(612f) Motion-Based Multiple Object Tracking of Ultrasonic-Induced nucleation: A Case Study of L-Glutamic Acid

Zhenguo Gao ^1,2 Dan
Zhu ^1,2 Yuanyi Wu ¹ Sohrab Rohani ¹* Junbo Gong ², Jingkang Wang ²

1. The
University of Western Ontario, Department of Chemical and Biochemical
Engineering, London, Ontario N6A 5B9, Canada

2. Tianjin
University, School of Chemical Engineering and Technology, Tianjin, 300072, P.
R. China

Keywords: Ultrasonic irradiation, Nucleation, Motion-based multiple
object tracking
Introduction

Nucleation from
solution has aroused great interest over the past decades.
Scientists have considered atomistic levels on the order of 10^-10 m
and time scales in the order of 10^-13 s to unravel the nucleation
mechanism with the help of advanced instruments¹ . However, in
industrial crystallization, there is still a limitation in monitoring
nucleation at the initial stage of critical nucleus formation² . Existing measuring
techniques used to monitor nucleation include the focused
beam reflection measurement (FBRM), turbidity, ultrasonic
velocity, electrical conductivity and light transmittance measurement of
bulk solution³ . The accuracy of the monitoring devices affects the measurement of
nucleation kinetics. In order to improve the accuracy, progress has been made
based on high-speed imaging, multivariate statistical process monitoring charts,
etc^4,5 . Recently, motion-based
multiple object tracking (MMOT) has been widely used in computer vision
field. The MMOT method detects moving objects in a video stream,
predicts their locations in the next frame, and records the object count
automatically, which shows its potential application in monitoring
the occurrence and disappearance of nuclei in a bulk solution.

External factors such as ultrasound,
microwave, magnetic field and electric field have been studied to control
crystal nucleation, which exhibit effectiveness in some cases^6,7 .
Ultrasonic irradiation creates sequential compression and expansion of the solution,
which leads to bubble formation and growth. Finally, the bubbles collapse and
release energy, promoting nucleation within a short induction time and at a
lower supersaturation level. Parameters, e.g. ultrasound power, frequency, with
or without pulse, have significant influence on nucleation process and crystal
qualities^8,9 . In this study, the
MMOT was introduced in the crystallization area, for the first time, to monitor
the nucleation of L-glutamic acid. Results showed that ultrasonic irradiation
shortened the induction time dramatically. The MMOT-based device showed higher
accuracy and precision compared to the FBRM technology as the bubbles were
transparent in the MMOT technology, but considered as nuclei by the FBRM.   
Materials
and Experiments

L-glutamic acid (LGA) was used as model compound. An ultrasonic processor was inserted into a
double jacketed crystallizer to generate ultrasound. During nucleation process,
a waterproof USB-based micro-camera (Magnification: 1-300X) was used to record
a video stream to monitor the nucleation process as shown in Figure 1(a). A
home-designed vial adaptor with a solution flow channel (Figure 1b) was printed by a 3D printer
and connected to the end of the micro-camera to allow
capture crystal images as the slurry flowed through the adaptor. Figure 2 shows
the experimental setup. In the first step, clear solutions were prepared in the
crystallizer at 80℃. Afterwards, nucleation temperature was reached as
soon as possible by switching the 3-way valve to the cooling refrigerated
circulator. The temperature was kept constant to study the
nucleation kinetics until the crystals appeared in the video stream. The images recorded by micro-camera
were parsed by MMOT in MATLAB.

Figure
1. (a) Portable USB-based micro-camera setup; (b) 3D print vial adaptor.

Figure
2. (a) Schematic of experimental setup; (b) Experimental setup picture. Results and Discussions

Different
conditions of solution concentration (30 g/L and 40 g/L) and nucleation temperature
(35℃ and 45℃) were
tested. The tracking model is solely based on motion,
in which the background is subtracted and a Kalman filter is used to predict
the assigned track’s location in next frame. The nuclei count in MMOT was smoothed by a moving average (MA) model
as shown in equations 1 and 2.

