Professional Documents
Culture Documents
Measurement and CFD Simulation of Single-Phase Flow in Solvent Extraction Pulsed Column
Measurement and CFD Simulation of Single-Phase Flow in Solvent Extraction Pulsed Column
www.elsevier.com/locate/ces
Received 5 July 2005; received in revised form 5 October 2005; accepted 10 October 2005
Available online 15 December 2005
Abstract
A CFD (computational fluid dynamics) model of a solvent extraction pulsed column has been developed and run with a single water phase.
The results are compared with experimental measurements taken on a pilot scale column using PIV (particle image velocimetry).
The pulsed column investigated had disk–doughnut internals and was operated under pulsing intensities ranging from 10 to 32.5 mm/s. PIV
measurements of velocity were used to validate the CFD model and to characterise the pulsing flow of a single phase through the column. The
CFD modelling was performed for the same geometry and operating conditions using a 2D computational grid and a low Reynolds Number
k– turbulence model. An improved velocity prediction was achieved by adding a gap between the doughnut internal and the pulsed column
wall. The combined measurements and predictions give insight into the effect of the geometry internals on the flow hydrodynamics in the
pulsed column.
䉷 2005 Elsevier Ltd. All rights reserved.
Table 1
Operating conditions of single-phase pulsed column for PIV measurements
Trial number Throughflow condition Pulsing amplitude Pulsing frequency Pulsation intensity Flux
(mm) (Hz) (mm/s) (m3 /m2 /hr)
1 No-flow 10 2 20 —
2 20 0.5 10
3 0.75 15
4 1 20
5 26 0.25 6.5
6 0.5 13
7 0.75 19.5
8 1 26
9 1.25 32.5
10 30 0.67 20
11 20 1 20 —
12 Down-flow (2 L/min) 20 1 20 15.3
13 Up-flow (2 L/min) 20 1 20 15.3
outside the cylindrical glass column to minimise the effect been made in the absence of through-flow, with one up-flow
of optical distortion at the curved surface. The positions of and one down-flow case also studied (see Trial 12 and 13 in
tracer particles illuminated by the light sheet were recorded at Table 1).
a frame rate that was a multiple of the pulsation frequency of The LDV (laser doppler velocimetry) measurements were
the liquid. An electrical encoder was installed on the crank of used as a crosscheck of the PIV results and the CFD sim-
the pulsator to allow selection of the single crank angle trigger ulations for Trial 4. LDV has been reported extensively in
for the PIV synchronisation system. This ensured that images literature for other systems (Fei et al., 2000) and is an ex-
could be taken repetitively in a sequence to a known pulsator tremely robust technique requiring no calibration. It is most
crank angle. suitable as a method for accurately determining turbulence
The field of view of the CCD camera was 80 × 64 mm2 characteristics (such as Reynolds stresses) of steady flows
using 1280 × 1028 pixels of CCD array. The smallest resolv- but is used here to cross-check against the PIV. LDV is
able length scale of the PIV set-up, which is the real length slow compared to PIV for velocity measurement which is
of each pixel, equals 62.5 m. The interrogation windows why the PIV was the preferred method for investigating the
were 64 × 64 pixels (4 × 4 mm2 ), with a 50% overlapping velocity field.
