Monitoring of itaconic acid hydrogenation in a trickle bed reactor using fiber-optic coupled near-infrared spectroscopy

Article abstract of DOI:10.1366/000370203321558209

Near-infrared (NIR) spectroscopy has been applied to determine the conversion of itaconic acid in the effluent stream of a trickle bed reactor. Hydrogenation of itaconic to methyl succinic acid was carried out, with the trickle bed operating in recycle mode. For the first time, NIR spectra of itaconic and methyl succinic acids in aqueous solution, and aqueous mixtures withdrawn from the reactor over a range of reaction times, have been recorded using a fiberoptic sampling probe. The infrared spectra displayed a clear isolated absorption band at a wavenumber of 6186 cm^-1 (wavelength 1.617 μm) resulting from the =C-H bonds of itaconic acid, which was found to decrease in intensity with increasing reaction time. The feature could be more clearly observed from plots of the first derivatives of the spectra. A partial least-squares (PLS) model was developed from the spectra of 13 reference samples and was used successfully to calculate the concentration of the two acids in the reactor effluent solution. Itaconic acid conversions of 23-29% were calculated after 360 min of reaction time. The potential of FT-NIR with fiber-optic sampling for remote monitoring of three-phase catalytic reactors and validation of catalytic reactor models is highlighted in the paper.

Full text of DOI:10.1366/000370203321558209

Monitoring of Itaconic Acid Hydrogenation in a Trickle Bed
Reactor Using Fiber-Optic Coupled Near-Infrared
Spectroscopy
JOSEPH WOOD* and PAUL H. TURNER
Centre for Formulation Engineering, Department of Chemical Engineering, University of Birmingham, Edgbaston, Birmingham,
B15 2TT, UK (J.W.); and Bruker Optics Ltd., Banner Lane, Coventry CV4 9GH, UK (P.H.T.)
Near-infrared (NIR) spectroscopy has been applied to determine the
conversion of itaconic acid in the ef uent stream of a trickle bed
reactor. Hydrogenation of itaconic to methyl succinic acid was car-
ried out, with the trickle bed operating in recycle mode. For the
Ž rst time, NIR spectra of itaconic and methyl succinic acids in aque-
stream. The method was preferable to analysis by gas
ous solution, and aqueous mixtures withdrawn from the reactor
over a range of reaction times, have been recorded using a Ž ber-
also being applied to problems of interest to the process
industries. For example, Coffey et al.6 used a Ž ber-optic
coupled acousto-optic tunable Ž lter NIR (AOTF-NIR) to
monitor the vapor content of a rotary dryer ef uent
chromatography (GC) or Karl Fisher titration, since the
need to wait 40 minutes for the results and the possible
optic sampling probe. The infrared spectra displayed a clear iso-
21
exposure of process operators to chemical vapor during
sampling were avoided. Drying times were more accu-
rately determined by AOTF-NIR, hence avoiding unnec-
essary drying periods at the end of a batch. Yalvac et al.7
analyzed a mixture of light alkenes using NIR and Coffey
et al.8 followed a reaction of industrial signiŽ cance using
Ž ber-optic NIR spectroscopy. Ward et al.9 used on-line
NIR spectroscopy to determine the end point of a bulk
pharmaceutical reaction in a closed loop hydrogenator. A
palladium/carbon catalyst was used and the reaction sol-
vent was tetrahydrofuran. Monitoring of the process in
real time enabled the key reaction to be stopped before
the concentration of undesirable by-products reached a
critical level.
lated absorption band at a wavenumber of 6186 cm (wavelength
1.617 mm) resulting from the 5C–H bonds of itaconic acid, which
was found to decrease in intensity with increasing reaction time.
The feature could be more clearly observed from plots of the Ž rst
derivatives of the spectra. A partial least-squares (PLS) model was
developed from the spectra of 13 reference samples and was used
successfully to calculate the concentration of the two acids in the
reactor ef uent solution. Itaconic acid conversions of 23–29% were
calculated after 360 min of reaction time. The potential of FT-NIR
with Ž ber-optic sampling for remote monitoring of three-phase cat-
alytic reactors and validation of catalytic reactor models is high-
lighted in the paper.
