Continuous flow iodination using an automated computer-vision controlled liquid-liquid extraction system

Article abstract of DOI:10.1016/j.tetlet.2017.01.029

A dynamic computer-vision control system incorporating open-source software technologies (Python, OpenCV) was used to automate a gravity based in-line liquid-liquid extraction during the iodination of enaminones in continuous flow. The system was able to cope with significant colouration of the organic reaction stream and with significant volume/flow changes in the aqueous and organic phase streams due to extraction of the water soluble acetonitrile co-solvent.

Full text of DOI:10.1016/j.tetlet.2017.01.029

Accepted Manuscript
Continuous flow iodination using an automated computer-vision controlled liq-
uid-liquid extraction system.
Matthew O'Brien, Dennis A. Cooper, Jonathan Dolan
PII:
S0040-4039(17)30051-5
DOI:
http://dx.doi.org/10.1016/j.tetlet.2017.01.029
TETL 48532
Reference:
Tetrahedron Letters
To appear in:
Received Date:
Revised Date:
Accepted Date:
28 November 2016
5 January 2017
10 January 2017
Please cite this article as: O'Brien, M., Cooper, D.A., Dolan, J., Continuous flow iodination using an automated
computer-vision controlled liquid-liquid extraction system., Tetrahedron Letters (2017), doi: http://dx.doi.org/
10.1016/j.tetlet.2017.01.029
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Graphical Abstract
To create your abstract, type over the instructions in the template box below.
Fonts or abstract dimensions should not be changed or altered.
Continuous flow iodination using an
automated computer-vision controlled
liquid-liquid extraction system
Leave this area blank for abstract info.
Matthew O’Brien,* Dennis A. Cooper and Jonathan Dolan

1
Tetrahedron Letters
Continuous flow iodination using an automated computer-vision controlled liquid-
liquid extraction system.
Matthew O’Brien*, Dennis A. Cooper and Jonathan Dolan
Lennard-Jones Laboratories, Keele University, Keele, Borough of Newcastle-under-Lyme, Staffordshire, ST5 5BG, UK.
ARTICLE INFO
ABSTRACT
Article history:
Received
Received in revised form
Accepted
A dynamic computer-vision control system incorporating open-source software technologies
(Python, OpenCV) was used to automate a gravity based in-line liquid-liquid extraction during
the iodination of enaminones in continuous flow. The system was able to cope with significant
colouration of the organic reaction stream and with significant volume/flow changes in the
aqueous and organic phase streams due to extraction of the water soluble acetonitrile co-solvent.
Available online
2009 Elsevier Ltd. All rights reserved.
Keywords:
Open-Source
Automation
Computer-Vision
Continuous-Flow
Iodination
Over the last decade or so, flow chemistry has emerged as an
alternative model to more traditional batch approaches to
synthesis.1 In addition to offering significant safety benefits when
using hazardous reagents, intermediates or conditions,2 flow
chemistry often enables superior control over interfacial
exchange parameters3 leading to highly efficient and scale-
invariant chemical processes. A particularly attractive feature of
flow chemistry is the ability to incorporate in-line purification
techniques. The use of solid-supported scavengers4 and phase-
switching reagents have enabled many noteworthy flow
syntheses.5 In-line liquid-liquid extraction is also emerging as an
alternative approach. Whilst it has limitations (obviously relying
on efficient extraction from one phase to another) it can also offer
advantages. Firstly, solid-supported reagents are often much
more expensive than the equivalent soluble form.6 Secondly,
unless some regeneration protocol is incorporated into the
system, solid-supported reagents become depleted over time and
at some point will need replacing, possibly requiring the system
to be temporarily shut down. Liquid extractants, which can be
pumped continuously into the system, do not suffer from this
issue. The use of solid supports also causes significant scale-
dependent dispersive effects.7 Whilst there will also be
unavoidable dispersive effects associated with liquid-liquid
extraction processes these can, in principle, be rendered scale-
invariant. Semi-permeable hydrophobic membranes, particularly
those based on expanded PTFE,8 have been used to effect
continuous flow separation of aqueous and organic flow streams.9
An alternative approach, essentially a flow adaptation of the
traditional separating funnel, relies on the vertical separation of
immiscible liquids under gravity. If a flow stream containing a
mixture of two immiscible liquids is fed into a separating vessel
with upper and lower exits, the dense phase will leave through
the lower exit and the light phase will leave through the upper
exit (Fig. 1).
