Available online at www.sciencedirect.

com

ScienceDirect
Energy Procedia 112 (2017) 217 – 224

Sustainable Solutions for Energy and Environment, EENVIRO 2016, 26-28 October 2016,
Bucharest, Romania

Assessment of global void fraction in a gas-liquid stirred vessel by

digital image processing
Andrei Dragomirescua*, Irina Pincovschia, Maria Miua
a
University Politehnica of Bucharest, Department of Hydraulics, Hydraulic Machinery and Environmental Engineering,
313 Splaiul Independentei, Bucharest 060042, Romania

Abstract

The main goal of the study presented in this paper is to propose a non-invasive method based on digital image processing to assess the
global void fraction for both the vortex and the air bubbles that form in a gas-liquid stirred vessel. Three impellers with different
geometries (circular, rectangular, and trapezoidal) were tested in a laboratory scale water tank for two aspect ratios of the water
column. The trapezoidal impeller has perforated blades, a secondary goal of the study being to investigate whether the perforations
could decrease the power demand of the impeller while providing a high global void fraction. The results obtained show that the
operation of the impellers is characterized by two critical rotational speeds, ncr,1 and ncr,2. The transition from the sub-critical (non-
aerated) to the super-critical (aerated) regime takes place at ncr,1. The vortex volume fraction increases up to ncr,2 and gets stabilized
beyond it. The evolution of the bubble void fraction can be described by the logistic function. When enough experimental data is
available, this function allows the estimation of a third critical speed beyond which the increase in bubble void fraction becomes
negligible. The usage of perforated blades did not bring a decrease but an increase in power demand. The method based on digital
image processing proved to be adequate for choosing the appropriate type of impeller and for optimizing new impeller designs in
applications involving gas-liquid stirred tanks.
© 2017
© 2017TheTheAuthors.
Authors. Published
Published by by Elsevier
Elsevier Ltd.Ltd.
This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of the organizing committee of the international conference on Sustainable Solutions for Energy
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
and Environment
Peer-review 2016.
under responsibility of the organizing committee of the international conference on Sustainable Solutions for Energy
and Environment 2016
Keywords: stirred vessel; global void fraction; vortex void fraction; bubble void fraction; digital image processing

* Corresponding author. Tel.: +4-074-056-9986; fax: +4-021-402-9865.

E-mail address :andrei.dragomirescu@upb.ro

1876-6102 © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the organizing committee of the international conference on Sustainable Solutions for Energy and Environment 2016
doi:10.1016/j.egypro.2017.03.1087
218 Andrei Dragomirescu et al. / Energy Procedia 112 (2017) 217 – 224