Index Headings: Near infrared; NIR; Hydrogenation; Itaconic acid;
Fiber optics; On-line; Trickle bed reactor; Partial least squares;
PLS.
Itaconic acid hydrogenation has been used as a model
reaction in the study of mass transfer and reaction ca-
pabilities of multiphase catalytic reactors such as the co-
current down ow contactor reactor.10 The reaction be-
tween itaconic acid and hydrogen involves the saturation
INTRODUCTION
Catalytic technology plays a major role in the bulk and
Ž ne chemical industry.1 Trickle beds, which consist of
Ž xed beds of catalyst contacted by cocurrent down ows
of gas and liquid, are one of the most common types of
three-phase catalytic reactors, and their design relies upon
models describing reaction kinetics, mass transfer, and
hydrodynamics of trickle  ow.2 The primary variables of
interest in trickle bed reactor models are the concentra-
tion of gaseous reactants dissolved in the liquid and the
conversion of liquid-phase reactants.3 This paper de-
scribes the application of Ž ber-optic coupled near-infrared
(NIR) spectroscopy to the monitoring of liquid-phase re-
actant composition at the outlet of a laboratory trickle
bed reactor. It is demonstrated that NIR spectroscopy pro-
vides a quick and accurate method of measuring liquid-
phase composition and hence conversion of key reactants.
The potential of the technique in evaluating models of
trickle bed reactors in order to gain conŽ dence in their
predictions for catalytic reactor design is identiŽ ed in the
paper.
5
of the C C double bond as follows:
The advantages of the above reaction for studying the
capabilities of catalytic reactors are that the kinetics are
Ž rst order with respect to hydrogen, and the reaction can
be carried out at low temperatures and pressures of
8
70 C and 1.5 bar(a), respectively. In trickle bed reactors,
complete consumption of hydrogen cannot be assumed,
and therefore the progress of the reaction must be mon-
itored by measuring the concentration of itaconic acid in
the liquid ef uent from the bed. Traditionally, the con-
centrations of itaconic acid and methyl succinic acid have
been measured using GC. Before injection to the GC, the
acids must Ž rst be esteriŽ ed, then extracted in chloroform
and dried over anhydrous sodium sulfate.11 This proce-
The use of NIR spectroscopy in remote monitoring has
been widely reported in the chemical literature4,5 and is
Received 29 August 2002; accepted 4 November 2002.
* Author to whom correspondence should be sent.
0003-7028 / 03 / 5703-0293$2.00 / 0
Volume 57, Number 3, 2003
APPLIED SPECTROSCOPY
293
q
2003 Society for Applied Spectroscopy

FIG. 1. Experimental setup of the laboratory trickle bed hydrogenation experiment.
dure takes 2–3 h and is subject to experimental errors
wool at the top and bottom with a layer of glass beads
at the top acting as the distributor. The liquid and gas
 ow out of the base of the bed and are passed through a
disengagement zone; after passing through this zone the
liquid was recycled to the supply tank and gas was vented
to the atmosphere.
associated with the esteriŽ cation and extraction. There-
fore, this study examined the application of NIR spec-
troscopy to the measurement of itaconic and methyl suc-
cinic acid concentrations in order to achieve a more ac-
curate measurement in a fraction of the time taken by
GC. This paper reports the Ž rst known application of NIR
to the monitoring of itaconic acid hydrogenation.
In the following section, the experimental setup of the
trickle bed hydrogenator and details of the applied infra-
red spectroscopy techniques are given. The development
of a partial least-squares (PLS) model for the calculation
of liquid-phase composition from the recorded spectra is
described. Then NIR spectra of samples from the reactor
ef uent stream and plots of the concentrations of itaconic
and methyl succinic acids as a function of reaction time
are presented and discussed. Finally, the potential of in-
line NIR for evaluating computer models of trickle bed
performance is highlighted.
