Ind. Eng. Chem. Res.

1997, 36, 903-914 903

Bubble Column Reactors for Wastewater Treatment. 3. Pilot-Scale

Solvent Sublation of Pyrene and Pentachlorophenol from Simulated
Wastewater
Jeffrey S. Smith† and Kalliat T. Valsaraj*
Department of Chemical Engineering, Louisiana State University, Baton Rouge, Louisiana 70803

Removal efficiencies from a three-phase continuous, countercurrent solvent sublation of pyrene

and pentachlorophenol from water are presented and compare well with Series CSTR Model
predictions. The effects of the air-to-water flow ratio, the solvent flow rate, and the gas sparger
design are discussed. Steady-state efficiencies as high as 96% for pyrene and 94% for
pentachlorophenol are reported for an annular shear gas sparger. These results demonstrate
the superiority of solvent sublation over bubble fractionation which is shown to be limited to a
removal efficiency of 68% or less for these compounds. Moreover, analyses of the solvent (mineral
oil) phase show that the separation factor (ratio of the solvent effluent concentration to the
water effluent concentration) in solvent sublation can be as high as 300, which is distinctly
greater than that observed in solvent extraction.

Introduction is that a nonvolatile organic solvent is floated upon the

In wastewater treatment plants, aeration devices,
such as diffused aerators and mechanical surface aera-
tors, are often used to facilitate the biological oxidation
of organic compounds. In the case of diffused aeration,
the process closely resembles that of a short, squat
bubble column reactor with multiple air injection points.
Recently (Chern and Yu, 1995), it has been reported that
the air stripping which takes place in these devices is a
serious and growing concern as waste treatment opera-
tors are being charged to further control their volatile
organic compound (VOC) emissions. The authors
stressed the need for improved mass-transfer models
which describe VOC emission from these devices and
introduced a new oxygen mass-transfer model. Al-
though such activities are indeed very good first steps,
the problem of preventing VOC emission from these
devices still remains. Moreover, the current system fails
to remove organic compounds which are resistant to
biological oxidation such as polynuclear aromatic hy-
drocarbons and chlorinated pesticides and insecticides.
Though regulations may not require action now, it is
anticipated that new or improved aeration technology
that reduces VOC emission as well as targets com-
pounds which resist biodegradation will soon be re-
quired. Liquid-liquid extraction is an alternative for
the removal of nonvolatile compounds if an appropriate
solvent is available (Cusack, 1996). Solvent extraction,
however, leaves significant residual solvent in the water
which requires a subsequent separation step. One
potential alternative to conventional solvent extraction
is solvent sublation.
The solvent sublation process, which was introduced
in parts 1 and 2 of this series (Smith et al., 1996a,b), is
a wastewater treatment technology that has the poten-
tial to remove volatile and nonvolatile compounds
without the problem of residual solvent. Like the
diffused aeration device, solvent sublation closely re-
sembles a squat bubble column reactor and is basically
operated in the same way; however, the only difference
* To whom correspondence should be addressed. E-mail:
kvalsar@lsuvm.sncc.lsu.edu. Telephone: (504) 388-6522.
† Separations Expertise Center, Dow Corning Corp. No. 128,
Midland, MI 48686. E-mail: usdccmmj@ibmmail.com.
S0888-5885(96)00524-6 CCC: $14.00
aqueous phase. The advantage of the solvent is that it
serves as a sink for VOCs as well as nonvolatile organic
compounds (NVOCs) which are resistant to biodegrada-
tion. Moreover, in the sublation process, the surface-
active nature of organic compounds is exploited. This
phenomenon has a profound favorable effect on the
removal efficiencies of these compounds, in particular
strongly hydrophobic compounds. The reasons why
efficiencies are higher in sublation than in other separa-
tion processes are because (i) the adsorbed phase and
the bubble wake, which is entrained into the overlying
solvent, experience solvent extraction, (ii) the overall
driving force for mass transfer in the air-water disper-
sion is increased, and (iii) the mechanism of adsorption
to the bubble-water interface circumvents the air-side
mass-transfer resistance which is the dominant resis-
tance for hydrophobic compounds (of low vapor pres-
sures) in conventional air-water contactors.
In this paper, results obtained from a pilot-scale
sublation process used to treat aqueous streams con-
taining pyrene and pentachlorophenol are reported. The
effects of the air-to-water flow ratio, the gas sparger
design, and the Henry enhancement factor on fractional
removal are discussed. The Henry enhancement factor,
which was defined in part 1, is the ratio of the effective
Henry law constant, which arises from the combination
of bulk phase partitioning and surface adsorption, to the
conventional Henry law constant. The results show
that solvent sublation is distinctly superior to bubble
fractionation (diffused aeration). Furthermore, the
Series CSTR Model (SCM) developed in Part 1 is tested
for predicting sublation performance.
Experimental Section
Materials and Methods. The equipment used in
this investigation included a 4 in. i.d. × 60 in. long
(0.1016 × 1.524 m) sublation column, equipped with
both a fine porous glass frit and an annular shear
sparger (0.1 ft2, 6.3 ft/s water jet) described in part 2 of
this series. By using the two different spargers, we were
able to investigate the effect of the different column
hydrodynamics. Pyrene and pentachlorophenol (PCP)
were selected as the target organics to be removed from
the aqueous phase. Pyrene is an example of a poly-
© 1997 American Chemical Society
904 Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997

Table 1. Operating Conditions and Steady-State Fractional Removal Results for the Solvent Sublation and Bubble
Fractionation of Pyrene and Pentachlorophenol
run I.D. compound sparger instr./control QA, SCCM Qw, cm3/min Qo, cm3/min FRss
Sublation
R-73-9-21-94 pyrene porous frit ultrasonic 959 41.8 4 0.809
R-79-9-27-94 pyrene porous frit ultrasonic 959 27.9 4 0.863
R-80-9-29-94 pyrene porous frit ultrasonic 959 41.8 4 0.796
R-87-10-12-94 pyrene porous frit ultrasonic 959 27.9 2 0.862
R-88-10-13-94 pyrene porous frit ultrasonic 2225 27.9 2 0.892
R-89-10-15-94 pyrene porous frit ultrasonic 4413 27.9 2 0.896
R-92-11-1-94 pyrene porous frit ultrasonic 443 27.9 2 0.771
R-94-11-15-94 pyrene porous frit ultrasonic 443 55.8 2 0.684
R-97-11-17-94 pyrene porous frit ultrasonic 1592 13.9 2 0.893
R-99-11-30-94 pyrene porous frit ultrasonic 2225 41.8 2 0.833
R-100-12-1-94 pyrene porous frit ultrasonic 8307 41.8 2 0.901
R-103-12-7-94 pyrene porous frit ultrasonic 959 27.9 2a 0.832
R-106-1-18-95 pyrene annular shear ultrasonic 2225 27.9 2 0.887
R-113-2-9-95 pyrene annular shear ultrasonic 4093 27.9 2 0.945
R-117-3-7-95 pyrene annular shear ultrasonic 5977 27.9 2 0.962
R-120-3-21-95 pyrene annular shear ultrasonic 5355 13.9 2 0.956
R-124-4-6-95 pyrene annular shear ultrasonic 2225 55.8 2 0.882
R-16-6-22-95 PCP (pH 9) annular shear ultrasonic 4732 27.9 2 0.127
R-24-9-20-95 PCP (pH 3) annular shear conductivity 4732 27.9 2 0.853
R-30-9-26-95 PCP (pH 3) annular shear conductivity 7156 27.9 2 0.907
R-33-9-29-95 PCP (pH 3) annular shear conductivity 7156 13.9 2 0.942
R-37-10-3-95 PCP (pH 3) annular shear conductivity 2840 55.8 2 0.744
R-40-10-9-95 PCP (pH 3) annular shear conductivity 2840 41.8 2 0.783
R-43-10-10-95 PCP (pH 3) annular shear conductivity 3454 27.9 2 0.858
Bubble Fractionation
R-131-6-13-95 pyrene annular shear ultrasonic 5977 27.9 no solvent 0.673
R-58-11-27-95 PCP (pH 3) annular shear conductivity 7156 27.9 no solvent 0.682
a R-103 was conducted with “spent solvent”; that is, the solvent used in R-100 was used again.