1. Introduction

Numerous industrial chemical processes, including water treatment, gas absorption, and aerobic fermentations,
take place between a gas phase and a liquid phase, by generating gas bubbles inside the liquid phase. For bubble
generation, spargers supplied with air by blowers are usually employed in bubble columns (BC). As an alternative,
in gas-liquid stirred tanks (G-L STs) bubbles are introduced by mechanical agitation into the liquid phase through
vortex entrainment from the free surface. In such a case, impellers are used to break up the vortex into bubbles.
For gas-liquid stirred tanks, the most important task is the generation of a high contact area between the gas and
liquid phases – which in turn allows for high mass transfer and improved reaction rates. G-L STs achieve high
specific areas (up to 100-500 m2/m3 [1]) by using an impeller to disperse the gas into the liquid in radial direction
(Rushton impeller, concave blade impeller, etc.) or axial direction (marine impeller, pitched blade impeller, etc.).
The flow induced by the impeller placed at the bottom of the tank generates a vortex, which is initiated at the free
surface of the liquid inside the tank and descends towards the impeller. When the agitation starts and as long as the
free surface shaped by the vortex does not reach the impeller, gas-liquid mass transfer occurs only through the free
surface. Bubbles are generated inside the tank as soon as the deformed free surface comes into contact with the
impeller. Until the vortex reaches the impeller, bubble formation does not occur and, consequently, no bursting takes
place at the free-surface. This is called the sub-critical or non-aerated regime. When the vortex reaches the impeller,
in the super-critical or aerated regime, the gas phase is dispersed inside the tank, thus providing better mixing.
Flow models with empirical correlations are generally used to describe the multiphase flow in the reactor. This
approach has provided a foundation for reactor design and scale-up. The empirical information available for
modeling usually describes global properties in the system of interest. The overall gas holdup or global void
fraction (GVF) is an important parameter required for flow regime identification as well as for modeling, design,
and scale-up of process equipment. It is defined as the fraction of the reactor dynamic volume occupied by the gas.
GVF and its radial profile govern liquid recirculation, flow pattern, mixing, and heat and mass transfer in bubble
column reactors and stirred tanks. In gas-liquid stirred tanks with vortex entrainment from the free surface, GVF is
given by the volume fractions of the vortex and of the bubbles.
There are two main challenges regarding multiphase stirred tank reactors: (i) scale-up and (ii) Computational
Fluid Dynamics Modeling [2]. Reactor scale-up – from laboratory scale to industrial scale reactor – is a big
challenge regarding multiphase reactors. By scaling up to larger geometric sizes, bubble, mixing, and kinetic lengths
and characteristic times do not scale in proportion.
Computational Fluid Dynamics (CFD) models have been developed to provide quantitative descriptions of flows
in multiphase STs. These CFD models require experimental validation. In multiphase STs, difficulties arise in
validating and improving the CFD codes. Rammohan et al. [3] and Khopkar et al. [4] have shown that current CFD
attempts fail to properly capture flows in gas-liquid STs. To empower the predictive capabilities of CFD, more
experimental data is required [5, 6].
GVF has been extensively studied experimentally with various measurement techniques. Some review articles in
the literature present the available measurement techniques for multiphase flows [7, 8]. There are two options when
considering a measurement technique: either invasive or non-invasive devices. The former are inserted into the reactor
so that they can interact with the flow, the latter are either mounted on or positioned closely to the reactor for
obtaining information about the flow without interacting with it. Among the non-invasive techniques one can mention:
laser techniques [9, 10], tomographic techniques [11, 12], physical tracers [13, 14], pressure drop measurements [15,
16], dynamic gas disengagement [17, 18], and visualization/high-speed photography techniques [19, 20].
The main goal of this work is to develop a non-invasive method based on high-speed photography and image analysis
techniques to quantify GVF in a laboratory scale stirred tank. The method proposed in this study, although not as accurate
as the laser technique, can offer quick and inexpensive information on GVF and interfacial area in an ST. This method can
be very useful when comparing different types of impellers with respect to their influence on aeration efficiency, where the
aeration efficiency is defined as oxygen transfer rate (from gas to liquid phase) per unit total power input [21].
In the present work, a laboratory scale stirred tank and three types of impellers were used in order to identify by
means of the proposed method the type of impeller and the operating conditions that maximize the two components
of GVF, namely the vortex and the bubble void fractions. A secondary goal of the study was to verify whether an
impeller with perforated blades could provide high bubble void fraction with low power demand. The independent
variables were the type of impeller, the impeller speed, and the height of the liquid phase inside the tank.
 Andrei Dragomirescu et al. / Energy Procedia 112 (2017) 217 – 224 219