Fourier Transform Near-Infrared System. A VEC-
TOR 33 FT-IR spectrometer (Bruker Optics) was used
for the measurements. The instrument was conŽ gured for
near-infrared operation using a tungsten halogen NIR
source, a thermoelectrically cooled InAs detector sensi-
tive over the range 12 800–3470 cm21, and a NIR quartz
beam splitter. A Ž ber-optic coupling unit was Ž tted to the
right side exit port after the interferometer to focus light
onto the Ž ber-optic cable and direct the return light from
the probe to the InAs detector, which was housed in the
unit. The probe used was of the quartz immersion type
with a Ž xed 2 mm gap (or path length) and Ž tted with
two single Ž bers 2 m in length. Spectrometer control and
measurements were performed using a Windows NT PC
running Bruker OPUS/IR software.
EXPERIMENTAL
Hydrogenation Experiment with Near-Infrared
Monitoring. Prior to carrying out the hydrogenation ex-
periment in the trickle bed reactor, the catalyst was ac-
tivated by reduction in a  ow of hydrogen for one hour.
Then, 9.5 L of itaconic acid solution was made up in the
Trickle Bed Reactor Apparatus. A trickle bed reactor
rig was constructed as displayed in Fig. 1. The reactor
vessel consisted of a glass column 60 cm in depth with
a diameter of 5 cm at the top and 7 cm at the base. The
feed liquid was supplied to the top of the reactor from a
heated tank by a peristaltic pump. Hydrogen gas was also
supplied to the top of the reactor, via a non-return valve.
The bed was Ž lled with Pd/Al2O3 catalyst pellets 4–8 mm
in diameter (Johnson Matthey), packed between glass
8
supply tank and heated to 60 C. The hydrogenation ex-
periment then commenced, with liquid feed supplied to
21
the top of the reactor at a rate of 280 mL min and
21
22 21
hydrogen gas at a rate of 1.5 L min (0.002 kg m
s )
294
Volume 57, Number 3, 2003

21
22
and 4.2 L min (0.005 kg m
s
21) respectively during
two different runs. In the feasibility study reported here,
50 mL liquid samples were taken from the reactor ef u-
ent for analysis at 45-min intervals up to a reaction time
of 6 h, and NIR spectra of the samples were taken. Mea-
surement conditions were 64 coadded scans with a mea-
21
surement time of 60.9 s, at a resolution of 4 cm (1.1
nm at 6000 cm21). A background or zero reference spec-
trum was taken with air.
RESULTS AND DISCUSSION
Calibration. A calibration set was constructed from
reference spectra of samples containing known concen-
trations of itaconic and methyl succinic acids. A set of
13 reference samples and one test sample were prepared
by adding the acids in powdered form to water. The con-
centrations of each acid covered the range 0–0.23
21
mol L across the samples, and the total concentration
of the two acids in each sample was 0.23 mol L21. NIR
reference spectra were recorded for each sample by dip-
ping the probe into the sample container. The same ex-
perimental measurement parameters were used as for the
reactor ef uent samples. Each calibration sample mea-
surement was repeated twice with the sample shaken in-
between in order to build slight sample-handling vari-
ability into the model. Any cross-sample contamination
was avoided by cleaning the probe with a tissue between
sample measurements.
A PLS calibration model was constructed from the ref-
erence spectra using the Bruker OPUS/QUANT software,
which is based on the PLS-1 algorithm. In order to make
the input spectra more consistent, a data preprocessing of
Ž rst derivative and vector normalization was applied. The
spectral ranges used in the calibration were 6381–5617
21
21
cm for itaconic acid and 6096–5714 cm for methyl
succinic acid. These ranges encompass the overtone C–H
stretch region of the two acids that remain accessible in
between the strong water absorption bands. Figure 2a
shows the original 26 absorbance spectra of the calibra-
tion samples, which all appear very similar on this scale,
whereas Fig. 2b shows the clear differences when the
spectral region of interest is expanded considerably after
the data preprocessing. Figure 2c highlights the differ-
ences between the spectra of pure itaconic and methyl
succinic acids in solution after preprocessing.