nuclear aromatic hydrocarbon (PAH) which is a byprod- Table 2. Settings for Chromatographic Equipment
uct from coal gasification, and PCP is an example of a Schimadzu Autoinjector
commercially manufactured insecticide. Both com- injection volume (µL) 25
pounds pose a significant threat to the environment and no. of replicates 3
are classified as priority pollutants by the U.S. Envi- HP 1090L HPLC (Reverse Phase)
ronmental Protection Agency under the Clean Water mobile phase 70-85% MeCN (variable)
Act of 1977. Light-white mineral oil was chosen as the stationary phase Phenomenex Envirosep-PP
solvent phase as it has several advantages over other (3.2 × 125 mm)
organic liquids. For instance, the vapor pressure is low; flow rate (mL/min) 0.5
the density is 0.86 g/cm3; and there is no foul odor. The pressure (bar) 100
retention time (min) 4.5
compounds used in study were obtained from Sigma-
Aldrich, while distilled water and air were supplied by HP 1046A Fluorescence Detector
excitation wavelength (nm) 237
LSU plant utilities. emission wavelength (nm) 385
Twenty-six experiments were conducted which inves- response time (ms) 5 (2000)
tigated the effects of the air, water, and solvent flow scan speed (nm/s) 6 (50)
rates, the gas sparger design, and the Henry enhance- photomultiplier gain 12
ment factor. A summary of the experiments and
operating conditions appears in Table 1. The water fed The detector was connected in series with a Hewlett-
to the process was prepared by spiking containers of Packard 3932A integrator, a Hewlett-Packard 1090L
distilled water (≈19 L each) with the target organic from high-performance liquid chromatograph, and a Shi-
gravimetrically prepared acetonitrile (MeCN) solutions. mazdu autoinjector. The settings for the analytical
For the pyrene experiments, the concentrations of the equipment are listed in Table 2. A three-step procedure
feed stocks ranged between 86 and 130 µg/L, while those was used for preparing aqueous standards for calibra-
for the PCP experiments ranged between 3 and 9 mg/ tion. First, a gravimetric solution of pyrene in MeCN
L, with most being equal to 6 mg/L. For those experi- (3192 mg/L) was prepared; second, 1000 µL of the
ments which investigated the removal of neutral PCP, gravimetric solution was diluted with 100 mL of MeCN
sulfuric acid was added to the feed stocks to adjust the to make a 30 mg/L solution; finally, the samples were
pH between 2.5 and 3.0. Since the pKA of PCP is 4.74 further diluted with distilled water to make solutions
(Montgomery and Welkom, 1990), this ensured that the ranging from 1 to 135 µg/L. The relative standard
PCP remained neutral throughout the experiments. In deviation (RSD) of the fluorescent detector was 13%,
a typical experiment, aqueous influent and effluent which after error propagation corresponds to a 2.5%
samples were taken intermittently and analyzed for the RSD in fractional removal.
organic compound being sublated. Each analysis in- For the analysis of pyrene in mineral oil, a Hewlett-
cluded two replicates. Fractional removal was calcu- Packard 8452 UV/vis spectrophotometer was used. A
lated as FR ) 1 - Aeff/Ainf, where Aeff/Ainf is the ratio of standard solution was prepared gravimetrically by
the average aqueous, effluent concentration to the dissolving 0.5 g of solid pyrene in 1 L of mineral oil. To
average aqueous, influent concentration. calibrate the detector, aliquots of the standard were
A Hewlett-Packard 1046A fluorescent detector was diluted with fresh oil to make solutions ranging from
used to determine the concentration of pyrene in water. 0.5 to 7.40 ppm (wt). The analytical wavelength used

