result. Using
common sense also helps. An
impeller will certainly be inadequate
if it cannot produce sufficient upflow
velocity to overcome the solids set
tling velocity.
An opportunity exists til the time
('. '\ant st~rt-up to me~sure and doc
,
f.:t performance. 01 course, there
often is a ~lrong desire to begin pro
duction as Sl)on as possinle. If oper

aling management can be persuaded

to permit the validation of the
design. however. this will add to the
knowledge base and pay big divi
dends for future operation and
design. Routine testing is done with
water in the vessel. Speed, vibration.
and power draw (using a wattmeter)
are the first things to check.
An inspection of the liquid sur
face tells how effective the baffling
is and how easily materials can be
added to the surface of the fluid.
Figures 7A-C show typical surface
conditions. Large-scale. high-energy
eddies, Figure 7A, promote large
surface waves. A large central vortex
is shown in Figure 7B, while static
conditions are shown in Figure 7C.
Conditions in Figures 7A and Bean
be used effectively for adding reac
tants; faster mixing is achieved by
adding reactants to the impeller
region. Conditions shown in Figure
7C should never be used to add reac
tants to a vessel. A camcorder can
help to record motion, which can
serve as a valuable future reference.
Mixing times can be measured
by a conductivity probe placed
inside the vessel. The probe is con
nected to a high-speed recorder to
monitor changes in conductance
(and, hence, concentration). A trac
er, often salt water, is injected
while the vessel is in operation and
the conductivities are recorded, if
appropriate. Tests can be repeated
at different water levels and agita
tor speeds.
Draining the vessel is important
to see if mechanical problems occur
as the liquid passes through the
plane of the impeller.

geneous liquids). however. can often

be treated via rules. If truly simple
systems are involved in scale up.
consider the rules given in Table I.
which have proven succe\\ful in
our work.
2. Acceptance (~r eXiSlil1!{ condi
tions: Energy is often wasted when
scale up is based on undefined needs.
In the case of a fermentation process.
it was found that reducing the original
mixing energy by 60% did not change
the process performance.
3. Undue secrecy: Honest. com
petent equipment vendors can ,serve
as important resources in helping
you choose the most suitable mixer
design. if they are given adequate
information. Secrecy is important,
but harm can result when only
selected information is provided.
4. Rigidity about flexibilit}:
System demands sometimes are too
broad for anyone design to handle
well. For instance, a single \es.sel
may have to serve as a blend tank. a
polymerizer, a devolatilizer. and
even a crystallizer. It would be
remarkable if any design could serve
each of those requirements equally
well. Therefore, it is essential to
assess the trade-offs between flexi
bility
and
performance.
Occasionally, particularly at small,
remote sites, multi task designs must
be used regardless of consequences.
Mixing is still largely an art. but
it is becoming more scientific. In the
future. quantitative tools such as
computational fluid dynamics
should provide a better link between
8D
design and performance.

Literature Cited.
"

Other pitfalls
Engineers also should guard
against the following common traps
for the unwary in the scale up of
mixing.
1.
Reliance
on
rules:
Preoccupation with scale-up rules
e.g.. "constant tip speed." "constant
power per unit volume," or "constant
torque per unit volume" - usually
means the engineer wanls an easy
answer. Depending on rules often
creates problems because rules trun
cate thinking. Simple mixing (slich
as the blending of nnnrc:li.ting homo

",-,'J

1. Oldshue; J. y.P."Fluid Mixing',

Technology," p.191, .;!,Vt.cGraw-HiIl, Ne~ ~~
York (1983).
.,:,:".J . . '.. .,
'2. Cutter,
A.,
p. 35~

L.

AiChllJ.. 12;

'(1966).
.
.:;1\'"
.; .' iij
.3. Angst, W., J. R~ Bourne, and P.",'
Dell'ava, Chem. Eng:.Sci .. 39
335,~
(1984).
~l~;,~.:.. . .
.
'. :~
4. Leng, D. E., and G. J~. Quarderer, 'i
Chern. Eng. Commun.,.. ~4, p. 177 (982).:;;
S. Hamby, N., M. F; Edwards, and A.;:
W. Nienow, "Mixing in the Proces(~
Industries," p. '226. 306. Bulterworths.~
Woburn. MA (1985). ;i,i
,
.l'~
.~ ,: Ulbrecht. J. J., a~d G. K. Patterson;'~
'. UMixing of Liquids 'by Mechanical ..;
. Agitation." p. 93, Goro,o", & Breach. New::!
_ '~'A
. York (I985)..,.,~" .

p:

CHEMICAL ENGINEERING PROGRESS JUNE 1991

31