The quantitative analysis method was checked by
cross-validation in the leave-one-sample-out mode (2
spectra left out in turn). Table I shows the concentrations
of itaconic, methyl succinic, and total acid for each of
the 13 calibration samples determined both gravimetri-
cally and by NIR cross-validation. Figure 3 shows the
cross-validation graphs of the actual vs. predicted con-
centrations of itaconic and methyl succinic acids, respec-
tively. The excellent correlation is indicated by the high
R2
value of 0.9997 for each acid and the low root mean
square errors of cross validation (RMSECV) of 0.00120
and 0.00117 for itaconic and methyl succinic acids, re-
spectively. The PLS rank used for both graphs was one.
The RMSECV values are an indication of the accuracy
when predicting the concentrations of unknown samples.
As a further check, we also measured a test sample,
which was prepared gravimetrically (see Table I) and not
FIG. 2. (a) NIR spectra for the calibration samples displayed in the
range 12 000–4000 cm 1; (b) normalized Ž rst-derivative spectra for the
2
calibration samples shown for the range 6300–5600 cm 1; (c) normal-
2
ized Ž rst-derivative spectra for pure itaconic acid and methyl succinic
acid in aqueous solution over the range 6300–5600 cm 1. The two spec-
2
tra are offset for clarity.
APPLIED SPECTROSCOPY
295

TABLE I. Details of the calibration and test samples. Itaconic, methyl succinic, and total acid concentrations determined (1) gravimetrically
from the weight of powdered acid used to prepare the sample and (2) from NIR analysis by internal cross-validation for the calibration
samples A through M (leave-one-sample-out mode) and by external validation using the Ž nal PLS calibration models for the test sample.
The data in parentheses represent the difference between measurements (2) and (1).
Acid concentration
determined gravimetrically
Acid concentration determined by NIR;
results are averaged for 2 spectra
Methyl
Methyl
Calibration
sample
Itaconic acid
succinic acid
Total acid
Itaconic acid
succinic acid
Total acid
1
2
1
1
1
1
1
2
2
2
2
2
(mol L
)
(mol L
)
(mol L
)
(mol L
)
(mol L
)
(mol L )
A
0.2321
0.2118
0.1922
0.1728
0.1538
0.1352
0.1165
0.0966
0.0763
0.0573
0.0388
0.0202
0.0000
0.1659
0.0000
0.0195
0.0385
0.0573
0.0769
0.0961
0.1156
0.1361
0.1542
0.1737
0.1923
0.2112
0.2306
0.0676
0.2321
0.2313
0.2307
0.2301
0.2307
0.2313
0.2321
0.2327
0.2305
0.2310
0.2311
0.2314
0.2306
0.2335
0.2304
(20.0017)
0.2141
(10.0023)
0.1940
(10.0018)
0.1716
(20.0012)
0.1534
(20.0004)
0.1354
(10.0002)
0.1151
(20.0014)
0.0969
(10.0003)
0.0765
(10.0002)
0.0564
(20.0009)
0.0385
(20.0003)
0.0204
(10.0002)
0.0014
20.0004
(20.0004)
0.0181
(20.0014)
0.0378
(20.0007)
0.0595
(10.0022)
0.0780
(10.0011)
0.0967
(10.0006)
0.1151
(20.0005)
0.1347
(20.0014)
0.1550
(10.0008)
0.1731
(20.0006)
0.1928
(10.0005)
0.2125
(10.0013)
0.2289
0.2300
0.2322
0.2318
0.2311
0.2314
0.2321
0.2302
0.2316
0.2315
0.2295
0.2313
0.2329
0.2303
0.2330
B
C
D
E
F
G
H
I
J
K
L
M
(10.0014)
0.1661
(20.0017)
0.0669
Test Sample
(10.0002)
(20.0007)
used for calibration. We then took the difference between
the predicted concentrations from the calibration model
used and the known values from the weight of powdered
acids used to prepare the test sample. This cross-check
Interpretation of the Spectra. Figure 2a displays the
original absorbance spectra of all calibration samples
with the main features resulting from water, speciŽ cally
21
the 6870 cm overtone O–H stretch and the onset of
21
21
gave actual discrepancies of 0.0002 mol L and 0.0007
strong absorption below 5500 cm
due to the O–H
21
mol L for itaconic and methyl succinic acids, respec-
stretch-bending combination. It is just possible to see
some weak absorption due to the acids in the region of
tively, suggesting that the method is capable of achieving
21
21
6
the estimated accuracy of 0.001 mol L
.
the C–H overtones between 6200 and 5850 cm located
between these large water peaks. A considerable y-scale
expansion of this spectral region after the data prepro-
cessing clearly shows the spectral variations due to the
changing acid concentrations, as observed from Fig. 2b.