Figure 1. Examples of fractional removal profiles obtained with
(a and b) ultrasonic Level Control and (c) conductivity-based level
control.
for measuring absorbence was 340 nm. The uncertainty
of the measurements (95% C.I.) was 6.48%.
For the analysis of PCP in water, the UV spectropho-
tometer was used to measure absorbence at 214 nm. A
520 mg/L solution of PCP in MeCN was prepared
gravimetrically, and aliquots of this standard were
diluted with distilled water to make solutions ranging
from 0.5 to 10 mg/L. Fifty microliters of sulfuric acid
was added to each standard so that the pH would be
similar to the actual aqueous samples from the subla-
tion column. The RSD for the fractional removal of
PCP was 1.5%.
Control Aspects. As described in part 2 of the
series, the control of the solvent-water interface posi-
tion at the top of the column required the most attention
as the flows to the process were all well controlled with
metering pumps and flow tubes (rotameter). An ultra-
sonic level switch, which measured the time for sound
to traverse the two phases, and a solenoid valve were
used in most of the experiments for controlling the
interface position. This arrangement served reasonably
well. Figure 1a shows a typical fractional removal
profile obtained from the sublation process with the
ultrasonic-based controller. On occasion, however, pro-
cess upsets did occur and were the result of the
ultrasonic switch failing to detect deviations from set-
point. For example, two upsets occurred during run
R-88 as shown in Figure 1b. In both occurrences, the
solvent was displaced with water because the switch
failed to detect that the interface position had increased.
This type of failure was more common than solvent
infiltration because of the large Qw/Qo ratios used in the
experiments. The process upsets were avoided with
simple manual adjustments that included resetting the
Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997 905
ultrasonic switch and adjusting the height of the weir
in the overflow reservoir.
To eliminate the need for human intervention, a
second control scheme, named DAC (data acquisition
and control), was constructed. In the DAC system, the
control action was based upon the electrical-conductivity
profile in the overflow reservoir. As illustrated in
Figure 2, six pairs of nichrome electrodes, encased in
glass tubes, were immersed in the overflow reservoir
to different depths. The probes were arranged to
maximize the sensitivity near the interface. Each pair
of probes was subjected to a 5 kHz ac potential. (High
frequency was necessary as it prevented polarization of
the probes.) A multimeter measured the conductances
across the probes and transmitted the information to a
PC where the interface position was calculated. A
model-based control algorithm then determined the
output to the solenoid valve. One of the unique features
of the controller was that it simulated the valve action
as if it had full dynamic range. (Details of the DAC
system are given in the Appendix, which is available
upon request.) With the DAC system, the control of the
process was improved significantly as deviations about
steady state were within (0.012 fractional removal
points. Figure 1c shows a representative run where
conductivity-based control was used.
Deoxygenation Experiments for the Determina-
tion of KLav. Oxygen-based mass-transfer coefficients
were determined with the sublation column operating
in a semibatch mode (Qo ) 0, Qw ) 0). A YSI Model 55
oxygen meter was used to measure the dissolved oxygen
in the water. In a typical experiment, air, at a known
flow rate, was sparged into the column until the meter
showed that the dissolved oxygen was near 8.6 mg/L,
the solubility at 22 °C, and elevation at 0. (Distilled
water was used.) Once steady state had been main-
tained for several minutes, the feed gas was switched
from instrument air to nitrogen. The switch to nitrogen
occurred quickly and smoothly without any fluctuation
in the gas flow rate. This was possible because the two
regulators controlling the source of air and nitrogen
were fed into one common regulator which fed the
column. During the experiment, the drop in dissolved
oxygen was recorded with time. Plotting the data as
ln Co/C versus time, as eq 1 suggests, resulted in a slope
equal to KLav
Co 3k
ln t ∼ KLavt (1)
C
)
a(1 - )
Results and Discussion
Effect of Operational Variables (QA, Qw, Qo) on
Fractional Removal. The most important variable
affecting the sublation of organic compounds from water
is the air-to-water flow ratio, QA/Qw. This ratio not only
influences the extent of stripping but also contributes
greatly to column hydrodynamic properties. In part 2,
it was argued that the homogeneous flow regime is the
appropriate operating regime for the sublation process.
Thus, the values of QA and Qw shown in Table 1 were
chosen to ensure homogeneous flow. In Figure 3,
fractional removal results obtained from the sublation
process are plotted versus QA/Qw. Each point represents
a 15-30 h steady-state average. The general trend is
that, as the flow ratio is increased, the fractional
removal rapidly approaches an upper limit. This be-
havior is consistent with SCM (Series CSTR Model)
predictions presented in part 1, which showed a very
906 Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997
Figure 2. Diagram of conductivity probes and data acquisition and control (DAC) system used to control the solvent-water interface.
Figure 3. Fractional removal of pyrene and neutral pentachlo-
rophenol (PCP) versus the air-to-water flow ratio, QA/Qw.
steep gradient between 0 e QA/Qw e 50 and a plateau
at QA/Qw g 150. In fact, the data in the figure suggest
the optimum operating condition is in the range of 200
e QA/Qw e 300, with the exact value depending on the
compound and sparger type. Defining the optimum
range is based on the observation that the gains made
in fractional removal are very small when QA/Qw is
increased beyond 300. Another reason is that ulti-
mately, as the gas velocity is increased, an unfavorable
transition occurs from homogeneous flow to heteroge-
neous flow. The turbulence associated with heteroge-
neous flow is undesirable, as it tends to disperse the
solvent throughout the system.
Sublation experiments were performed to examine the
relative effect of the solvent flow rate versus the water
flow rate. First, runs R-73 and R-79, which appear at
the top of Table 1, were conducted to demonstrate the
effect of a -33% step change in water flow rate while
holding the solvent flow rate constant at 4 cm3/min. As
one can see from the table, the fractional removal of
pyrene increased from approximately 80% to 86% in
response to decreasing Qw from 41.8 to 27.9 cm3/min.
Given the strong influence of the air-to-water flow ratio
observed in Figure 3, this was not surprising. When
the water flow rate was returned to its original value
(run R-80), the fractional removal returned to 80%. A
second experiment, run R-87, was performed which
investigated the effect of reducing the solvent flow rate
from 4 to 2 cm3/min while holding the water flow rate
constant at 27.9 cm3/min. As shown in Table 1, for runs
R-79 and R-87, there was no effect of reducing the
solvent flow rate on the fractional removal. The reason
for this is that the mineral-oil-partition coefficient (Kow)
for pyrene is very large. In fact, since most hydrophobic
compounds possess very large partition coefficients, the
amount of solvent needed for sublation is only that
which ensures a steady state. This was shown in part
1. Thus, the solvent flow rate used in most of the
experiments was maintained at the lowest pump set-
ting, 2 cm3/min.
Because of the large capacity of the solvent, one
variation of the process is to recycle the solvent. This
is referred to as using “spent” solvent and is a way of
minimizing the overall solvent usage. Therefore, to
further show the weak dependence of the solvent, run
R-103 was conducted using a feed solvent (oil) having a
pyrene concentration of 1.57 mg/L; however, the operat-
ing conditions of R-103 were identical to those used in
R-87. The effect of the spent solvent was to reduce the
fractional removal by only 0.03. Since this difference
was outside the statistical error limits ((0.02), the effect
was statistically significant but, nonetheless, very weak.
Effect of Gas Sparger on Mass Transfer. As
illustrated in Figure 3, sublation experiments involving
pyrene were conducted which investigated the effect of
gas sparger type. For QA/Qw < 150, there was no
significant difference between the fractional removal
data collected from the shear sparger and those collected
from the porous frit. However, for QA/Qw g 150, the
fractional removal of pyrene was improved by as much
as 5%, suggesting that the benefit of the shear sparger
is only realized at large QA/Qw ratios. The reason for
this behavior may be explained by considering the