Figure 1. Gravity based liquid-liquid extraction in continuous flow.
It is important that the position of the liquid-liquid interface be
kept within the vessel, otherwise the liquids may exit the vessel
through the wrong channels. Due to material extraction between
phases, the flow rates of each phase (and therefore the position of
the interface) may vary. It is therefore desirable to acquire a
means of dynamic positional control. This requires a method of
determining the interface position. Sprecher and co-workers have
shown that differences in the electrical impedance of the aqueous
and organic phases can be used to achieve this in a continuous
flow system.10 Refractive index differences have also been used
in automated batch systems.11 In the growing context of camera-
enabled synthesis,12 we have been interested in the use of
computer-vision systems, together with inexpensive digital ‘web-
cams’, to facilitate interface location. Rather than using the
system to directly ‘see’ the liquid-liquid interface, we instead use
a coloured plastic float which, having a density between that of
the two phases, sits at the interface. With hue filtering, it is
straightforward to determine the position of the float in a digital
image stream. This system has been successfully employed for a
range of continuous flow transformations including hydrazone
formation, dithiane synthesis, alkene epoxidations,13 amino acid

2
Tetrahedron
diazotisations,14 ethylene-Heck reactions,15 and enamine
the freely available OpenCV22 libraries incorporated into a
Python23 script.
brominations.16 Recently, the Ley group reported the use of this
system in a multi-step flow synthesis of 2-aminoadamantane-2-
carboxylic acid, where computer-vision was also used to
dynamically measure liquid volumes in solution reservoirs.17 The
use of a gravity based separator in continuous flow was also
described by the Wirth group, although it was not reported
whether any positional control system was used.18 Herein, we
report our findings on the use of a computer-vision controlled
liquid-liquid extraction system in the continuous flow iodination
of a series of enaminones with N-iodosuccinimide,19 wherein the
succinimide by-product is efficiently removed using a solution of
aqueous alkali. We present data that suggest the use of a water
soluble component in the injected reaction solvent (7:3 CH2Cl2-
MeCN) leads to a significant perturbation of the interfacial
position during the liquid-liquid extraction process. This
perturbation was mitigated by the dynamic control system.
Additionally, we describe how the computer-vision system
tolerates significant colouration of the organic phase during the
extraction process.
Results and Discussion
The substrate enaminones 3a-c were synthesised by condensing
dimedone 1 with primary amines 2a-c at reflux in a 20:1 mixture
of toluene and ethanol (Scheme 1).20 Enaminones 3d-m were
synthesised in a similar manner.13 Our iodination studies began
with enaminone 3a using the flow apparatus setup shown in
Figure 2. The starting material and N-iodosuccinimide are
introduced into separate flow streams (which have the same flow
rate) via injection loops.
Figure 2. Flow system for enamine iodination
Whilst our previously reported script used the original OpenCV
API and required the VideoCapture library24 to access frames
which were then converted between PIL (Python Image
Library)25 and OpenCV formats, the script used in this work (see
ESI) uses the newer CV2 OpenCV API and does not require
these additional file conversions. Based on our previous
bromination studies using N-bromosuccinimide,13 we began with
neat dichloromethane as the reaction solvent. However, N-
iodosuccinimide proved to be significantly less soluble than N-
bromosuccinimide in this solvent and only concentrations of
0.025 M could be obtained. We quickly abandoned our
investigations with this solvent due to the low reaction rates and
the low material throughput. Changing to a binary solvent (7:3
CH2Cl2-MeCN) enabled much higher concentrations of N-
iodosuccinimide (0.077 M) to be used whilst still providing
complete phase separation, and efficient removal of succinimide,
following aqueous extraction. An immediate consequence of the
change of solvent, however, was that the float we were using for
neat dichloromethane, made by melting together 186 mg of
polyacetal (from a green Keck clip) and 29 mg of low density
polyethylene (from the plunger of a 1 mL disposable syringe),
was no longer suitable. When 3 mL of CH2Cl2-MeCN (7:3) was
extracted with 6 mL of water in a glass vial, this ‘float’ sank to
the bottom of the vessel, clearly indicating that some MeCN
remained in the organic phase, lowering its density. This was
rectified by using a different mixture of plastics. Experiments
determined that a float made by melting together 138 mg of
