Batch kinetics in flow: Online IR analysis and continuous control

Article abstract of DOI:10.1002/anie.201306468

Currently, kinetic data is either collected under steady-state conditions in flow or by generating time-series data in batch. Batch experiments are generally considered to be more suitable for the generation of kinetic data because of the ability to collect data from many time points in a single experiment. Now, a method that rapidly generates time-series reaction data from flow reactors by continuously manipulating the flow rate and reaction temperature has been developed. This approach makes use of inline IR analysis and an automated microreactor system, which allowed for rapid and tight control of the operating conditions. The conversion/residence time profiles at several temperatures were used to fit parameters to a kinetic model. This method requires significantly less time and a smaller amount of starting material compared to one-at-a-time flow experiments, and thus allows for the rapid generation of kinetic data. Go with the flow: By continuously manipulating the flow rate and temperature, classical batch-reactor time-series data were obtained with microreactors under conditions of low dispersion with inline IR analysis. The approach requires significantly less time and a smaller amount of starting material compared to one-at-a-time flow experiments, which allows for the rapid generation of kinetic data.

Full text of DOI:10.1002/anie.201306468

.
Angewandte
Communications
DOI: 10.1002/anie.201306468
Kinetics in Flow
“Batch” Kinetics in Flow: Online IR Analysis and Continuous
Control**
Jason S. Moore and Klavs F. Jensen*
Abstract: Currently, kinetic data is either collected under
steady-state conditions in flow or by generating time-series data
in batch. Batch experiments are generally considered to be
more suitable for the generation of kinetic data because of the
ability to collect data from many time points in a single
experiment. Now, a method that rapidly generates time-series
reaction data from flow reactors by continuously manipulating
the flow rate and reaction temperature has been developed.
This approach makes use of inline IR analysis and an
automated microreactor system, which allowed for rapid and
tight control of the operating conditions. The conversion/
residence time profiles at several temperatures were used to fit
parameters to a kinetic model. This method requires signifi-
cantly less time and a smaller amount of starting material
compared to one-at-a-time flow experiments, and thus allows
for the rapid generation of kinetic data.
time in the plug-flow reactor.[1] These reactors are typically
treated differently only because of deviations from ideality,
such as concentration or temperature gradients from imper-
fect mixing.
In a recent contribution by Mozharov et al., a method that
takes advantage of the ideality of microreactors to derive
time-series data by flow manipulation was presented:[9]
A
Knoevenagel condensation in a microreactor was allowed to
reach steady state at a low flow rate. Then, a step change in
flow rate rapidly flushed the contents out of the reactor. As
the contents exited, an inline Raman probe measured the
product concentration. Although this enabled generation of
a conversion curve in agreement with steady-state experi-
ments, the exact reaction times during this flow-rate step
change could not be determined because the step change was
not actually instantaneous, so that graphical and empirical
estimation of reaction times was required. As stated by
Mozharov et al., “The step increase in flow rate is never
perfect. The system always needs some time to speed up to the
higher flow rate… The exact function F(t) during this transi-
tional period is uncertain.”[9] This non-ideality is caused by
several effects, including non-rigidity of the tubing walls and
the syringe plunger, that prevent an immediate change in the
pressure profile throughout the system.
With the method developed herein, a microreactor system
is allowed to come to steady state at short residence time,
which significantly reduces the initial waiting period before
flow manipulation can begin. Uncertainty in the accurate
determination of residence time is avoided through the use of
a controlled ramp, rather than a step change in flow rate. This
enables the rate of change in residence time to be set, which in
turn allows control over the trade-off between more exper-
imental data and experiment duration.