unique hydrodynamic properties of the shear sparger.
It was observed in part 2, that, for gas velocities greater
than about 1 cm/s, the processes of bubble coalescence
and breakage were buffered by the combined actions of
the water jet and the milky-white plume of microbubbles
exiting the shear sparger. It was speculated that this
so-called “buffering effect” was responsible for the weak
bubble size dependence on gas velocity. As it turns out,
in the cases where the shear sparger performed better
than the frit, the gas velocities varied from 0.84 to 1.23
cm/s. For the low QA/Qw experiments where the re-
moval efficiencies were about the same for both sparg-
ers, the gas velocities were less than or equal to 0.46
cm/s. To investigate whether or not the improved
performance was just the result of operating at the
higher gas velocities, run R-100 was conducted using
the porous frit sparger and a gas velocity of 1.71 cm/s
(QA/Qw ) 199). The result of this experiment was no
different than those observed at 0.33 (QA/Qw ) 115) and
0.91 cm/s (QA/Qw ) 158) for the porous frit. Thus, one
may reasonably conclude that the shear sparger does
have an advantage over the frit for superficial gas
velocities g 1 cm/s and that the effect is not simply the
result of gas velocity. The advantage is attributed to
improved hydrodynamic properties of the shear sparger
and how they affect the air-water mass-transfer coef-
ficient. The following discussion elucidates this rela-
tionship.
Consider eq 2, which is the classical two-resistance
model for the overall liquid-side mass-transfer coef-
ficient. For many organic compounds such as VOCs,
1 1 1 phase penetration model describing the overall liquid-
(2)
k κl Hcκg
) +
air-side resistance is unimportant and the term 1/Hcκg
is neglected. However, for strongly hydrophobic com-
pounds such as pyrene and pentachlorophenol, air-side
resistance is important, if not the dominant resistance,
because of the very low Henry law constants, Hc, that
they possess. Thus, the common practice of scaling
oxygen mass-transfer coefficients (which are liquid-side-
controlled) by the Schmidt number is not correct for
these compounds. Therefore, other means must be used
to determine k.
One common approach for predicting the mass-
transfer coefficient for gas-liquid dispersions is to use
a penetration theory such as the Higbie penetration
model (Taylor and Krishna, 1993). In this model the
average mass-transfer coefficient is defined as
x x
4DAB 4DAB
κl ) ) (3)
πte λ fluid (eddy) in a bubble column is the Kolmogoroff
π
ω mixing length. Baird and Rice (1975) and Kantak et
where DAB is the binary diffusivity and te is the exposure
time. The exposure time may be considered as the time
a small fluid element of uniform concentration comes
in contact with a phase boundary through which mass
transport occurs. Like film theory, where the film
thickness is difficult to determine, the exposure time is
also difficult to determine; however, in some cases, it
can be calculated. As suggested in eq 3, the exposure
time may be taken as the ratio of some characteristic
length, λ, which the fluid element must traverse to some
characteristic velocity, ω. For example, in the case of
noninteracting bubbles rising up through a liquid (string
bubbling flow regime), the exposure time is the mean
Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997 907
bubble diameter, ds, divided by the rise velocity, U
(Taylor and Krishna, 1993). This is because the liquid
in immediate contact with a gas bubble is replaced in a
time equal to that for the mean bubble to rise one
diameter. However, it is implied that the resistance to
mass transfer resides totally in the liquid phase, that
is, the phase which experiences the penetration. Thus,
eq 3 is valid for single-phase resistance only. Addition-
ally, it is implied that the bubbles rise in pure plug flow
since the effects of backmixing or circulation are ig-
nored. In order to generalize, the key questions which
must be addressed are as follows: How does one
combine gas-phase resistance into eq 3? How does one
determine the exposure time when the flow regime is
different from plug flow?
Sherwood et al. (1975) derived that the total moles
transferred during an exposure period for two resis-
tances in series may be expressed as
{x ( ( )
4DABte DAB κi2te
∫0t N dt )
e
π
+
κi
exp
DAB
×
( x ) )}( )
te Cg
erfc κi -1 Cl - (4)
DAB Hc
where κi is the resistance at the interface due to a
surfactant. In the derivation of eq 4, the authors
assumed that the resistance 1/κi may be added to the
liquid-side resistance in a fashion similar to the way
gas-side resistance was added to eq 2. Thus, eq 4 may
be used as a starting point to generate a general two-
side mass-transfer coefficient. In the spirit of the
conventional film theory, we assume that the interface
itself offers no resistance but that the gas phase offers
significant resistance. Substituting 1/Hcκg for 1/κi and
after some rearrangement, one can show
x
4DAB
k πte exp(z2) erfc(z) - 1
(5)
Hcκg Hcκg
) +
z2
where z is defined as Hcκg(te/DAB)0.5, and Hc is the bulk
phase Henry law constant. When z is approximately
greater than or equal to 1, the “correction function”, f(z),
becomes effectively zero and eq 5 reduces to eq 3;
however, if z is <1, then the effect of air-side resistance
is significant.
The argument, z, depends upon the characteristic
length and velocity. It is proposed that the character-
istic length describing the exposure time of a packet of
al. (1994) have suggested expressions describing the
degree of mixing, or axial dispersion, in bubble columns
in terms of mixing lengths. In the homogeneous flow
regime, these lengths may range from the bubble
diameter to as large as the column height. In part 2 of
this series, mixing lengths, referred to as dispersivities,
were determined for the porous frit and the annular
shear sparger. A simple expression for the dispersion
coefficient suggested by Kantak et al. (1994) was used,
D ) λug. In the case of the porous frit, the dispersivity,
based on the superficial gas velocity, was on the order
of the column height, L. However, for the shear sparger,
the dispersivity was based on the interstitial gas velocity
and was found to be on the order of the column
908 Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997
Figure 4. Oxygen-based mass-transfer coefficient determined
from the deoxygenation of water versus ug.
diameter, Dc. The reason why the dispersion coefficient
for the shear sparger scaled with the interstitial velocity
instead of the superficial velocity was not fully under-
stood; however, it was observed that the circulatory
motion set up by the water jet tended to buffer the
processes of coalescence and breakage, which ultimately
led to a weak bubble size dependence on the gas velocity.
This weak dependence may suggest that the energy
dissipated for mixing, coalescence, and breakage is due
more to the energy input by the water jet than that of
the gas. Kantak et al. (1994) reported that the inter-
stitial gas velocity in the definition of the axial disper-
sion coefficient makes it independent of the operating
parameters and may explain why the dispersion coef-
ficient scaled much better with the interstitial gas
velocity.
The characteristic velocity is chosen as the superficial
gas velocity, ug. The choice is largely based on the
commonly observed fact that overall volumetric mass-
transfer coefficients, KLav, scale well with ug (Shah et
al., 1982). Typical power dependencies on gas velocity
range from 0.6 to 1.7. Data collected from deoxygen-
tation experiments conducted by our group have shown
dependencies similar to those illustrated in Figure 4.
When eq 3 is combined with correlations for gas holdup
and bubble size obtained from the porous frit and the
shear sparger (part 2 of series), it can be shown that
the volumetric mass-transfer coefficients have 1.1 and
1.3 overall power dependencies in ug, respectively.
Thus, the choice of ug as the characteristic velocity
appears to be appropriate.
Having defined the exposure time as the character-
istic mixing length (dispersivity) divided by ug for mass
bubbling in bubble columns, the criteria for neglecting
air-side resistance in eq 5 may be written in terms of
the Sherwood and Peclet numbers. Equation 6 defines
x
DAB
Hcκg g
te
DAB
(Hcκg) g 2
λ/ug
(λHcκg) 2 2
λ DAB
g
DAB2 DAB2(λ/ug)
Shλ g Peλ1/2 (6)
the criteria for which air-side resistance may be ne-
glected in terms of the relative degree of backmixing or
circulation, Peλ. Given that Shλ is dependent upon the
Figure 5. Argument of the Higbie correction function versus gas
velocity.
bulk phase Henry law constant, meeting the criteria for
strongly hydrophobic compounds requires extreme con-
ditions. Figure 5 shows the argument of the correction
function, f(z), versus the superficial gas velocity for
pyrene and PCP. As one can see in the figure for the
cases where the conventional Henry law constant is
used to determine Shλ, eliminating air-side resistance
requires extremely low gas velocities. In this regime
string bubbling flow prevails, and the example cited
earlier applies. However, operating at such low veloci-
ties is impractical, as the overall removal efficiency
under these conditions is very small. The situation
where the conventional Henry law is used applies to
processes where the principal driving force for mass
transfer is bulk phase partitioning. However, in the
solvent sublation process, the material adsorbed to the
bubble-water interface is removed and recovered in the
overlying solvent layer. Thus, unlike bubble fraction-
ation where the adsorbed material is not removed from
the water (i.e., no sink), the driving force for mass
transfer in sublation is enhanced. The effective Henry
law constant, H, which combines bulk phase partition-
ing and surface adsorption (Hc + 3KA/a) is then used in
the determination of Shλ. As shown in Figure 5, the
criteria for neglecting the correction function in eq 5 are
met for solvent sublation; however, the appropriate
mixing length must still be used for calculating k. It is
emphasized that Figure 5 applies only to the homoge-
neous flow regime as the nature of mixing changes
greatly as the flow becomes heterogeneous.
Inserting the dispersivities obtained from the porous
frit (L) and the shear sparger (Dc) into eq 3, one may
estimate the relative effectiveness of the shear sparger
over the porous frit
x x
4DAB πtefrit tefrit
x
kshear L
) x15 )
kfrit πteshear 4DAB teshear Dc
) ) )
3.87 (7)
In terms of the volumetric mass-transfer coefficients,
this ratio becomes
3
(kav)shear
) 3.87
( a) shear
) 3.87 × 0.503ug0.15 (8)
(kav)frit
(3a )
frit
where the gas velocity power laws determined in part
2 have been substituted for the hydrodynamic proper-