2. Experimental set-up and working methods

The experiments were performed in a prismatic tank made of Plexiglas (Fig. 1), having a total height of 700 mm
and a square cross-sectional area. The prismatic form was chosen so that, when photographing the tank, no image
distortion appears because of the wall curvature. The side of the square, T, is of 300 mm. The tank was filled with
tap water up to a height H, measured under no agitation conditions. The water volume is Vt = HT 2. The ratio H/T
will be denoted further as aspect ratio. Experiments were carried out for two values of H: 300 mm and 400 mm. The
corresponding aspect ratios are H/T = 1 and H/T = 1.33 respectively. In all runs, the water temperature inside the
tank remained between 24°C and 25°C and the pressure at the free surface equaled the atmospheric pressure.
Three impellers with different geometrical forms – rectangular (RI), circular (CI), and trapezoidal (TI) – were
tested in order to evaluate their effect on the GVF inside the tank. Impellers RI and CI were made of 0.75 mm thick
ordinary aluminum sheet, while TI was made of a 0.75 mm thick perforated aluminum sheet with orifices arranged
in dense arrays that span over the entire surface of the blades. The trapezoidal form of the perforated impeller was
chosen to avoid scalloped edges. Sketches of the impellers are presented in Figure 2. As it can be seen, each impeller
has two blades that are flat for impellers CI and TI and slightly curved for impeller RI. The clearances C with respect
to the bottom of the tank were of roughly 40 mm.
All impellers were driven by a Heidolph RZR 2102 stirrer of 140 W, foreseen with a speed controller and a
torque transducer. The speed of the stirrer is adjustable over two scales, in the ranges 8–400 rpm and 40–2000 rpm.
During the experiments, various constant speeds were adjusted in a range from 280 rpm to 900 rpm, depending on
impeller type, in order to explore different flow regimes that occur inside the tank. For each operating point, the
power absorbed by the impeller was calculated with the formula
Sn M
P , (1)
30 100
where n and M are the values of speed in rpm and of torque in Ncm, respectively, as displayed on the stirrer LCD. The
specific power, defined as ratio between absorbed power and water volume inside the tank, P/V, was also calculated.
The method used to estimate the vortex void fraction, Hv, is based on the observation that the vortex can be
approximated to a stack of infinitely thin cylinders with diameters depending on water depth. These diameters can
be assessed by photographing the vortex and by measuring the distances between its left and right boundaries at
different water depths on the resulting photographs.
The photographs used for this study were taken by means of the photography setup presented in Figure 1. A
DSLR camera was mounted on a sturdy tripod in front of the tank, with the lenses set at a focal length of 70 mm, an
aperture of f/11, and a shutter speed of 1/1000 s. Black curtains were used so that the vortex and the bubbles appear
on the photographs as lighter areas on a dark background and to avoid reflexions on the front wall of the tank. A

Fig. 1. Sketch of the laboratory scale stirred tank and the photography setup.
220 Andrei Dragomirescu et al. / Energy Procedia 112 (2017) 217 – 224

Fig. 2. Tested impellers: a) circular, CI, b) rectangular, RI, c) trapezoidal, TI.

paper ruler was glued on the right edge of the tank to allow the estimation of the image scale. Series of five
photographs were taken for each operating point characterized by an impeller type, a constant aspect ratio, and a
constant speed. A total number of 42 operating points were investigated, for which 210 photos were taken and
subsequently processed. One reason for taking a series of only five photos for each operating point was the fact that
one of the processing steps was carried out by means of a cumbersome and time consuming procedure, as it will be
detailed below. Another reason was based on the hypothesis that the vortex dimensions do not change dramatically
in time at a given operating point. This hypothesis is, certainly, arguable and requires further investigation.
Although a number of five photos cannot be regarded as being enough to provide an accurate value of the mean
volume of each vortex, it was assumed that they still can offer a good assessment of that volume.
The photographs were cropped to the relevant areas comprised between the upper edge of the impeller and the
free surface of the water inside the tank. A photograph obtained with the setup and procedure described above is
presented in Figure 3a. It can be seen that both the vortex and the air bubbles are clearly visible and the image does
not suffer from motion blur.
The photographs were then manually processed with the graphical image manipulation program GIMP to extract
the vertical projections of the vortices. Separate white layers were created, on which the left and right contours of
the vortices were outlined with 1 pixel-thick curves and the areas between the outlines were filled with black colour.
Figure 3b depicts the result obtained for the vortex in Figure 3a. The vertical projection of the same vortex is
presented in Figure 3c together with a few characteristic dimensions.
Assuming that any horizontal cross-section of the vortex is circular, its diameter, d, is a function of time, t, due to
the continuous water movement, and of water depth, y. Under this assumption, at constant impeller speed, the vortex
volume has the expression
S h (t ) 2
4 ³0
V (t ) d ( y ) dy , (2)

Fig. 3. Example of vortex fill obtained for impeller RI operating at the aspect ratio H/T = 1.33 and the speed n = 380 rpm: a) original photo,
b) vortex fill with opacity reduced to 40%, superimposed on the original photo, c) vortex fill with dimensions for geometry calculations.
 Andrei Dragomirescu et al. / Energy Procedia 112 (2017) 217 – 224 221

Fig. 4. Examples of areas cropped immediately above the runner, processed for the estimation of bubble void fraction in the super-critical regime,
at the aspect ratio H/T = 1.33 and the speed n = 380 rpm.