C
urrent methods for generating kinetic data can be catego-
rized as either sampling steady-state conditions in flow or
generating time-series data in batch.[1] The latter method has
proven particularly useful in identifying complex kinetic
mechanisms.[2] Unfortunately, both of these techniques have
significant limitations. Whereas continuous-flow experiments,
especially in microreactors, have advantages over batch
systems in terms of mixing times,[3,4] temperature control,[5,6]
materials savings,[7] and the ability to perform sequential
experiments without intermediate cleaning steps, batch
experiments are seen as better suited to generating kinetic
data because of the ability to collect data from many time
points in a single experiment.[2] However, with continuous
online measurement, flow experiments can generate such
time-series data by a continuous variation of the flow rate in
a low-dispersion reactor.[8] This analysis is possible because,
under ideal conditions, a batch reactor and a plug-flow reactor
have the same kinetics equation; they will have the exact
same conversion as a function of conditions and time for any
reaction, as time in the batch reactor corresponds to residence
This new method is intended to be broadly applicable to
a wide range of chemical studies for which time-series data
are desired. Ideally, with inline analysis, this method could be
applied to any chemical reaction that can be quenched
chemically, thermally, or otherwise. However, an integrated
online sensor at the end of the reaction zone would allow this
method to be expanded even further. Attenuated total
reflectance (ATR)-FTIR, UV/Vis, flow NMR, Raman, and
fluorescence spectroscopy and other techniques could poten-
tially be implemented.
[*] J. S. Moore, Prof. K. F. Jensen
Department of Chemical Engineering
Massachusetts Institute of Technology
77 Massachusetts Avenue, 66–342, Cambridge, MA 02139 (USA)
E-mail: kfjensen@mit.edu
The efficacy of this new technique was demonstrated for
the Paal–Knorr reaction between 2,5-hexanedione (1) and
ethanolamine (2) in dimethyl sulfoxide (DMSO; Scheme 1
and 2)[10,11] in an automated flow platform (Supporting
Information, Figure S1),[12,13] which used an inline Mettler
Toledo ReactIR ATR-FTIR flow cell[14] to continuously
monitor the effluent from a silicon microreactor.
Homepage: http://web.mit.edu/jensenlab
J. S. Moore
The Dow Chemical Company
2301 North Brazosport Blvd., B-1603, Freeport, TX 77541 (USA)
[**] We thank Novartis for financial support.
Supporting information for this article is available on the WWW
under http://dx.doi.org/10.1002/anie.201306468.
470
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2014, 53, 470 –473

Angewandte
Chemie
S
a
ð3Þ
t ¼ t0 þ Stf
where S is the slope of t versus tf:
Scheme 1. The Paal–Knorr reaction.[15,16]
S ¼ ð1ꢀeꢀa
Þ
ð4Þ
An example of the resulting residence time profile is given
in Figure S2, along with additional derivations. This method
results in a predictable and accurate residence time profile.
Furthermore, the reactor effluent can be measured for
a longer period to reduce variability and increase data
collection. This approach enables a greater sampling rate of
the experimentally collected conversion data (Figure S3) with
a data density that is ten times higher than that reported by
Mozharov et al.,[9] which reduces the error in the estimated
kinetic parameters. In combination with a more rapid
analytical technique, this method could also become appli-
cable to much faster chemical reactions. Furthermore, a lower
value of S would allow the residence time to be ramped more
slowly, enabling analysis at residence time intervals that are
much closer together than the sampling frequency, which
leads to a further increase in data density. Moreover, a lower
value of S would allow the generation of time-series data that
cannot be captured in batch systems because of reaction
kinetics that result in complete conversion too quickly for
conventional in situ analysis techniques. However, this
approach would require analysis of a larger number of
reactor volumes (Figure S4).