polyacetal (green Keck clip) with 64 mg of the low density
polyethylene (syringe plunger) was of a suitable density to
remain at the interface, both before and after extraction. The
quantities of each plastic contained in the floats are only
approximate as the method used to combine them (holding the
pieces together with tweezers and heating with a heat gun) led to
some mechanical loss of material. With a more accurate melting
technique, it should be possible to create floats of any required
density. Having obtained a suitable float, we continued with the
Scheme 1. Enaminone formation
The reagent loop is longer than the starting material loop, to
ensure that starting material always passes through the system in
the presence of reagent. The starting material and reagent streams
are mixed at a T-piece and pass into a residence-time loop where
they react at room temperature. The reaction mixture is then met
by a stream of aqueous extractant. Sodium thiosulfate ensures
that excess N-iodosuccinimide is reduced to succinimide and
potassium carbonate helps to extract the mildly acidic
succinimide into the aqueous phase. Efficient mixing of the
aqueous and organic phases is facilitated by passing the biphasic
stream through a small glass column containing several tiny
magnetic stirrer beads, sitting on the plate of a magnetic stirrer.21
Upon exiting the mixer, the phases rapidly settle back into plug-
flow and pass into the separating chamber where they separate
into bulk phases according to gravity. The computer-vision
system, incorporating an inexpensive webcam, monitors the
position of the interfacial float. Filtration of the digital images in
the webcam stream according to hue values allows the position of
the float, and therefore the liquid-liquid interface, to be
ascertained in a straightforward manner. In simple terms, a fall in
the level of the interface below the desired set point causes the
aqueous-out pump to increase its flow rate, thereby lifting the
interface level. If the interface level goes above the desired point,
the flow rate of the aqueous-out pump decreases, causing the
interface level to drop. The software system used was based on

3
iodination reactions. Both the N-iodosuccinimide and the starting
material were dissolved in the 7:3 CH2Cl2-MeCN mixture,
although these solutions were injected, via the loops, into flow
streams of neat dichloromethane (both at 1 mL min-1). With the
N-iodosuccinimide at 0.077 M, a concentration of 0.051 M was
used for the starting material 3a. An excess of N-iodosuccinimide
was used to facilitate complete and rapid reactions. An aqueous
extraction stream of Na2S2O3/K2CO3 (0.1 M) was fed into the
system at a rate of 4 mL min-1. A small number of experiments
revealed that an 11 mL reaction loop (5.5 min reaction time) was
sufficient to ensure complete conversion of 3a to iodoenaminone
5a at room temperature. The N-iodosuccinimide flow stream was
injected 20 seconds prior to the starting material. This
corresponds to an overlap of 20 seconds at the injection front and
64 seconds at the tail. The product stream was collected for 40
minutes. A very high yield of 5a (99%), which was analytically
1
pure by H and 13C NMR analysis, was obtained simply by
removing the solvent under reduced pressure. No drying step was
used. Liquid-liquid separation in batch, e.g. using a separating
funnel, typically leads to an organic phase visibly contaminated
with small droplets of aqueous liquid (held in place due to
surface forces) and generally necessitates the use of a dessicant.
As the mechanical agitation and settling processes are confined to
separate vessels, the flow extraction system does not suffer from
this problem. Although trace quantities of dissolved water were
presumably present before solvent removal on the rotary
evaporator, no visible water droplets were ever observed in the
product streams. Importantly, all of the succinimide, and excess
N-iodosuccinimide, had been efficiently removed by the in-line
liquid-liquid extraction system (Fig. 3). Using the same setup and
conditions, the iodination of a series of enaminones 3a-m led to
the efficient synthesis of the corresponding iodoenaminones 5a-
m in very high yield (Scheme 2).
Figure 3. A: 1H NMR spectrum of product 5a isolated after continuous flow
synthesis incorporating liquid-liquid extraction. B: 1H NMR spectrum of
product 5a isolated when the extraction unit is bypassed. C: 1H NMR
spectrum of succinimide alone. D: 1H NMR spectrum of N-iodosuccinimide
alone. All spectra obtained in CDCl3.
We were interested in quantifying how the colouration affected
the hue recognition during this reaction. For this purpose, the
iodination of 3f was repeated but all flow streams were halted
part way through (after 10 min.) whilst observations were made.
Figure 4 shows the feed from the digital webcam at this point.