between consecutive interrogation cells providing a velocity
vector spacing of 32 pixels (2 mm). Each instantaneous 2D
velocity measurement contained a total of 1209 vectors. The 2.3. Computational method
seeding particles used were TSI silver-coated hollow glass
beads with mean diameter of 14 m and relative density of The experimental pulsed column geometry has been mod-
1.65. The relaxation time of the particles was 18 s, which is elled using an axi-symmetric 2D geometry. This should be an
negligible compared to the separation time of 1–3 ms between excellent approximation particularly away from the upper and
pairs of images used for computation of particle displace- lower ends of the column. The axi-symmetric modelling as-
ments. The uncertainty of pixel displacement in the measure- sumption of the flow field is valid since the large number of axi-
ment was about 0.2 pixel within the interrogation window of symmetric internal plates present in the column means that any
64 × 64 pixels. Therefore the error of PIV measurements is variations in velocity with angular position around the column
about 0.31%. present at the top and bottom of the column will be damped out
Measurements were taken over a period of approximately for sections away from the top and bottom. Furthermore, the
100 cycles, and the flow fields averaged for each of 8 phases tangential component of flow is negligible when compared to
within a cycle. In this way the effects of turbulence, which mean the radial and axial components (Aoun Nabli et al., 1998). The
that the detailed flow at each cycle is different, were averaged model computational domain extends over three disk–doughnut
out. For the preliminary single-phase flow field investigation, sections in which the single-phase flow is considered to be typ-
the PIV measurements were carried out under the experimental ical of the flow over most of the column height. Although there
operating conditions in Table 1. These conditions correspond will be entrance regions at the top and bottom of the column, it is
to the range of pulsation intensity for which the column can expected that the flow will be the same in most of the sections of
be operated in the mixer-settler or emulsion mode according to the column.
the experimentally determined regime map and flooding curve The computational domain was terminated at top and bot-
(Logsdail and Slater, 1991). Most of the measurements have tom by a 1 mm thick section of a disk, and the associated flow
J.M. Bujalski et al. / Chemical Engineering Science 61 (2006) 2930 – 2938 2933
passage between the disk and the wall, at which inlet/outlet flow U = Af sin(2f t). (7)
boundary conditions must be defined. The height of the com- At the top boundary (the “outlet”) pressure was set to be uni-
putational domain was 156 mm whilst its radius was 50 mm. form, with zero normal gradients for the other variables (i.e.,
To model the pulsed column an axi-symmetric uniform grid of Dirichlet condition) whereas, when the flow was entering the
1 mm cell edge length was used with 7800 computational cells computational domain, the velocity was set to be uniform and
as shown in Fig. 2. equal to Eq. (7). The lower boundary (the “inlet”), velocity was
To simulate the disks and doughnut internal plates, 2 mm set to be uniform and equal to Eq. (7) at all times. The wall
thick walls were inserted in the geometry at correct heights. boundary conditions for the low Reynolds number k– turbu-
In the simulation only half the 6 mm diameter shaft holding lence model were set to no-slip with the kinetic energy and
the disks was modelled, as it was positioned on the axis of energy dissipation set to 0.
symmetry, running from the bottom to the top of the simulated The oscillation in the flow boundary condition required the
geometry. The properties of the single phase were those of water use of a transient simulation. The number of time steps used
with a density of 1000 kg/m3 and viscosity 0.001 kg/ms. The for each pulsation period was from 25 to 80 and the number
Reynolds number for the pulsating flow, ReP , can be defined of iterations per time step was from 35 and 100, respectively.
as (Ni et al., 2003): No difference was found in the results in the two different
simulation procedures. The time for the simulations to reach
D(f A) periodic steady state was 15 oscillation periods.
ReP = . (1)
3. Results and discussion
For the case with frequency, f, equal to 1 Hz and peak-to-peak
amplitude, A, equal to 20 mm, ReP = 2000. 3.1. PIV validation
The general purpose CFD code, CFX 4, was used as the
basis for the model development. The conservation of mass LDV was used to verify the PIV results at a selected mea-
and momentum were determined using the Reynolds averaged surement point A over the period of a whole oscillation. The
2934 J.M. Bujalski et al. / Chemical Engineering Science 61 (2006) 2930 – 2938
Mean Velocity Vectors at Time Phase - 0 Degree Mean Velocity Vectors at Time Phase - 45 Degree
Mean Velocity Vectors at Time Phase - 90 Degree Mean Velocity Vectors at Time Phase - 135 Degree
135°
90°
1m/s 1m/s
Fig. 4. Flow field comparison between PIV measurement and CFD predictions at pulsator phases 0◦ –135◦ .
in the middle of the campaign, the gap was slightly different circuiting and possibly the creation of a fast ‘jet’ flow along
for the throughflow cases. Closer to centreline all sets of mea- the column wall. This additional vertical flow will reduce size
surements are quite similar as the flow is mostly radial in this of the generated vortex, thus changing the local flow and shear
region, as shown also in Fig. 5. rate in the column.