Scheme 2. Mechanism of the Paal–Knorr reaction.[11,17,18]
For the microreactor system used, the narrow channel
widths (400 mm) allow diffusion to rapidly eliminate radial
concentration gradients, which leads to minimal dispersion
under the conditions tested.[1,8] Under such conditions of low
dispersion, a flow reactor may be treated as a series of batch
reactors (Figure 1). This treatment allows the use of well-
controlled system transients to generate kinetic information
much more rapidly than with traditional flow experiments;
A first test of the linear-residence-time-ramp method at
1308C with several values of S yielded residence time profiles
that follow the same trend as the steady-state values, but
display a slight deviation in the product concentration from
the steady-state values that increased as S approached one
(Figure S5). This deviation results from the small volume
between the end of the reaction zone and the inline IR flow
cell, owing to the thermal quench zone of the silicon reactor,
the cooling block of the reactor, and the tubing that connects
the reactor to the IR flow cell. This delay volume, Vd, between
the reactor exit and the actual measuring point is included in
the model by determining the time difference between tf,
when a fluid element exits the reaction zone, and tm, when the
concentration is actually measured.
Figure 1. Treatment of a low-dispersion flow reactor as a series of well-
mixed batch reactors. Color represents the extent of conversion from
low (green) to high (red).
the only requirement entails that the history of each fluid
element is known when it exits the reactor.
The method used to generate this time profile is to initially
set the flow rate to give a short residence time, t0, in the
system of volume Vr. After this system has reached steady-
state conditions, the residence time is gradually increased at
a constant rate, a, times the experiment time, t, by reducing
the flow rate, Q, in a controlled manner, so that instantaneous
residence time, tins, of the system is always known.
Z
tm
Vr
t0 þ at
Vd ¼
dt
ð5Þ
tf
Vr
QðtÞ
Solving for tm in terms of tf gives:
tins ¼ t0 þ at ¼
ð1Þ
ꢀ
ꢁ
Vd
Vr
V
t0
a
d a
tm ¼ e atf þ
e
ꢀ 1
ð6Þ
Vr
Each “pseudo-batch” reactor passes through the reactor in
a time t, from an initial time, ti, to the final time, tf; both values
are unique for each fluid element. The residence time that
each fluid element spends in the reactor is a function of when
it exits the reactor:
Substituting the relationship between tf and t analogously to
Eq. (2) gives the residence time profile as a function of tm:
ꢀ
ꢁ
V
Vr
t0
a
d a
t ¼ Seꢀ
tm
þ
ð7Þ
ꢀ
ꢁ
t0
a
t ¼ tf ꢀ ti ¼ ð1 ꢀ eꢀaÞ tf þ
ð2Þ
Replotting the residence time profiles as a function of this
residence time (Figure 2) reveals product concentrations for
each residence time profile that are in good agreement with
This expression can be rewritten as the linear residence time
ramp,
Angew. Chem. Int. Ed. 2014, 53, 470 –473
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
471

.
Angewandte
Communications
Figure 2. Residence time ramp results using corrected residence
times, with S=1/4 (blue), S=1/3 (red), S=1/2 (green), and S=2/3
(orange). Steady states (ꢀ).
each other for different values of S and with the results from
steady-state concentration measurements.
Next, this method was applied for the generation of
conversion profiles at several temperatures. The data were
then fit to a kinetic model (based on Scheme 2), in which
a second-order reaction, which is first-order in each starting
material, forms an intermediate that then undergoes first-
order conversion, which is followed by rapid conversion into
the product. The reverse reaction of the first step is assumed
to be significantly slower, and thus it only has a small effect on
the overall reaction rate. The automated platform was used to
run reactions under multiple conditions (Figure 3a). The data
generated at sample intervals of 15 seconds during the first
experiment are shown in Figure 3b. These product concen-
tration curves were then assigned to their corresponding
temperatures (Figure 3c). The experiments were repeated
three times, and the two kinetic parameters k1 and k2
(Scheme 2) were then fitted by least-squares regression in
Matlab at each temperature by using every 20th data point as
a test set and all other data points as the validation set. The
activation energies thus obtained are 12.2 ꢁ 0.4 kJmolꢀ1 and
20.0 ꢁ 0.9 kJmolꢀ1 for k1 and k2, respectively.