Although the lower organic layer is very dark, the green float,
which happened to be situated towards the front of the separation
vessel, is still easily seen but it is clear that the brown colouration
has perturbed its appearance. Hue histograms of three regions of
the image stream (A, B and C) were obtained and are shown on
the right of the figure. Region A has a hue histogram practically
identical to that obtained for the float in clean solvent. Almost all
of the pixels are in the green region of the HSV colourspace with
a narrow distribution around a maximum of 75 (the hue values in
the OpenCV HSV images used have a range of 0-180 from red
through to red, see colourbar at the bottom of Figure 4). The hue
histogram for region B (the organic solution) is quite clearly in
the red/yellow hue region.
Effect of colouration on hue filtering
Although the majority of iodination reactions in Scheme 2 led to
colourless or pale yellow product streams, in the case of the
pyridyl compound 3f the reaction was accompanied by
significant yellow/brown colouration. Despite this, the tolerances
in the computer-vision algorithm were such that the position of
the green float could be discerned by the system throughout the
entire reaction.
Scheme 2. Continuous flow iodinations.

4
Tetrahedron
In region C, the colour of the solution has clearly affected the
selected when starting each run. In B2, many more pixels fall
within the wider tolerance limits of ± 20 hue units. In C2, the
tolerance limits of ± 30 lead to a similar situation, with the green
float being correctly identified and located. With tolerance limits
of ± 40, however, there is insufficient filtration of pixels (D2) and
the system fails to correctly locate the float, as can be seen from
the position of the turquoise circle in D1. With suitable tolerance
limits (± 20) the system was able to cope with the significant
colouration seen in the iodination of 3f.
observed hue of the float. A significant portion of the histogram
is now in the yellow region, with the most populated histogram
bin having a hue value of 58, a shift of 16 units compared with
region A. When a flow chemistry run is initiated, the user of the
system selects a portion of the image (i.e. the green float) from
which the system obtains a hue histogram and identifies the most
common hue. Subsequently, all images are filtered based on
whether pixels have matching hue values, within a specified
tolerance. Clearly, very narrow tolerance limits might affect the
ability of the system to recognise the green float in the event of
such colouration. Shown in Figure 5 are several images (A1, B1,
C1, D1) taken from the webcam feed during the paused
iodination of 3f, and the corresponding images (A2, B2, C2, D2)
following hue filtering with various hue tolerances.
Positional Control
The presence of MeCN in the reaction solvent raised the
possibility of significant mass transfer from the organic phase to
the aqueous phase during extraction. This would cause a
substantial increase in the volumetric flow rate of the aqueous
outlet stream following injection of the reactant solutions into the
flow streams of neat CH2Cl2. We were interested in the system’s
response to this perturbation and in order to gauge the potential
for interfacial transfer of MeCN during the extraction step, we
carried out a simple batch liquid-liquid partitioning experiment
(Fig. 6).
Figure 6. Illustration of the batch liquid-liquid extraction experiment using
an aqueous Na2S2O3/K2CO3 solution and CH2Cl2-MeCN solvent mixture.
Fig 4. Hue histograms (right) for selected regions of the web-cam feed (left),
during the halted iodination of 3f.
To 60 mL of CH2Cl2-MeCN (7:3, vol:vol) in a graduated
cylinder, 120 mL of the aqueous extraction solution (0.1 M
Na2S2O3/K2CO3) was carefully added without mixing. The
interface level remained at the 60 mL mark. The cylinder was
stoppered and the contents shaken vigorously for 5 minutes.
After settling, the volume of the lower organic phase was now 52
mL and that of the upper aqueous phase was 128 mL (the total
volume was unchanged at 180 mL). Unsurprisingly, a significant
amount of material had clearly passed from the organic phase to
1
the aqueous phase. H NMR analysis of the two organic solvent
mixtures before and after extraction indicated that the vol.-vol.
ratio of CH2Cl2-MeCN had changed from 70:30 to approximately
78:22. To measure the response of our in-line liquid-liquid
extraction system to such mass transfer, we carried out a run
using the same setup shown in Figure 2, where we recorded both
the position of the green float as well as the flow rate for the
aqueous-out pump. For simplicity, this was done without any
starting materials or N-iodosuccinimide, although the
Na2S2O3/K2CO3 (0.1 M) solution was still used as the aqueous
phase. The two injection loops were filled with the CH2Cl2-
MeCN (7:3) solvent mixture and injected into the flow stream of
neat CH2Cl2. The results are shown in Figure 7. Both the
interfacial position and the aqueous-out flow rate share the same
horizontal time axis but have separate vertical axes as indicated.