ties. The average ratio is then determined by integrat-
ing eq 8 over the gas velocity limits defining the
homogeneous flow regime,
( ) ∫0.14.0cm/s
cm/s
(kav)shear 1.95ug0.15 dug
) 2.1
∫0.14.0cm/s
) cm/s
(kav)frit avg dug
Interestingly, this result compares very well with
experimental data (Burns, 1995) collected from the 5.5
in. diameter column described in part 2. The data
indicated that mass-transfer coefficients obtained from
a shear sparger (porous surface area ) 0.05 ft2) are
approximately 2 times greater than those obtained from
a flexible rubber membrane (Flexisparger).
Effect of the Henry Enhancement Factor (Sur-
face Activity). The effectiveness of the sublation
mechanism is largely based upon the surface activity
of the organic compound. This activity can be quantified
in terms of the Henry enhancement factor, H. For
example, based on the thermodynamic properties, Hc
and KA, cited in part 1 for pyrene and neutral PCP, the
enhancement factor can be calculated as a function of
the bubble radius, a. In this study, the enhancement
factors for both compounds are significantly greater
than 1. However, the enhancement factor for pyrene
is consistently predicted to be 46% greater than that
for PCP for the range of bubble radii, 0.25-2.5 mm. This
implies not only a greater driving force for the mass
transfer of pyrene but also a greater removal efficiency.
As illustrated in Figure 4, the fractional removal of
pyrene was greater than that for PCP for all cases. Even
the pyrene data collected from the porous frit showed
better performance than the PCP data collected from
the shear sparger.
To further illustrate the influence of surface activity,
run R-16 was conducted in which ionic PCP was
sublated. As shown in Table 1, the pH of the water was
maintained at 9. The fractional removal for the run was
0.127, which was significantly less than the 0.853 value
obtained from run R-24 which was conducted under
similar operating conditions but with a pH of 3. The
reason for this dependence is that the pH affects the
available amount of the neutral PCP which is suscep-
tible to the sublation mechanism. Because the pKA of
PCP is 4.74, the degree of dissociation of PCP at pH )
9 is nearly 100%. Consequently, a very low removal
efficiency was observed for the run; however, under
acidic conditions (R-24), the removal efficiency was
significantly larger because essentially all of the PCP
was surface active.
Surface adsorption at an interface is certainly not
limited to just solvent sublation. In any interface-mass
transfer process, surface adsorption may play a role in
the overall transport. However, whether or not the
effect of the enhancement factor is exploited is a
completely different issue. For example, in most con-
ventional separation devices such as absorption, strip-
ping, and extraction, the phases which are used to
convect the species of interest are only the bulk phases.
That is, the interface phase, which is present in each of
the devices, is not convected in or out. As such, the
material which adsorbs to the interface remains in the
device. To support this argument, bubble fractionation
runs, R-131 and R-58, were conducted which investi-
gated the removal in the absence of mineral oil. In
Figure 6, fractional removal results of these experiments
are compared to those obtained from corresponding
Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997 909
(9)
Figure 6. Effect of the Henry enhancement factor on fractional
removal in solvent sublation and bubble fractionation.
sublation experiments, R-117 and R-30. For the bubble
fractionation runs, the removal efficiencies for both
neutral PCP and pyrene were about 68% or less. For
the sublation experiments, however, the removal ef-
ficiencies of PCP and pyrene were significantly greater,
90% and 96%, respectively. Clearly, these data are
evidence that solvent sublation has a distinct advantage
over bubble fractionation (or diffused aeration). More
importantly, the data in Figure 6 also confirm the notion
that most conventional processes do not exploit the
interface phase. As shown in the figure, the enhance-
ment factors for PCP and pyrene under the conditions
specified were 77 and 117, respectively; and for the
sublation experiments, these were good indicators of the
relative removal efficiencies. However, in the case of
the bubble fractionation experiments, these enhance-
ment factors gave no indication of the relative perfor-
mance. The reason for this is that in these experiments
the sorbate at the interface phase remained in the
column.
Model Predictions. Predicting sublation perfor-
mance is a challenging problem given the large number
of parameters and the relative dependencies they have
on one another (Smith et al., 1996a). Many parameters,
however, are easily obtained such as the column dimen-
sions, sparger type, fluid properties (e.g., water viscosity,
density, surface tension), and operating conditions,
while others are obtained through simple calculations
and experiments. The values of the parameters used
in this study are listed in Table 3. The thermodynamic
properties, Hc and KA, appearing in the table were
obtained from Montgomery and Welkom (1990) and Hoff
et al. (1993), respectively. The Kow for neutral PCP was
determined experimentally by extracting three samples
of an aqueous solution (8 mg/L of PCP) with different
amounts of mineral oil and measuring the extraction
efficiencies. The oil-to-water volume ratios used were
0.018, 0.030, and 0.050. Plotting the data as FR/(1 -
FR) versus the oil-to-water volume ratio yielded a slope
equal to Kow (Kow of PCP ) 732). The Kow for pyrene
was estimated as the ratio of its oil solubility to its water
solubility. Equation 10 (Prausnitz et al., 1986) was used
to estimate the solubility of pyrene in mineral oil. Table
4 summarizes the values used in the calculation of the
solubility as well as the value of Kow which results.
[ ( )]
Tmelt
1 ∆Hfus
x∼ exp 1- (10)
γ RTmelt T
910 Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997

Table 3. Model Parameters Used in the Series CSTR The procedure for determining di was to adjust it until
Model (SCM) the fractional removal as predicted by the SCM matched
pyrene pyrene shear neutral PCP the experimental fractional removal to at least three
porous frit sparger shear sparger decimal places. In Table 5, the column labeled “exact
Thermodynamic di” contains those values which gave the best agreement
Hc 7.65 × 10-4 7.65 × 10-4 1.39 × 10-4 between theory and experiment. Interestingly, all of the
KA (cm) 5.39 × 10-3 5.39 × 10-3 6.70 × 10-4 adjusted values are nearly of the same order of magni-
Kow 76412 76412 732 tude, ≈O(10-3) cm, regardless of compound type and
Design sparger type. This observation is a good indication that
Dc (cm) 10.16 10.16 10.16 the transport mechanisms in sublation have been ac-
L (cm) 152.4 152.4 152.4 counted for appropriately. On closer inspection, how-
Hydrodynamic ever, the corresponding wake-to-bubble volume ratios,
a (cm) 0.122ug0.573 0.174ug0.24 0.174ug0.24 Vw/Vb ≈ 3di/a, are only on the order of ≈O(10-2). Fan
 0.0491ug1.21 a 0.0352ug1.031 0.0352ug1.031 and Tsuchyia (1990) have compiled wake-to-bubble
λ (cm) 152.4 10.16 10.16
volume data from numerous investigators which show
Kinetic that, for spherically-capped bubbles with effective di-
no. of CSTR stages n ) 1 (i ) 2) n ) 1 (i ) 2) n ) 1 (i ) 2)
ameters of 4-20 mm, Vw/Vb is on the order of 5. The
DAB (cm2/s) 6.70 × 10-6 6.70 × 10-6 7.19 × 10-6
bubbles observed in the sublation experiments, however,
x x x
k, Higbie model 4DAB 4DAB 4DAB
(cm/s) were much more spherically-shaped and smaller. Thus,
πL/ug πDc/ug πDc/ug the wakes associated with these bubbles should be very
kl, film theory 10-4 10-4 10-4 small and most likely are the results of entirely different
(cm/s) physics than those discussed by Fan and Tsuchyia.
a The power-law model here varies slightly from that cited in Based on the relative magnitude of di/a, it is believed
part 2 as it is based only on the data obtained from the porous that di is related to the viscous boundary layer that
frit and not the flexible rubber sparger. develops around the bubble.
In the development of the velocity distribution for a
Table 4. Values Used To Calculate the Solubility of rising gas bubble in a liquid, Levich (1962) found that
Pyrene in Mineral Oil and the Oil-Water Partition
Coefficient the boundary layer thickness may be estimated from
the bubble radius, the kinematic viscosity, and the rise
T (K) 298 velocity, U, as shown in eq 11. (The same result is
Tmelt (K) 424
∆Hfus (estimated) (J/g) 127.3
µa 1/2
γ pyrene in oil (UNIFAC)
water solubility (mol frac)
3.4
1.20 × 10-8
di ∼ (FU ) (11)
oil solubility (mol frac) 0.0134
Kow (lt/lt) 76412 predicted by Boundary Layer Theory (Bird et al., 1960).)
Thus, substituting typical values into the equation, one
The liquid diffusivities, DAB, were estimated from the can estimate the boundary layer around a rising bubble,

( )
Wilke-Chang correlation (Geankoplis, 1993), and the
hydrodynamic correlations shown in the table are based 0.01 g/cm‚s × 0.1 cm 1/2
di ∼ ) 5.8 × 10-3 cm (12)
on the data obtained in part 2. Of all the SCM 1 g/cm3 × 30 cm/s
parameters, the amount of wake (water), di, which is
entrained into the solvent is the most difficult to Interestingly, this prediction of di is consistent with
determine and is therefore chosen as the single regres- those shown in Table 5, indicating that the water which
sion variable. is entrained into the solvent is indeed the result of the