3.4. Effect of the gap between doughnut edge and column wall 3.5. Pulsation intensity
In Figs. 6 and 7 the main discrepancy between the PIV and Fig. 8 shows the comparison of the maximum velocity (up-
the CFD is in the flow near the column wall. It was observed that ward and downward) at point A as a function of pulsation in-
in the experimental set-up the doughnut rings did not fit tightly tensity for trials 1 to 11. The results show the close agreement
against the column wall and a gap of varying size was present between the PIV and the CFD results for the downward velocity
in individual doughnut configurations. The CFD geometry was in the pulsation cycle, but there is an over prediction of the up-
modified to include a gap of 2 mm between the column wall and ward velocity component at this point. This discrepancy could
the doughnut edge. The CFD results for the modified geometry be attributed to the turbulence model used as supported by sim-
in Figs. 6 and 7 give a better prediction of the axial velocity ilar work of Fei et al. (2000) who found a similar discrepancy
than the original CFD results without the 2 mm gap. between their CFD modelling using the k– turbulence model
The gap between the wall and the doughnut could become with their LDV measurements. The results indicate that the ve-
significant in the scale-up of the pulsed column, leading to short locity component of the pulsed column can be scaled up with
2936 J.M. Bujalski et al. / Chemical Engineering Science 61 (2006) 2930 – 2938
Mean Velocity Vectors at Time Phase - 180 Degree Mean Velocity Vectors at Time Phase - 225 Degree
180° 225°
1m/s 1m/s
Mean Velocity Vectors at Time Phase - 270 Degree Mean Velocity Vectors at Time Phase - 315 Degree
270°
1m/s 1m/s
315°
CFD Prediction PIV Measurement CFD Prediction PIV Measurement
Fig. 5. Flow field comparison between PIV measurement and CFD predictions at pulsator phases 180◦ –315◦ .
the pulsation intensity and that there is reliability in extrapolat- and oscillation period T:
ing the CFD simulation results in single phase flow to higher
3
pulsation intensities. A
= 778 . (8)
T
3.6. Energy dissipation The value of the average energy dissipation for the system of
1 Hz and 20 mm amplitude by Eq. (8) is 0.044 m2 /s3 whilst
As shown in Figs. 4 and 5 the pulsed flow generates vortices the CFD predicted value of was 0.025 m2 /s3 (k– turbulence
in the flow field which in turn generate the shear rate neces- models are known to underpredict the turbulent properties in
sary for the dispersed phase break-up and coalescence needed stirred tanks by a similar amount). The average energy dissipa-
for mass transfer. In turbulent flow the value of shear is linked tion in the pulsed column might not be representative of the av-
to the turbulent energy dissipation, . Previously in the litera- erage level of turbulence and mixing in the operation, and may
ture (Milot et al., 1990; Aoun Nabli et al., 1998) only average only be a crude estimate in the analysis of the column opera-
values of the kinetic energy dissipation over a pulsation cy- tion. For example in the analysed system the maximum value
cle have been reported. Eq. (8) is a correlation proposed by of at a given time, ranged from 0.0446 m2 /s3 to 0.698 m2 /s3 ,
Milot et al. (1990) to estimate the average energy dissipation but the question remains whether the generated maximum oc-
in a pulsed column as a function of peak-to-peak amplitude, A, curs at the point in the pulsation cycle and at a location in the
J.M. Bujalski et al. / Chemical Engineering Science 61 (2006) 2930 – 2938 2937
0.3 0.2
Velocity, [m/s]
-0.1 Intensity (Frequency x Amplitude), [mm/s]
0.0
-0.2
-0.1
-0.3
0°
-0.2 Maximum velocity Down (CFD)
-0.4 Regression for Down
Maximum velocity Up (CFD)
-0.3 Regression for Up
-0.5 Maximum velocity Down (PIV)
Point A
Maximum velocity Up (PIV)
-0.4 -0.6
0.00 0.01 0.02 0.03 0.04 0.05
Radial distance, [m] Fig. 8. Point A maximum and minimum velocity at different pulsation inten-
sities: PIV measurement and CFD simulation comparison.