MA =
 =                           
(1)

 MA
_current = MA _prev +                                              
(2)

Where c
indicates the nuclei counts along
the time series and MA _prev is the previous averaged
value. The moving window n was optimized to be 30 which tightly
fitted the nuclei counts. The time at which a
continuous increase in the micro-camera counts was sensed was taken as the
onset of nucleation. The induction time was calculated as the period between
the time the ultrasound processor was switched on and the onset of nuclei
occurrence. Under the ultrasonic irradiation at 40 g/L and 35℃, Figure 3
and Figure 4 show the results at 38 s and 95 s of same process after the
ultrasonic processor was started.

Figure
3. Detection results at 38 s with
ultrasonic irradiation (14 W). (a) In-situ plot of nuclei counts
versus time based on MMOT; (b) Plot of nuclei counts against time of
the whole process. The blue line and black line are raw data and smoothed data.
(c) Detected objects marked in an original video. (d) Detected
objects marked in the video after background subtraction.

Figure
4. Detection results at 95 s with ultrasonic
irradiation (14 W). (a) In-situ plot of nuclei counts
versus time based on MMOT; (b) Plot of nuclei counts against time of
the whole process. The blue line and black line are raw data and smoothed data.
(c) Detected objects marked in an original video. (d) Detected
objects marked in the video after background subtraction.

As shown in Figure
3c and 4c, only one object had been detected, which indicates no nucleation
occurred before 38 s, while large quantities of
objects had been detected after 95 s. A continuous increase of nuclei counts
was detected after 72 s in Figure 4a, with the constant nucleation occurring in
raw image (Figure 4c), which confirmed the onset of nucleation at 72 s.
Comparing Figure 4a with Figure 3a, it can be concluded that the threshold of
background noise is around 5, which can be ignored before the continuous
increase of nuclei count. At least six experiments were repeated under the same
condition and average results were reported in Table 1.

Table
1. Comparison of induction time and crystal form of
LGA with/without ultrasonic irradiation.

Note: Standard deviation was calculated by , and X
means average of all (i=1, 2, 3…, n).

There is a sharp decrease in induction time compared with the
nucleation kinetics without ultrasonic irradiation ¹⁰. For a
rising of nucleation temperature, the induction time is longer. Ultrasonic irradiation shows
potential application to promote nucleation in crystallization process design.  Experimental results show the measurement accuracy based on MMOT is
much better than FBRM as the latter does not distinguish between the nuclei and
the bubbles generated by ultrasound. However, limitation exists in MMOT method,
which is similar to the PVM (Particle Vision and Measurement) technology, the detection limit of objects
count is around 100 and the occurrence of multiple crystals can affect the
accuracy. Conclusions

In this study, the
MMOT method was, for the first time, introduced in the crystallization field to
monitor nucleation process. Results show better accuracy compared with other
methods such as the FBRM technology. The nuclei count measured by MMOT, which
is different from the FBRM count, can be used as a parameter to control
nucleation in designing crystallization processes. Based on the MMOT
technology, potential applications can be developed to measure solubility,
metastable zone width, particle size and even solution-mediated
polymorphic transformation.  
Reference

1.         J. Zhang, P. Chen, B. Yuan, W. Ji, Z. Cheng, and X. Qiu, Science,
2013, 342, 611–4.

2.         A. S. Myerson,
Handbook of industry crystallization, Butterworth-Heinemann, 2002.

3.         C. Brandel and
J. Horst, Faraday discussions, 2015, 179, 199–214.

4.         S. Bauerecker,
P. Ulbig, V. Buch, L. Vrbka, and P. Jungwirth, J Phys Chem C, 2008, 112,
7631–7636.

5.         L. Simon, K.
Oucherif, Z. Nagy, and K. Hungerbuhler, Chem Eng Sci, 2010, 65,
4983–4995.

6.         S. Jiang,
University of Leeds, 2012.

7.         D.-C. Yin, Prog
Cryst Growth Ch, 2015, 61, 1–26.

8.         S. Bhangu, M.
Ashokkumar, and J. Lee, Cryst Growth Des, 2016, 16, 1934–1941.

9.         M. Jiang, C.
Papageorgiou, J. Waetzig, A. Hardy, M. Langston, and R. Braatz, Cryst Growth
Des, 2015, 15, 2486–2492.

10.       Schöll, Vicum, Müller,
and Mazzotti, Chem Eng Technology, 2006, 29, 257–264.