where h(t) is the vortex height that also depends on time as long as the vortex does not reach the impeller. The
height becomes constant when the speed is high enough to continuously keep the vortex in contact with the impeller.
The dependence of h on time is introduced so that relation (2) remains valid at any impeller speed.
For each vortex, an average volume was estimated based on the set of five photographs taken for that vortex. The
vertical projections of the vortex were exported as binary (only black and white) images. A Python script was used
to count the number of black pixels on each pixel row of each image. The resulting number is the vortex diameter
expressed in pixels at that position. The vortex volume, expressed in cubic pixels, can be estimated with the formula
S N 2
Vj ¦ d i 'y ,
4 i1
(3)

where N is the number of pixel rows of the image, 'y is the height of a pixel row, i is the row index, and j is the
image index in the set of five images. Taking into account the fact that each pixel row has a height of 1 pixel and
introducing the image scale, s, expressed in meters per pixel, the average vortex volume, expressed in m3, is

¦ j 1V j ,
1 5
V s3 (4)
5
Finally, the vortex void fraction was calculated as

H v V / Vt , (5)

where Vt is the water volume inside the tank.

In order to estimate the bubble void fraction, Hb, two surfaces of equal areas were cropped from the left and right
of each image. The areas were chosen as large as possible, provided that they do not contain any part of the vortex,
but only bubbles. Each area was converted to a binary image by thresholding, so that the bubbles appeared
completely white on a black background. The bubble void fraction was estimated as the ratio between total area of
the bubbles in the image and image area, both expressed in square pixels. The total area of the bubbles was obtained
by counting the white pixels on the two processed areas. All operations – cropping, thresholding, and counting of
white pixels – were performed with another Python script. As an example, Figure 4 presents portions of two
processed areas, cropped from immediately above the runner.

3. Results and discussion

In diagrams presented in this paper, black symbols mark the critical conditions – critical speed ncr,1, critical
vortex void fraction, Hv,cr,1, and critical bubble void fraction, Hb,cr,1 – corresponding to the transition from the sub-
222 Andrei Dragomirescu et al. / Energy Procedia 112 (2017) 217 – 224

Table 1. Values of the first and second critical speed depending on impeller type and aspect ratio.

H/T First critical speed, ncr,1 (rpm) Second critical speed, ncr,2 (rpm)

impeller CI impeller RI impeller TI impeller CI impeller RI impeller TI

1 600 300 459 738 358 532

1.33 691 348 571 835 397 665

critical to the super-critical regime. ncr,1 will be further denoted as first critical speed. It represents the minimum
speed at which it is visually observed that the vortex comes into contact with the impeller, yet no bubble bursting
occurs. The values of ncr,1 are presented in Table 1 for each impeller and aspect ratio.
Figure 5 shows that, for some of the impellers and aspect ratios investigated, Hv increases as the impeller speed is
increased, as long as the speed is lower than a threshold value. This is visible for all impellers and aspect ratios
investigated, except for impeller TI at H/T = 1.33. The threshold value will be further referred to as second critical
speed, ncr,2. For n > ncr,2 the vortex void fraction does not increase any longer. Instead, it gets stabilized or even
decreases slightly regardless of impeller speed. The increase seems to be linear, as indicated by the sloped lines
obtained by fitting the data up to ncr,2 with linear functions. The method for obtaining the values of ncr,2 is detailed
below. In the particular case of impeller TI at H/T = 1.33, additional data is required in order to draw a clear
conclusion. However, the low maximum power of the stirrer (140 W) did not allow the increase of the speed beyond
700 rpm in order to get more data.
Figure 6 presents the effect of impeller speed on bubble void fraction, Hb. As expected, the void fraction increases with
increasing speed, its profile being influenced by the flow regime. In the subcritical regime, the void fraction increases only
slightly with impeller speed. At the beginning of the super-critical regime, the increase becomes very steep. However, it is
clear that the bubble void fraction cannot increase indefinitely. Instead, it is expected for Hb to reach a saturation (i.e.
maximum) value, Hbs, with increasing speed. Based on this observation, the regression curves in Figure 6 were obtained by
fitting the data with the logistic function, which describes population growth (in this case, the growth of the bubble
population). The function has three fitting parameters: the abscissa of the curve midpoint, the maximum value of the
curve, and the steepness of the curve. A careful examination of the data points plotted in Figures 5 and 6 suggested that the
abscissa of the midpoint of the logistic function is close to the threshold value of the impeller speed, ncr,2, observed
previously for Hv. Therefore, in the study presented here, the logistic function was written in the following form:
H bs
Hb . (6)
1  exp( k (n  ncr , 2 ))

The steepness of the curve, k, is the slope at ncr,2. The values of ncr,2 were obtained by fitting the data with function (6).
Their values are presented in Table 1 and are marked with vertical dashed lines in Figures 5 and 6.