21
The feature between 6220 and 6160 cm is clearly as-
5
signed to the vinylic C–H overtone stretch of itaconic
acid, which disappears in pure methyl succinic acid so-
lution, as shown in Fig. 2c. The absorbance decreases as
the double bond of itaconic acid becomes saturated dur-
ing hydrogenation to methyl succinic acid.
Figure 4 shows the raw absorbance spectra of itaconic
acid, methyl succinic acid, and four samples taken from
the reactor ef uent. It is observed that the spectra of the
pure samples and reactor ef uent are very similar, except
in the range 6220–6160 cm21, where a feature that is
sensitive to the extent of reaction occurs. The inset of
Fig. 4 shows an enlargement of this feature. The pro-
gressive change can be more clearly observed from a plot
of the Ž rst-derivative spectra, shown in Fig. 5. It is seen
that the Ž rst-derivative spectra for the reaction mixtures
FIG. 3. Cross-validation plots of predicted vs. actual concentrations of
itaconic acid (Ž lled circles) and methyl succinic acid (open inverted
triangles).
296
Volume 57, Number 3, 2003

FIG. 5. First-derivative spectra of itaconic acid solution (0.23 mol L 1),
2
FIG. 4. NIR spectra of itaconic acid solution (0.23 mol L 1), methyl
2
methyl succinic acid solution (0.23 mol L 1), and samples withdrawn
2
succinic acid solution (0.23 mol L 1), and samples withdrawn from the
2
from the reactor ef uent at various reaction times shown over the range
reactor ef uent at various reaction times shown over the range 6400–
6260–6100 cm 1. Arrows indicate the direction of reaction progress.
2
1
5600 cm . Inset: Enlarged Ž gure for the region 6220–6160 cm 1 show-
ing a feature that is sensitive to the reactant conversion. The arrow
indicates the direction of reaction progression.
2
2
actors. In this paper, it has been shown that NIR spec-
troscopy with Ž ber-optic probes can be used to determine
the composition of liquid ef uent streams from trickle
bed reactors and that it is a faster and more accurate
analysis method than GC for this application. Since the
NIR method used could be easily adapted for in-line anal-
ysis, it could be used to evaluate whether the reactor ef-
 uent compositions predicted by computer simulations
are conŽ rmed by experiments carried out under the same
conditions. At Birmingham University, the use of NIR in
validating reactor models will form part of on-going stud-
ies of three-phase catalytic reactor performance.
gradually approach that of methyl succinic acid with in-
creasing reaction time, indicating a progressive increase
in itaconic acid conversion.
Quantitative Analysis. The PLS calibration was used
to calculate simultaneously the concentration of itaconic
acid and methyl succinic acid for each sample withdrawn
from the reactor ef uent. In Fig. 6 the concentrations of
the two acids are plotted as a function of reaction time
for two different hydrogen  ow rates through the reactor.
X,
t
The fractional conversion,
of itaconic acid at time
was calculated from the equation:
C0 2 Ct
CONCLUSION
X 5
C0
Near-infrared spectroscopy with a Ž ber-optic sampling
probe has been used to record spectra of itaconic and
methyl succinic acids in aqueous ef uent solution from
a trickle bed hydrogenation reactor operating in recycle
mode. A PLS calibration model was constructed from
spectra recorded for 13 different reference samples, and
cross-validation showed excellent correlation between ac-
C
where
Ct
is the initial concentration of itaconic acid, and
0
t.
is the concentration at time After 360 min of reaction
time, the conversion of itaconic acid was 23% and 29%
for hydrogen  ow rates of 4.2 L min and 1.5 L min
21
21
,
respectively. The lower conversion at the higher  ow rate
of hydrogen may have resulted from thinning of the liq-
uid Ž lm on the catalysts and associated reduction in liq-
uid hold up at the higher gas  ow rate. Also, it is thought
that the higher  ow rate of hydrogen may lead to a great-
er temperature decrease over the bed and hence a lower
rate of reaction. It is noted that the NIR technique pro-
vides scope for in-line reaction monitoring of the reactor
ef uent stream composition. We will investigate the pos-
sible in-line monitoring of reactions in future studies.