Table 5. Adjusted and Regressed Bubble-Wake Thicknesses Used To Evaluate the Theoretical Fractional Removal
SCM (i ) 2)
bubble
Reynolds no. di (cm) FR relative
sublation run (FUa)/µ sparger a (cm) exact regressed regressed experimental error (%)
R-73-9-21-94 121 porous frit 0.048 2.45 × 10-3 1.95 × 10-3 0.779 0.809 -3.75
R-79-9-27-94 121 porous frit 0.048 2.43 × 10-3 1.95 × 10-3 0.841 0.863 -2.60
R-80-9-29-94 121 porous frit 0.048 2.21 × 10-3 1.95 × 10-3 0.779 0.796 -2.17
R-87-10-12-94 121 porous frit 0.048 2.40 × 10-3 1.95 × 10-3 0.840 0.862 -2.49
R-88-10-13-94 517 porous frit 0.078 1.83 × 10-3 1.38 × 10-3 0.873 0.892 -2.14
R-89-10-15-94 345 porous frit 0.115 8.26 × 10-4 1.03 × 10-3 0.904 0.896 0.93
R-92-11-1-94 32 porous frit 0.031 1.87 × 10-3 2.61 × 10-3 0.818 0.771 6.16
R-94-11-15-94 32 porous frit 0.031 2.49 × 10-3 2.61 × 10-3 0.693 0.684 1.28
R-97-11-17-94 192 porous frit 0.064 8.59 × 10-4 1.58 × 10-3 0.924 0.893 3.51
R-99-11-30-94 234 porous frit 0.078 1.58 × 10-3 1.38 × 10-3 0.821 0.833 -1.45
R-100-12-1-94 498 porous frit 0.166 7.76 × 10-4 7.85 × 10-4 0.901 0.901 0.03
R-106-1-18-95 432 annular shear 0.144 2.69 × 10-3 2.83 × 10-3 0.890 0.887 0.32
R-113-2-9-95 501 annular shear 0.167 4.07 × 10-3 2.29 × 10-3 0.926 0.945 -2.01
R-117-3-7-95 549 annular shear 0.183 4.49 × 10-3 2.00 × 10-3 0.943 0.962 -1.94
R-120-3-21-95 534 annular shear 0.178 6.93 × 10-4 2.08 × 10-3 0.968 0.956 1.30
R-124-4-6-95 432 annular shear 0.144 6.99 × 10-3 2.83 × 10-3 0.802 0.882 -9.12
R-24-9-20-95 519 annular shear 0.173 1.55 × 10-3 2.20 × 10-3 0.880 0.853 3.13
R-30-9-26-95 573 annular shear 0.191 2.20 × 10-3 1.90 × 10-3 0.898 0.907 -0.90
R-33-9-29-95 573 annular shear 0.191 1.66 × 10-3 1.90 × 10-3 0.947 0.942 0.46
R-37-10-3-95 459 annular shear 0.153 2.60 × 10-3 2.63 × 10-3 0.745 0.744 0.12
R-40-10-9-95 459 annular shear 0.153 2.60 × 10-3 2.63 × 10-3 0.796 0.783 1.57
R-43-10-10-95 480 annular shear 0.160 2.60 × 10-3 2.46 × 10-3 0.864 0.858 0.77

Figure 7. Frequency distribution of measured water droplets
entrained in the mineral oil during sublation.
boundary layer and not the wakesat least not the wake
defined in the conventional sense. To seek further
confirmation, attempts to measure the sizes of the water
droplets in the mineral oil were made with the JAVA
video image analyzer described in part 2 and a 10 power
Olympus video camera. Although measuring the drop-
let sizes was not difficult, capturing still images of the
droplets was. The reasons for this were 2-fold. The first
was that the fast speed at which the droplets traversed
the field of view caused streaking in many of the images.
The second problem was that the mineral oil became
opaque due to aeration and had to be frequently
replaced. Nonetheless, some still images were captured
and analyzed. A histogram of the droplet sizes captured
at QA ) 1592 cm3/min and Qw) 0 is shown in Figure 7.
The weighted average for the data is 0.13 cm. When
the values of di and a in Table 5 (for example, run R-106)
are used to calculate the equivalent water droplet
diameter, d̃, that would result in the oil phase, one
discovers that d̃ is, in fact, nearly the same as the
measured value.
d̃ ) 2[(a + di)3 - a3]1/3 )
2[(0.144 + 2.69 × 10-3)3 - 0.1443]1/3 ) 0.11 cm (13)
Thus, it is reasonable to conclude that bubble wake
entrainment into the solvent is dictated by the viscous
boundary layer. It is important, though, to emphasize
that this finding is only applicable to the homogeneous
flow regime. At higher gas velocities the dynamics
governing wake phenomena are vastly different. Though
some light has been shed on the physics of the wake
entrainment mechanism, the issue of determining the
proper variable with which to regress di still remains.
Levich (1962) showed that, for bubble Reynolds
numbers, (FUa/µ), < 600, the rise velocity depended
upon the square of the bubble radius. Experimental
data (Clift et al., 1978) confirm this type of dependence;
however, for contaminated water the dependence has
been shown to be much more linear. Substituting a2
for U and rearranging eq 11, one finds that the theoreti-
cal dependence of the wake-to-bubble volume ratio is
Vw di
∝ ∝ a-3/2 (14)
Vb a
(An a-1 dependence results if the contaminated water
dependence is used; however, the subtle differences
between the two do not affect the conclusion.) When
the correlations in Table 3 for bubble size are used to
Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997 911
eliminate the bubble radius, the wake ratio then be-
comes a function of the superficial gas velocity
( ) ()
Vw
Vb frit
∝
di
a frit
∝ ug-3/4
( ) ()
Vw
Vb shear
∝
di
a shear
∝ ug-3/8 (15)
Like most hydrodynamic properties, eq 15 suggests that
the appropriate variable to regress against is ug. Fur-
thermore, it predicts that the wake ratio for the shear
sparger has a weaker dependence on ug than that for
the porous frit. As shown in eq 16, this was indeed the
case; however, the actual dependencies are somewhat
larger than the predicted values. Moreover, the R2 for
the shear sparger was only 43%, indicating perhaps that
di/a is independent of gas velocity or that there is an
insufficient amount of data on which to base a conclu-
sion.
()
di
) 0.024ug-0.955 R2 ) 94%
a frit
()
di
) 0.038ug-0.600 R2 ) 43% (16)
a shear
When the entire data set is used in the regression, the
dependence on ug is ≈-3/4, and the correlation coefficient
is 84%. This at least tends to confirm the functional
form of eq 15.
The regressed di’s which result from eq 16 are shown
in Table 5. These values were used in the SCM to
calculate the removal efficiencies. As expected, the
regressed values do introduce error between the pre-
dicted and experimental removal efficiencies; however,
the error in most cases is less than (3%. The largest
deviation between theory and experiment was 9.12%
and occurred for run R-124. In this particular run,
however, the steady-state value was suspect due to
interface control problems and may explain the large
error. A better way to assess the model predictions is
to consider the parity plot in Figure 8a. The absence
of biases among the data sets is a strong indication of
model validity. Moreover, this observation suggests
that a generalized correlation of the type described in
part 1 may be sought. For cases where the water-
solvent capacity factor, Pw ≡ Qw/Kow/Qo, is sufficiently
small (Pw e 0.013), it was found that the fractional
removal may be expressed in terms of the water-side
entrainment number, Ew, and the air-water Stanton
number, StwA. (Ew describes the mass transfer associ-
ated with the entrainment of the bubble wake into the
solvent, and StwA describes the mass transfer from the
water phase to the air bubbles.) Since the PCP-
mineral oil system does not meet the condition, the
following correlation is valid only for the pyrene data
FR
) 3.42E0.24 0.39
w StwA R2 ) 93% (17)
1 - FR
where FR is the fractional removal. Equation 17
compares well with the theoretically-based correlation
derived in part 1 in that the mass transfer associated
with the air bubbles (StwA) is weighted more than that
associated with entrainment (Ew). The values of the
exponents appearing in eq 17 and the leading coefficient,
however, do differ from those in the theoretical correla-
912 Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997

analysis, may be compared to that recovered in the oil.