Fig. 5. Vortex void fraction depending on impeller speed and aspect ratio for a) impeller CI, b) impeller RI, and c) impeller TI.
 Andrei Dragomirescu et al. / Energy Procedia 112 (2017) 217 – 224 223

Fig. 6. Bubble void fraction depending on impeller speed and impeller type for a) H/T = 1 and b) H/T = 1.33.

Fig. 7. Bubble void fraction depending on specific power and impeller type for a) H/T = 1 and b) H/T = 1.33.

Figure 7 presents the bubble void fraction depending on specific power and type of impeller for the two aspect
ratios investigated. The data was fitted using another expression of the logistic function, in which n was replaced
with P/V. The diagrams show an interesting behaviour of the impellers, that seem to be influenced by the presence
or absence of perforations on the blades. In case of impellers RI and CI, that have normal (i.e. unperforated) blades,
for the sub-critical regime, the impeller geometry does not have a significant influence on Hb. In the super-critical
regime, impeller CI is more efficient than impeller RI. It can be noticed that impeller TI, with perforated blades,
requires a much higher specific power input to generate a given bubble void fraction, throughout the entire operating
range investigated. It can be also noticed that, when compared to impellers RI and CI, the efficiency of impeller TI
deteriorates faster as the aspect ratio increases.

4. Conclusions

The main purpose of the present study was to develop a non-invasive method based on image analysis techniques
to quantify the global void fraction by assessing the vortex and bubble void fractions in gas-liquid stirred tanks,
which have numerous applications in water treatment, gas absorption, aerobic fermentation, etc. A secondary
purpose was to verify whether an impeller with perforated blades could decrease the power demand while
maintaining a bubble void fraction comparable with that produced by usual impellers having unperforated blades.
The results obtained with the proposed method show that the operation of an impeller inside a stirred water tank
is characterized by two critical speeds: ncr,1 and ncr,2. The former is the speed that separates the sub-critical and
super-critical regimes. The latter represents the speed beyond which the vortex void fraction has the tendency to
stabilize. It was also found that the dependence of the bubble void fraction on impeller speed can be described by
the logistic function. When enough experimental data is available, the function could be used to estimate a third
224 Andrei Dragomirescu et al. / Energy Procedia 112 (2017) 217 – 224

critical speed, ncr,3, beyond which a further increase in speed does not produce any significant increase in bubble
volume fraction. From the practical point of view, ncr,3 should be regarded as the maximum operating speed of an
impeller in a gas-liquid stirred tank. However, this assumption requires further experimental research.
The bubble void fraction as a function of specific power is influenced by the presence or absence of perforations on
the blades. In case of impellers with unperforated blades, the impeller geometry does not have a significant influence in
the sub-critical regime. As far as the impeller with perforated blades is concerned, the bubble void fraction was found
to be significantly lower than that of the other two impellers. Hence, an impeller with perforated blades cannot provide
high bubble void fraction with low power demand.
Despite not being extremely accurate, the proposed photographic method and its accompanying image processing
technique described in this paper, proved to be able to offer inexpensive information that allowed to compare the
performance of the impellers from the point of view of obtaining a high GVF value with a low specific power input.
The method can be used to choose the appropriate impeller type for a specific application involving water stirred
tanks or in the first stages of optimizing a new impeller design. Further studies should focus on the usage of
powerful edge detection algorithms to automatically find the contours of the vortices, even in the presence of
bubbles. A significantly larger number of photographs could be processed in this way, possibly providing more
accurate values of the mean volumes of the vortices in a short time.