The interfacial position is displayed in terms of the fraction of the
vertical distance between the lower and upper bounds selected
during initialisation (corresponding to the 1 mL and 3 mL marks
on the syringe barrel that formed the separation vessel). The time
shown is relative to the injection of the CH2Cl2-MeCN mixtures
into the flow streams of neat CH2Cl2. To remove the high-
frequency oscillatory component from the raw data (shown in
Fig 5. Web-cam feed images during interrupted iodination of 3f (A1, B1, C1,
D1) and corresponding filtered images (A2, B2, C2, D2) using different hue-
tolerance values (± 10 means that pixels would be selected if they had hue
values within ± 10 units of the most populated hue value from the histogram
of the originally selected coloured object, in this case the green float). The
position of the centroid of the largest filtered ‘object’ is indicated by a
turquoise circle.
In A2, very few pixels in the image are between ± 10 hue units of
the originally selected green colour. The centroid of the largest
group of ‘correct’ pixels (after additional noise removal steps) is
indicated with a turquoise circle in A1 while the smaller blue
circles indicate the positions of the upper and lower bounds

5
light colours), a Butterworth low-pass filter26 was applied to the
two sets of data. The resulting lines (bold, Fig. 7) allow the
general trend to be more easily discerned. As can be seen,
following an initial delay as the CH2Cl2-MeCN mixture works its
way through the reaction loop and in-line mixer, there is a
noticeable lowering of the interface level at around 500 sec.
Supporting Information
NMR spectral data and spectra of compounds 3a-c, 5a-m, and
the Python control script are included in the ESI.
References
1
(a) Hessel, V.; Kralisch, D.; Kockmann, N.; Noel, T.; Wang, Q.
ChemSusChem 2013, 6, 746; (b) Brzozowski, M.; O’Brien, M.;
Ley, S. V.; Polyzos, A. Acc. Chem. Res. 2015, 48, 349; (c)
Pastre, J. C.; Browne, D. L.; Ley, S. V. Chem. Soc. Rev. 2013,
42, 8849; (d) Newman, S. G.; Jensen, K. F. Green Chem. 2013,
15, 1456; (e) Knowles, J. P.; Elliott, L. D.; Booker-Milburn, K. I.
Beilstein J. Org. Chem. 2012, 8, 2025; (f) Wiles, C.; Watts, P.
Chem. Commun. 2011, 47, 6512; (g) Webb, D.; Jamison, T. F.
Chem. Sci. 2010, 1, 675; (h) McQuade, D. T.; Seeberger, P. H. J.
Org. Chem. 2013, 78, 6384; (i) Wegner, J.; Ceylan, S.;
Kirschning, A. Adv. Synth. Catal. 2012, 354, 17; (j) Yoshida, J. I.
Chem. Rec. 2010, 10, 332; (k) Kobayashi, J.; Mori, Y.; Okamoto,
K.; Akiyama, R.; Ueno, M.; Kitamori, T.; Kobayashi, S. Science
2004, 304, 1305; (l) Gutmann, B.; Cantillo, D.; Kappe, C. O.
Angew. Chem. Int. Ed. 2015, 54, 6688; (m) Ley, S. V. Chem.
Rec. 2012, 12, 378; (n) Hartman, R. L.; McMullen, J. P.; Jensen,
K. E. Angew. Chem. Int. Ed. 2011, 50, 7502.
2
(a) Mueller, S. T. R.; Wirth, T. ChemSusChem 2015, 8, 245; (b)
Mastronardi, F.; Gutmann, B.; Kappe, C. O. Org. Lett. 2013, 15,
5590; (c) Dallinger, D.; Pinho, V. D.; Gutmann, B.; Kappe, C. O.
J. Org. Chem. 2016, 81, 5814; (d) Fukuyama, T.; Totoki, T.;
Ryu, I. Green Chem. 2014, 16, 2042; (e) Brancour, C.;
Fukuyama, T.; Mukai, Y.; Skrydstrup, T.; Ryu, I. Org. Lett.
2013, 15, 2794; (f) Gross, U.; Koos, P.; O'Brien, M.; Polyzos,
A.; Ley, S. V. Eur. J. Org. Chem. 2014, 6418; (g) Koos, P.;
Gross, U.; Polyzos, A.; O'Brien, M.; Baxendale, I.; Ley, S. V.