This analysis demonstrates the significantly higher effi-
ciency of this method to generate data for kinetic analysis
compared with traditional steady-state techniques. The
experimental data in Figure 2 could be acquired within
approximately eight hours, and only 5 mL of each 2m reactant
solution was required. In contrast, had traditional steady-state
reactions been performed at each temperature and at
residence times of 10, 20, 30, and 40 min, allowing four
residence times for steady state,[19–21] the experiment would
have taken nearly two days and 13.5 mL of each reaction
solution, and fewer data points would have been acquired.
For this example, the sample interval of 15 seconds that
was used in our new method allowed the collection of a large
number of data points for residence times between 0.5 and
40 min, and smooth conversion trends were observed. How-
ever, as mentioned previously, even with a significantly faster
Figure 3. a) Residence time ramps (c) and temperature steps
(c) used in the experiments. b) Product concentration determined
by inline IR analysis at sample intervals of 15 seconds. c) Product
concentration as a function of residence time at different temper-
atures; from top to bottom: 170, 150, 130, 110, 90, 70, and 508C.
reaction, the use of a value of S closer to zero would enable
the generation of significantly larger amounts of data than the
use of typical in situ batch techniques, allowing for elucidation
of the reaction profile.
472
www.angewandte.org
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
Angew. Chem. Int. Ed. 2014, 53, 470 –473

Angewandte
Chemie
time ramp to cover nearly the full range of reaction conversions, with
additional time for the reactor effluent to reach the IR flow cell.
We have developed a method that rapidly generates time-
series reaction data from flow reactors, analogous to classical
batch reaction data and in agreement with traditional steady-
state flow analysis. The approach was implemented with an
automated microreactor system, allowing for rapid and tight
control of operating conditions. The resulting conversion/
residence time profiles at several temperatures were used to
fit parameters in a kinetic model, in agreement with experi-
ments. This approach could be used to generate time-series
data in flow for a wide range of chemical reactions.
Furthermore, this method could be extended to other
parameters, such as performing concentration ramps.
Received: July 24, 2013
Published online: November 29, 2013
Keywords: automation · continuous flow · IR spectroscopy ·
.
kinetics · microreactors
[1] O. Levenspiel, Chemical Reaction Engineering, 3rd ed., Wiley,
New York, 1999.
[2] F. E. Valera, M. Quaranta, A. Moran, J. Blacker, A. Armstrong,
J. T. Cabral, D. G. Blackmond, Angew. Chem. 2010, 122, 2530 –
2537; Angew. Chem. Int. Ed. 2010, 49, 2478 – 2485.
[3] K. F. Jensen, Chem. Eng. Sci. 2001, 56, 293 – 303.
[4] V. Hessel, H. Lçwe, Chem. Eng. Technol. 2005, 28, 267 – 284.
[5] A. J. deMello, Nature 2006, 442, 394 – 402.
Experimental Section
The microfluidic system that was used in this work is depicted in the
Supporting Information, Figure S1, which includes a schematic rep-
resentation of the silicon microreactor. Two Harvard stainless steel
syringes were loaded separately with solutions of 2,5-hexanedione (1)
and ethanolamine (2) in dimethyl sulfoxide (DMSO; 2m), which
resulted in a solution with 1m concentrations upon mixing in the
reactor. One syringe was placed on each of two Harvard syringe
pumps (PHD 2000) to control the residence time. The flow rates were
updated every second through linked RS-232 communications to
a Dell (Optiplex 960) computer. These syringe pumps were connected
to a silicon microreactor[22] with a 120 mL reaction zone and a cooled
inlet/outlet zone, the temperature of which was maintained at 228C
with a recirculating VWR chiller (model 1171MD). The cooled
section allowed the reactant streams to mix fully before the reaction
occurred and thermally quenched the reaction. The temperature of
the reaction zone was controlled with an Omega (CN9311) controller
and an Omega (CSH-102135/120 V) heating cartridge. The controller
was connected to the computer with an RS-232 cable to read the
measured reactor temperature and program the temperature set
point. Furthermore, because of the high heat transfer coefficient of
silicon, the temperature of the reaction zone could be quickly
changed between set points and the fluid stream rapidly reached the
desired temperature in both the reaction and quench zones. A Mettler
Toledo ReactIR iC 10 that was outfitted with a DiComp ATR flow
cell (10 mL) was used for continuous inline monitoring, averaging 30
spectrum scans, which were saved to an Excel file once every 15 s.