References

[1] Middleton JC. Gas-liquid dispersion and mixing. In: Harnby N, Edwards MF, Nienow AW, editors. Mixing in the Process Industries. Oxford:
Butterworth-Heinemann; 1985.
[2] Mueller S. Optical Measurements in gas-liquid stirred tanks. All Theses and Dissertations (ETDs). Paper 253; 2009.
[3] Rammohan, AR. Characterization of single and multiphase flows in stirred tank reactors. Ph.D Thesis, Washington University in St. Louis; 2002.
[4] Khopkar AR, Rammohan AR, Ranade VV, Dudukovic MP. Gas liquid flow generated by a Rushton turbine in a stirred vessel: CARPT/CT
measurements and CFD simulations. Chem Eng Sci 2005; 60:2215-2229.
[5] Scargiali F, D’Orazio A, Grisafi F, Brucato A. Modeling and simulation of gas-liquid hydrodynamics in mechanically stirred tanks. Trans
IChemE 2007; 85(A5):637-646.
[6] Sommerfield M, Decker S. State of the art and future trends in CFD simulation of stirred vessel hydrodynamics. Chem Eng Technol 2004;
27(3):215-224.
[7] Boyer C, Duquenne AM, Wild G. Measuring techniques in gas-liquid and gas-liquid-solid reactors. Chem Eng Sci 2002; 57:3185-3215.
[8] Yang GQ, Du B, Fan LS. Bubble formation and dynamics in gas-liquid-solid fluidization – a review. Chem Eng Sci 2007; 62:2-27.
[9] Montante G, Paglianti A, Magelli F. Experimental analysis and computational modeling of gas-liquid stirred vessels. Trans IChemE 2007;
85(A5):647-653.
[10] Simonnet M, Gentric C, Olmes E, Midoux E. Experimental determination of the drag coefficient in a swarm of bubbles. Chem. Eng. Sci., 62,
858-866.
[11] Ford JJ, Heindel TJ, Jensen TC, Drake JB. X-ray computed tomography of a gas-sparged stirred-tank reactor. Chem Eng Sci 2008; 63:2075- 2085.
[12] Hubers JL, Striegel AC, Heindel TJ, Gray JN, Jensen TC. X-ray computed tomography in large bubble columns. Chem Eng Sci 2005;
60:6124-6133.
[13] Gupta P, Al-Dahhan MH, Dudukovic MP, Mills PL. A novel signal filtering methodology for obtaining liquid phase tracer responses from
conductivity probes. Flow Meas Instrum 2000; 11:123–131.
[14] Vrabel P, van der Lans RGJM, Luyben KCAM, Boon L, Nienow AW. Mixing in large-scale vessels stirred with multiple radial or radial and
axial up-pumping impellers: modeling and measurements. Chem Eng Sci 2000; 55:5881-5896.
[15] Joshi JB, Patil TA, Ranade VV, Shah YT. Measurement of hydrodynamic parameters in multiphase sparged reactors. Rev Chem Eng 1990;
6(2-3):73-227.
[16] Vial C, Camarasa E, Poncin S, Wild G, Midoux N, Bouillard J. Study of hydrodynamic behaviour in bubble columns and external loop airlift
reactors through analysis of pressure fluctuations. Chem Eng Sci 2000; 55:2957-2973.
[17] Camarasa E, Vial C, Poncin S, Wild G, Midoux N, Bouillard J. Influence of coalescence behaviour of the liquid and of gas sparging on
hydrodynamics and bubble characteristics in a bubble column. Chem Eng Proc 1999; 38:329-344.
[18] Lee DJ, Luo X, Fan LS. Gas disengagement technique in a slurry bubble column operated in the coalesced bubble regime. Chem Eng Sci
1999; 54:2227-2236.
[19] Kazenin DA, Chepura IV, Petrov IA, Zhavoronkov VA. Hydrodynamics, mass transfer, and power consumption in air-core stirred tank
reactors. Theor Found Chem Eng 2008; 42(2):118-124.
[20] Martin M, Montes FJ, Galan MA. Bubbling process in stirred tank reactors I: agitator effect on bubble size, formation and rising. Chem Eng
Sci 2008; 63:3212-3222.
[21] Measurement of Oxygen Transfer in CleanWater, American Society of Civil Engineers Standards ASCE/EWRI 2-06, 2006.