Models representing the behavior of trickle bed reac-
tors need to account for the effect of parameters such as
liquid hold up, gas–liquid contacting, and catalyst effec-
tiveness upon conversion of reactants in the liquid  ow-
ing through the bed. Earlier models of trickle bed reactors
assumed plug  ow of liquid and used an effectiveness
factor to describe the fraction of the external catalyst area
wetted by liquid.12 More recent phenomenological mod-
els are based on rigorous analyses of multicomponent
mass and energy transfer on the reactor and catalyst
scale.13 Such models will have to be further validated
over a wide range of reactor operating conditions in order
to gain conŽ dence in their predictions for scaling up re-
FIG. 6. Concentrations of itaconic and methyl succinic acids in the
reactor ef uent stream as a function of reaction time at two hydrogen
 owrates.
APPLIED SPECTROSCOPY
297

3. M. P. Dudukovic, F. Larachi, and P. L. Mills, Chem. Eng. Sci. 54,
1975 (1999).
4. A. G. Cavinato, D. M. Mayes, Z. H. Ge, and J. B. Callis, Anal.
Chem. 62, 1977 (1990).
5. R. Mackison, S. J. Brinkworth, R. M. Belchamber, R. E. Aries, D.
J. Cutler, C. Deeley, and H. M. Mould, Appl. Spectrosc. 46, 1020
(1992).
6. C. Coffey, A. Predoehl, and D. S. Walker, Appl. Spectrosc. 52, 717
(1998).
tual and predicted concentrations of the two acids. The
PLS model was applied to determine the concentrations
of the acids in the reactor ef uent stream, and hence, the
conversion of itaconic acid was calculated. After a re-
action time of 360 min, the conversion of itaconic acid
was 23–29%, depending inversely upon the hydrogen
 ow rate used in the reactor. The potential of the tech-
nique for in-line monitoring of three-phase catalytic re-
actors and veriŽ cation of trickle bed reactor models was
identiŽ ed.
7. E. D. Yalvac, M. B. Seasholtz, and S. R. Crouch, Appl. Spectrosc.
51, 1303 (1997).
8. C. Coffey, B. E. Cooley, and D. S. Walker, Anal. Chim. Acta 395,
335 (1999).
9. H. W. Ward, S. S. Sekulic, M. J. Wheeler, G. Taber, F. J. Urbanski,
F. E. Sistare, R. Norris, and P. K. Aldridge, Appl. Spectrosc. 52, 17
(1998).
ACKNOWLEDGMENTS
J. Wood is grateful to Bruker Optics Limited for carrying out the
NIR spectroscopic measurements and to Johnson Matthey plc for sup-
plying the catalyst used in the trickle bed reactor without charge.
10. X. X. Lu, A. P. Boyes, and J. M. Winterbottom, Chem. Eng. Sci.
51, 2715 (1996).
11. M. A. Harmon and H. W. Doelle, J. Chromatogr. 42, 157 (1969).
12. D. E. Mears, Chem. Eng. Sci. 26, 1361 (1974).
13. M. R. Khadilkar, P. L. Mills, and M. P. Dudukovic, Chem. Eng.
Sci. 54, 2421 (1999).
1. H.-U. Blaser and M. Studer, Appl. Catal., A 189, 191 (1999).
2. M. H. Al-Dahhan, F. Larachi, M. P. Dudukovic, and A. Laurent,
Ind. Eng. Chem. Res. 36, 3292 (1997).
298
Volume 57, Number 3, 2003