Org. Biomol. Chem. 2011, 9, 6903; (h) Maurya, R. A.; Park, C.
P.; Lee, J. H.; Kim, D. P. Angew. Chem. Int. Ed. 2011, 50, 5952.
(a) Hook, B. D. A.; Dohle, W.; Hirst, P. R.; Pickworth, M.;
Berry, M. B.; Booker-Milburn, K. I. J. Org. Chem. 2005, 70,
7558; (b) O'Brien, M.; Baxendale, I. R.; Ley, S. V. Org. Lett.
2010, 12, 1596; (c) O'Brien, M.; Taylor, N.; Polyzos, A.;
Baxendale, I. R.; Ley, S. V. Chem. Sci. 2011, 2, 1250; (d)
Mansur, E. A.; Ye, M.; Wang, Y.; Dai, Y. Chin. J. Chem. Eng.
2008, 16, 503; (e) Wada, Y.; Schmidt, M. A.; Jensen, K. F. Ind.
Eng. Chem. Res. 2006, 45, 8036; (f) Yang, L.; Jensen, K. F. Org.
Proc. Res. Dev. 2013, 17, 927; (g) Park, C. P.; Kim, D. P. J. Am.
Chem. Soc. 2010, 132, 10102.
Ley, S. V.; Baxendale, I. R.; Bream, R. N.; Jackson, P. S.; Leach,
A. G.; Longbottom, D. A.; Nesi, M.; Scott, J. S.; Storer, R. I.;
Taylor, S. J. J. Chem. Soc. - Perkin Trans. 1 2000, 3815.
O'Brien, M.; Denton, R.; Ley, S. V. Synthesis 2011, 1157.
E.g. at time of writing: Sigma-Aldrich: polymer-bound
thiosulfate (589977-25G) costs £ 6.43/mmol. Regular solid
sodium thiosulfate (217263-250G-D) costs £ 0.02/mmol. .
Lange, H.; Carter, C. F.; Hopkin, M. D.; Burke, A.; Goode, J. G.;
Baxendale, I. R.; Ley, S. V. Chem. Sci. 2011, 2, 765.
(a) Kralj, J. G.; Sahoo, H. R.; Jensen, K. F. Lab Chip 2007, 7,
256; (b) Castell, O. K.; Allender, C. J.; Barrow, D. A. Lab Chip
2009, 9, 388; (c) Atallah, R. H.; Ruzicka, J.; Christian, G. D.
Anal. Chem. 1987, 59, 2909; (d) Kolehmainen, E.; Turunen, I.
Chem. Eng. Process. 2007, 46, 834; (e) Adamo, A.; Heider, P.
L.; Weeranoppanant, N.; Jensen, K. F. Ind. Eng. Chem. Res.
2013, 52, 10802.
(a) Hornung, C. H.; Mackley, M. R.; Baxendale, I. R.; Ley, S. V.
Org. Proc. Res. Dev. 2007, 11, 399; (b) Sahoo, H. R.; Kralj, J.
G.; Jensen, K. F. Angew. Chem. Int. Ed. 2007, 46, 5704; (c)
Delville, M. M. E.; van Hest, J. C. M.; Rutjes, F. P. J. T.
Beilstein J. Org. Chem. 2013, 9, 1813; (d) Maurya, R. A.; Min,
K. I.; Kim, D. P. Green Chem. 2014, 16, 116; (e) Delville, M. M.
E.; Koch, K.; van Hest, J. C. M.; Rutjes, F. P. J. T. Org. Biomol.
Chem. 2015, 13, 1634; (f) Varas, A. C.; Noël, T.; Wang, Q.;
Hessel, V. ChemSusChem 2012, 5, 1703.
Figure 7. Data for the position of the interfacial float (red) and the flow rate
of the aqueous-out pump (blue). Raw data is shown in light colours while the
smoothed data (Butterworth low pass filter) is shown as a dark line.