Figure 8. (a) Parity plot of the theoretical fractional removal
calculated by the SCM versus experimental data. (b) Experimen-
tally determined separation factors (corrected for dilution due to
startup inventory) versus the theoretical separation factor.
tion cited in part 1. Nonetheless, the correlation does
confirm that, for strongly hydrophobic compounds,
sublation performance may be estimated quickly and
accurately by considering just the two important mass-
transfer mechanisms associated with rising air bubbles.
Overall Mass Balance and the Separation Fac-
tor. One of the concerns with past sublation studies
has been the lack of an overall mass balance, which
shows that the sublate is recovered in the solvent phase.
Thus, for the pyrene experiments conducted in this
study, the total amount of oil collected from the experi-
ments was analyzed at the end of each run. In this way
the mass of pyrene removed, based on the water
As shown in the third column of Table 6, the mass
recovered was less than what was removed in all but
two cases (actually four; two are within experimental
error). This suggests that, in general, there were
unaccounted losses. Though one possibility is that some
of the pyrene may have been lost to the atmosphere, it
is highly unlikely that this would occur given the low
volatility of pyrene. The most probable loss mechanism
is due to the fact that, during the course of the sublation
experiments, small droplets of oil were continuously
sprayed onto and, to a limited extent, over the walls of
the overflow reservoir. The oil which was sprayed on
the walls formed a film of residue which remained until
it was wiped clean at the end of a run. These oil losses
were estimated to be on the order of 3% of the total oil
processed and can account for as much as a 10% mass
deficiency. For those runs in which the mass balances
disagree by more than 10%, the discrepancies may in
part be attributed to the fact that the variation between
the feed stock concentrations was as much as 15 µg/L.
Since the volumes of the feed stocks fed to the process
were not necessarily equal, calculations involving the
average feed water concentration may result in errors
greater than 10%.
An important measure of the utility of the solvent
phase in any liquid-liquid type separation process is
the separation factor. It is defined as the ratio of the
solvent effluent concentration to the water effluent
concentration. As shown in Table 6, the separation
factors determined from the pyrene sublation data
ranged from about 30 to well over 300. Such large
values are unique to sublation and are indicative of its
superiority over solvent extraction. For example, in
solvent extraction, the ratio of the solvent flow rate to
the water flow rate is generally much greater than the
ratios used in this study. Consequently, the separation
factor for the corresponding liquid-liquid extraction
process would be much less than those shown in Table
6. Also appearing in the table are the theoretical oil
concentrations predicted by the SCM. A comparison of
these values with the experimentally determined con-
centrations indicates that, in all but one case, the theory
overpredicts the concentration. The reason for this is
attributed to the volume of oil used to fill the overflow
reservoir during startup. The initial oil loadings were
a significant fraction (25-46%) of the total oil collected.
Correcting the oil concentrations for dilution by the

Table 6. Overall Mass Balance and Separation Factor Data for the Sublation of Pyrene
effluent oil separation
mass removed average feed average effluent conc. (Co) effluent oil factor
pyrene (water analysis) mass recovered water conc. water conc. experiment conc. (Co) theory (CoFoil/Cw)
sublation run (g) (oil analysis) (g) % error (Cwo) (µg/L) (Cw) (µg/L) (ppm (wt)) (ppm (wt)) experiment
R-73-9-21-94 5.59 × 10-3 5.52 × 10-3 -1.2 96 18.3 0.817 0.908 38
R-79-9-27-94 8.05 × 10-3 6.42 × 10-3 -20.2 119 16.3 0.543 0.811 29
R-80-9-29-94 7.34 × 10-3 5.99 × 10-3 -18.4 119 24.3 0.742 1.126 26
R-87-10-12-94 3.62 × 10-3 3.63 × 10-3 0.3 109 15.0 1.120 1.486 64
R-88-10-13-94 6.98 × 10-3 6.15 × 10-3 -11.9 126 13.6 1.030 1.784 65
R-89-10-15-94 4.78 × 10-3 4.55 × 10-3 -4.8 127 13.2 0.960 1.863 63
R-92-11-1-94 3.58 × 10-3 3.52 × 10-3 -1.6 100 22.9 0.944 1.328 35
R-94-11-15-94 5.52 × 10-3 5.25 × 10-3 -4.9 106 33.5 1.170 2.382 30
R-97-11-17-94 1.70 × 10-3 2.04 × 10-3 20.2 97 10.4 0.622 0.725 52
R-99-11-30-94 8.18 × 10-3 6.75 × 10-3 -17.5 120 20.0 1.220 2.394 52
R-100-12-1-94 5.06 × 10-3 5.04 × 10-3 -0.4 106 10.5 1.570 2.322 129
R-106-1-18-95 4.59 × 10-3 4.22 × 10-3 -8.0 92 10.4 0.970 1.333 80
R-113-2-9-95 4.67 × 10-3 4.27 × 10-3 -8.6 103 5.7 0.840 1.555 128
R-117-3-7-95 4.81 × 10-3 5.60 × 10-3 16.3 96 3.6 1.370 1.476 323
R-120-3-21-95 2.32 × 10-3 3.19 × 10-3 37.5 106 4.7 0.870 0.832 160
R-124-4-6-95 1.01 × 10-2 1.00 × 10-2 -0.2 120 14.2 1.860 3.142 113