Fig. 6. Comparison of the PIV measurements and CFD simulation of axial
velocity component at a line positioned 7 mm under the doughnut at a
pulsation phase set at 0◦ (20 mm amplitude, 1 Hz frequency).
4. Conclusions
0.30
CFD- No through flow (no gap)
The pulsed column operation in single phase flow was inves-
0.25 Exp. - No through flow (Trial 4) tigated with PIV and LDV experimental techniques. Extensive
Exp. - Up through flow (Trial 13)
Exp. - Down through flow (Trial 12) sets of PIV measurements have been taken of the unsteady flow
Vertical velocity, [m/s]
0.20 CFD - No through flow(2mm gap) within the column for a range of operating conditions and ac-
180°
curate point measurements taken with LDV have confirmed the
0.15 PIV data. The PIV velocity measurements were taken at seven
different phases in the pulsation cycle. The system was mod-
0.10
elled with CFD and the simulations were validated by compar-
0.05 Point A ing with the measurements at various phases in the pulsation
cycle: the flow field was well matched by the CFD simulation.
0.00 The comparison of the CFD and PIV results showed that the
flow field in certain regions of the pulsed column can depend
-0.05
sensitively on the positioning of the doughnut internal relative
0.00 0.01 0.02 0.03 0.04 0.05
to the column wall if a small gap exists between the two. PIV
Radial distance, [m]
measurements and CFD model predictions for the pulsed col-
Fig. 7. Comparison of the PIV measurements and CFD prediction of axial umn operating in single phase mode show an oscillating sin-
velocity component at a line positioned 7 mm under the doughnut at a uous flow through the column, with significant regions of low
pulsation phase set at 180◦ (20 mm amplitude, 1 Hz frequency). velocity and hence low shear. The range of in the pulsed col-
umn is narrower than in a stirred tank which would lead to
different performance e.g. narrower droplet size distribution.
column where the energy can be used effectively for phase
dispersion.
The maximum for the pulsed column is lower than the pre- Notation
dicted maximum values in stirred tanks under fully turbulent
conditions. Zhou and Kresta (1998) found the values for Rush- A amplitude of oscillation, m
ton turbines to be in a range of 85.2–528 m2 /s3 , but these are B body force, N
localised values inside the impeller generated vortex which is C constant in low Re k– turbulence model
comparatively small in volume, whilst 57% of occurs at a D pulsed column diameter, m
lower value over majority of vessel volume (Ng and Yianneskis, f frequency of oscillation, s−1
2000). Stirred vessels have a large difference between maxi- f function to determine the turbulent viscosity
mum and minimum energy dissipation—ranging by at least two k turbulent kinetic energy, m2 /s2
orders of magnitude (Ng and Yianneskis, 2000) which gener- p pressure, Pa
ates a larger number of fine droplets and a broader size distri- Re Reynolds number, dimensionless
bution. The range of in the pulsed column is smaller (max- t time, s
imum value ranged from 0.0446 to 0.698 m2 /s3 ) generating a T oscillation period, s
narrower droplet size distribution. U velocity, m/s
2938 J.M. Bujalski et al. / Chemical Engineering Science 61 (2006) 2930 – 2938