Labview software (version 8.5.1) on the computer communicated
with the syringe pumps and temperature controller and read the IR
Excel export files to determine the reaction conversion based upon
calibrations to peak heights. Matlab scripts (version 2010b) within
Labview controlled the reaction temperature set points and syringe-
pump flow rates.
[6] K. Jꢀhnisch, V. Hessel, H. Lçwe, M. Baerns, Angew. Chem. 2004,
116, 410 – 451; Angew. Chem. Int. Ed. 2004, 43, 406 – 446.
[7] K. F. Jensen, MRS Bull. 2006, 2, 101 – 107.
[8] K. D. Nagy, B. Shen, T. F. Jamison, K. F. Jensen, Org. Process
Res. Dev. 2012, 16, 976 – 981.
[9] S. Mozharov, A. Nordon, D. Littlejohn, C. Wiles, P. Watts, P.
Dallin, J. M. Girkin, J. Am. Chem. Soc. 2011, 133, 3601 – 3608.
[10] V. Amarnath, D. C. Anthony, K. Amarnath, W. M. Valentine,
L. A. Wetterau, D. G. Graham, J. Org. Chem. 1991, 56, 6924 –
6931.
[11] B. Mothana, R. J. Boyd, J. Mol. Struct. 2007, 846, 97 – 107.
[12] J. S. Moore, K. F. Jensen, Org. Process Res. Dev. 2012, 16, 1409 –
1415.
[13] J. S. Moore, K. F. Jensen in Microreactors in Organic Chemistry
and Catalysis, 2nd ed. (Ed.: T. Wirth), Wiley-VCH, Weinheim,
2013, pp. 81 – 100.
[14] DS Series Sampling Technology; Mettler Toledo, LLC. http://us.
mt.com/us/en/home/products/L1_AutochemProducts/L2_in-
situSpectrocopy/AgX-FiberConduit-Sampling-Technology-DS-
Series.html; accessed November 7, 2010.
[15] L. Knorr, Ber. Dtsch. Chem. Ges. 1885, 18, 299 – 311.
[16] C. Paal, Ber. Dtsch. Chem. Ges. 1885, 18, 367 – 371.
[17] Name Reactions in Heterocyclic Chemistry (Eds.: J.-J. Li, E. J.
Corey), Wiley, Hoboken, 2005.
[18] V. Amarnath, K. Amarnath, W. M. Valentine, M. A. Eng, D. G.
Graham, Chem. Res. Toxicol. 1995, 8, 234 – 238.
[19] J. P. McMullen, K. F. Jensen, Org. Process Res. Dev. 2010, 14,
1169 – 1176.
[20] J. P. McMullen, M. T. Stone, S. L. Buchwald, K. F. Jensen,
Angew. Chem. 2010, 122, 7230 – 7234; Angew. Chem. Int. Ed.
2010, 49, 7076 – 7080.
[21] J. P. McMullen, K. F. Jensen, Org. Process Res. Dev. 2011, 15,
398 – 407.
For the data in Figure 3, once the temperature had equilibrated at
a set point, and the residence time had equilibrated at t0, the
residence time ramp ran for 70 min. The temperature set point was
then changed to the next set of conditions, and the process was
repeated. The duration was set at 70 min by adding 10 min to the time
that would be necessary for tf to reach 40 min, allowing the residence
[22] M. W. Bedore, N. Zaborenko, K. F. Jensen, T. F. Jamison, Org.
Process Res. Dev. 2010, 14, 432 – 440.
Angew. Chem. Int. Ed. 2014, 53, 470 –473
ꢀ 2014 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
www.angewandte.org
473