initial loadings greatly improves the agreement, as
shown by Figure 8b.
Conclusion
Three-phase, continuous solvent sublation of pyrene
and pentachlorophenol (PCP) was successfully demon-
strated. The conductivity-based controller proved to be
an effective and reliable method for controlling the
height of the solvent-water interface and was superior
to ultrasonic level control. The benefits of the annular
shear sparger were realized as steady-state removal
efficiencies as high as 96% for pyrene and 94% for PCP
were observed.
Sublation results also supported that for hydrophobic
organic compounds the effect of the solvent flow rate is
very weak, and as model predictions suggested, only an
amount of solvent which ensures a steady state is
necessary. This unique feature of solvent sublation had
a profound effect on the separation factor, which is a
measure of solvent utility. Separation factors as high
as 300 were calculated, which demonstrated the supe-
riority of solvent sublation over solvent extraction.
It was shown that the Henry enhancement factor,
which is a measure of the additional air-water parti-
tioning that arises from surface adsorption, is an
excellent indicator of a compound’s susceptibility to
solvent sublation. For example, removal efficiencies
obtained for pyrene, which has an enhancement factor
of approximately 117, were consistently greater than
those obtained for PCP, which has an enhancement
factor of only 77. It was further shown that for
processes which do not exploit the surface activity of
the air-water interface, such as bubble fractionation,
the removal efficiencies for pyrene and PCP were
limited to 68%.
An analysis of the overall resistance to mass transfer
showed that, for nonvolatile hydrophobic compounds,
gas-side resistance is dominant in conventional separa-
tion processes. However, in solvent sublation where the
surface-active nature of hydrophobic compounds is
exploited, gas-side resistance is circumvented. Accord-
ingly, a modified Higbie penetration model of the overall
mass-transfer coefficient was developed as a function
of the characteristic mixing length, λ, and characteristic
velocity, ω. It was argued that λ is the dispersivity
determined by axial dispersion experiments, and ω (for
conditions of mass bubbling in the homogeneous flow
regime) is the superficial gas velocity. Under this
criteria, Higbie model predictions showed that the use
of the annular shear sparger improved mass transfer
by a factor of 2 over the fine porous frit. Agreement
between experimental data and model predictions sup-
ported this assertion.
Lastly, the Series CSTR Model (SCM) proved to be
an excellent tool for predicting fractional removal, as
most predictions were within (3%. For the sublation
of pyrene, fractional removal data were correlated in
terms of the air-water Stanton number, StwA, and the
water-side entrainment number, Ew. The correlation
supported the idea that the two most important mass-
transfer mechanisms are air-water mass transfer and
bubble-wake entrainment.
Acknowledgment
This paper was prepared by Louisiana State Univer-
sity as an account of work sponsored by AIChE/CWRT.
Neither AIChE/CWRT, members of AIChE/CWRT, RE-
Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997 913
SEARCHER, nor any person acting on behalf of any of
them (a) makes any warranty or representation, ex-
pressed or implied, with respect to the accuracy, com-
pleteness, or usefulness of the information contained in
this report, or that the use of any information, ap-
paratus, method, or process disclosed in this report may
not infringe on privately-owned rights or (b) assumes
any liability with respect to the use of, or for damages
resulting from the use of, any information, apparatus,
method, or process disclosed in this report. We give
special thanks to Mike Kurtz for his valuable assistance
with the model-based control algorithm.
Nomenclature (also applies to parts 1 and 2 of
the series)
a ) bubble radius, cm
A ) dimensionless air concentration, A ≡ CA/HCow
CA ) effective concentration of pollutant in air, mol/cm3
Co ) concentration of pollutant in solvent, mol/cm3
Cw ) concentration of pollutant in water, mol/cm3
CP ) chlorinated phenol
di ) bubble-wake thickness, cm
d̃ ) water droplet diameter, cm
D ) dispersion coefficient, cm2/s
Dc ) column diameter, cm
DAB ) binary diffusivity, cm2/s
Ew ) water-side entrainment number, Ew ≡ 3QAdi/Qwa
FR ) fractional removal, FR ≡ 1 - Cw/Cow ) 1 - Aeff/Ainf
H ) effective Henry law constant, Hc + 3KA/a
Hc ) Henry law constant, cm3/cm3
∆Hfus ) heat of fusion, J/mol
i ) number of stages comprising the SCM
k ) water-side mass-transfer coefficient (air/water), cm/s
kl ) overall water-side mass-transfer coefficient (solvent/
water), cm/s
KLav ) overall volumetric mass-transfer coefficient, s-1
KA ) surface adsorption constant, cm
Kow ) solvent-water partition coefficient, cm3/cm3
L ) length of bubble column, cm
n ) number of stages comprising the water column, n ) i
-1
Pw ) water-solvent capacity factor, Pw ≡ Qw/QoKow
PAH ) polynuclear aromatic hydrocarbon
QA ) air flow rate, cm3/s (cm3/min)
Qo ) solvent flow rate, cm3/s (cm3/min)
Qw ) water flow rate, cm3/s (cm3/min)
R ) gas constant, 8.314 J/mol‚K
S ) dimensionless solvent concentration, S ≡ Co/KowCow
Stw ) water-side Stanton number (solvent/water), Stw ≡
klπrc2(1 - )/Qw
StwA ) water-side Stanton number (air/water), StwA ≡
3kπrc2L/aQw
te ) exposure time
T ) temperature, K
Tmelt ) melting point, K
ug ) superficial gas velocity, cm/s
U ) bubble rise velocity, cm/s
Vb ) bubble volume, cm3
Vo ) volume of solvent, cm3
Vw ) wake volume, cm3
W ) dimensionless water concentration, W ≡ Cw/Cow
x ) solubility, mole fraction
Greek Letters
 ) gas holdup, cm3/cm3
H ) Henry enhancement factor, H ≡ 1 + 3KA/aHc
γ ) activity coefficient
κl ) liquid-side mass-transfer coefficient, cm/s
κg ) gas-side mass-transfer coefficient, cm/s
λ ) characteristic length or dispersivity, cm
914 Ind. Eng. Chem. Res., Vol. 36, No. 3, 1997
µ ) fluid viscosity, g/cm‚s
F ) fluid density, g/cm3
σ ) surface tension, dyn/cm
Σ ) separation factor, Σ ≡ CoKowCw
ω ) characteristic velocity, cm/s
Literature Cited
Baird, M. H. I.; Rice, R. G. Axial Dispersion in Large Unbaffled
Columns. J. Chem. Eng. 1975, 9, 171-174.
Bird, R. B.; Stewart, W. E.; Lightfoot, E. N. Transport Phenomena;
John Wiley & Sons: New York, 1960; Chapter 4.
Burns, L. F. The Effect of Reduced Surface Tension on Mass
Transfer and Fluid Dynamics in Bubble Column Reactors. M.S.
Thesis, Louisiana State University, Baton Rouge, LA, 1995.
Chern, J. M.; Yu, C. F. Volatile Organic Compound Emission Rate
from Diffused Aeration Systems. 1. Mass Transfer Modeling.
Ind. Eng. Chem. Res. 1995, 34, 2634-2643.
Clift, R.; Grace, J. R.; Weber, M. E. Bubbles, Drops, and Particles;
Academic Press: San Diego, 1978; Chapter 7.
Cusack, R. W. Solve Wastewater Problems with Liquid/Liquid
Extraction. Chem. Eng. Prog. 1996, 4, 56-63.
Fan, L. S.; Tsuchiya, K. Bubble Wake Dynamics in Liquids and
Liquid-Solid Suspensions; Brenner, H., Ed.; Butterworth-
Heinemann: Stoneham, MA, 1990; Chapter 5.
Geankoplis, C. J. Transport Processes and Unit Operations;
Prentice-Hall: Englewood Cliffs, NJ, 1993; Chapter 6.
Hoff, J. T.; Mackay, D.; Gillham, R.; Shiu, W. Y. Partitioning of
Organic Chemicals at the Air-Water Interface in Environmen-
tal Systems. Environ. Sci. Technol. 1993, 27, 2174-2180.
Kantak, M. V.; Shetty, S. A.; Kelkar, B. G. Liquid Phase Back-
mixing in Bubble Column ReactorssA New Correlation. Chem.
Eng. Commun. 1994, 127, 23-34.
Levich, V. G. Physicochemical Hydrodynamics; Prentice-Hall:
Englewood Cliffs, NJ, 1962; Chapter 8.
Montgomery, J. H.; Welkom, L. M. Groundwater Chemicals Desk
Reference, 2 ed.; Lewis Publishers, Inc.: Chelsea, U.K., 1990; p
640.
Prausnitz, J. M.; Lichtenthaler, R. N.; Azevedo, E. G. Molecular
Thermodynamics of Fluid-Phase Equilibria; Prentice-Hall: En-
glewood Cliffs, NJ, 1986; Chapter 9.
Shah, Y. T.; Kelkar, B. G.; Godbole, S. P.; Deckwer, W. D. Design
Parameters Estimations for Bubble Column Reactors. AIChE
J. 1982, 28, 353-379.
Sherwood, T. K.; Pigford, R. L.; Wilke, C. R. Mass Transfer;
McGraw-Hill: New York, 1975; Chapter 5.
Smith, J. S.; Valsaraj, K. T.; Thibodeaux, L. J. Bubble Column
Reactors for Wastewater Treatment. 1. Theory and Modeling
of Continuous-Countercurrent Solvent Sublation. Ind. Eng.
Chem. Res. 1996a, 35, 1688-1699.
Smith, J. S.; Burns, L. F.; Valsaraj, K. T.; Thibodeaux, L. J. Bubble
Column Reactors for Wastewater Treatment. 2. The Effect of
Sparger Design on Sublation Column Hydrodynamics in the
Homogeneous Flow Regime. Ind. Eng. Chem. Res. 1996b, 35,
1700-1710.
Taylor, R.; Krishna, R. Multicomponent Mass Transfer; Wiley:
New York, 1993; Chapter 9.
Received for review August 22, 1996
Revised manuscript received December 12, 1996
Accepted December 18, 1996X
IE9605241
X Abstract published in Advance ACS Abstracts, February
1, 1997.