This is consistent with an increase in the aqueous phase
volumetric flow rate and a corresponding decrease in the organic
phase volumetric flow rate, brought about by extraction of MeCN
between the two phases. This was confirmed by 1H NMR
analysis of the organic outlet stream, which had the same
CH2Cl2-MeCN ratio as in the batch extraction experiment shown
in Figure 6. The perturbation in the interface level is
accompanied by a corresponding increase in the aqueous-out
flow rate as the system tries to compensate. This continues for
some time until the position of the interface and the aqueous-out
flow rate begin to return to their original values. Clearly, the
results shown in Figure 7 reinforce the need for dynamic
positional control in order to keep the liquid-liquid interface
within the desired bounds. The ability of the system to respond to
perturbations becomes more critical when low organic phase
volumes are used in the separator as there is less ‘buffer’ volume
between the interface and the organic outlet. This is an important
consideration from a process standpoint as the use of a low
organic phase volume is desirable to minimise unwanted
dispersion effects.27 In addition, the ability to keep the organic
phase volume within desired bounds provides control over such
dispersion, leading to time and/or scale invariance.
3
4
5
6
7
8
Conclusions
The computer-vision controlled liquid-liquid extraction system
used in this work enabled the automated continuous-flow
iodination of a series of enaminones using N-iodosuccinimide as
the halogen source. The system efficiently extracted all of the
succinimide by-products from the product stream, affording
analytically pure compounds in high yield. The system was able
to cope well with significant colouration of the product stream. It
also responded well to the perturbation in the relative flow rates
of the two phases brought about by the interfacial transfer of
MeCN. We are currently investigating the application of this
system to a wide range of other reaction types. We are also
continuing our investigations into the perturbation of interface
position due to the extraction of miscible solvents into the
aqueous phase and will report our findings in due course.
9

6
Tetrahedron
Sprecher, H.; Payán, M.; Weber, M.; Yilmaz, G.; Wille,
10
G. J. Flow. Chem. 2012, 2, 20.
11 Maslana, E.; Schmitt, R.; Pan, J. J. Autom. Methods Manag.
Chem. 2000, 22, 187.
12 Ley, S. V.; Ingham, R. J.; O'Brien, M.; Browne, D. L. Beilstein J.
Org. Chem. 2013, 9, 1051.
13 (a) O'Brien, M.; Koos, P.; Browne, D. L.; Ley, S. V. Org.
Biomol. Chem. 2012, 10, 7031; (b) O’Brien, M.; Cooper, D. A.;
Mhembere, P. Tetrahedron Lett. 2016, 57, 5188.
14 Hu, D. X.; O'Brien, M.; Ley, S. V. Org. Lett. 2012, 14, 4246.
15 Bourne, S. L.; O'Brien, M.; Kasinathan, S.; Koos, P.; Tolstoy, P.;
Hu, D. X.; Bates, R. W.; Martin, B.; Schenkel, B.; Ley, S. V.
ChemCatChem 2013, 5, 159.
16 O'Brien, M.; Cooper, D. Synlett 2016, 27, 164.
17 Ingham, R. J.; Battilocchio, C.; Fitzpatrick, D. E.; Sliwinski, E.;
Hawkins, J. M.; Ley, S. V. Angew. Chem. Int. Ed. 2015, 54, 144.
18 Muller, S. T. R.; Murat, A.; Hellier, P.; Wirth, T. Org. Proc. Res.
Dev. 2016, 20, 495.
19(a) Ramesh, N. G.; Heijne, E. H.; Klunder, A. J. H.; Zwanenburg,
B. Tetrahedron 2002, 58, 1361; (b) Koutentis, P. A.; Loizou, G.;
Lo Re, D. J. Org. Chem. 2011, 76, 5793.
20 Gerasyuto, A. I.; Hsung, R. P.; Sydorenko, N.; Slafer, B. J. Org.
Chem. 2005, 70, 4248.
21 Gugger, R. E.; Mozersky, S. M., US 4054270, 1977; Vol.
4054270.
22 (a) Bradski, G. Dr. Dobbs J. 2000, 25, 120; (b)
http://opencv.org (last accessed 25/8/2016).
23 http://www.python.org/ (last accessed 25/8/2016).
24 http://videocapture.sourceforge.net/ (last accessed 25/8/2016).
25 http://www.pythonware.com/products/pil/
25/8/2016).
(last
accessed
26 (a) Butterworth, S. Exp. Wireless 1930, 7, 536; (b)
http://docs.scipy.org/doc/scipy-
0.14.0/reference/generated/scipy.signal.butter.html
(last
accessed 25/8/2016).
27 Tyson, J. F. Analyst 1990, 115, 587.

7
Highlights
Open-source technologies were harnessed to provide low-cost automation.
Computer-vision was used in a continuous-flow liquid-liquid extraction process.
Iodo-enaminones were synthesised in high yield and purity.