An Experimental Characterization of Liquid Films in Downwards Co-Current Gas-Liquid Annular Flow by Particle Image and Tracking Velocimetry

The hydrodynamics of downwards gas-liquid annular ﬂows and falling ﬁlms in a pipe were studied experimentally using simultaneous planar laser-induced ﬂuorescence and a combination of particle image and particle tracking velocimetry techniques. The investigated conditions covered the range of liquid and gas Reynolds numbers: Re L = 306 − 1 532 and Re G = 0 − 84 600. The results presented in this paper concern: (i) information on the local and instantaneous velocity ﬁelds underneath the interfacial waves and the appearance of recirculation zones within the liquid ﬁlms under certain conditions, and (ii) mean velocity, velocity ﬂuctuation rms and kinetic energy proﬁles within the liquid ﬁlms. The results indicate that large waves contain multiple recirculation zones, which may play an important role in the gas and liquid phase entrainment mechanisms as well as in the mass and momentum transfer from the near-wall region towards the gas-liquid interface.


Introduction
1 Here, Re L = U Nu δ Nu /ν L is based on the mean liquid film thickness and the bulk-mean film velocity, which in the absence of entrainment is also equal to that defined based on the Nusselt thickness δ Nu = (3qν L /g) 1/3 and the Nusselt velocity U Nu = gδ 2 Nu /3ν L where subscript 'L' denotes the liquid phase and q is the flow-rate per unit width.  into these flows and focusing specifically, and with high spatiotemporal detail, 86 tion within large waves with significant velocities normal to the wall have 114 been locally observed by Karimi and Kawaji (1999), by using a laser-induced 115 PDT technique. Some attempts have also been made to predict these recir-116 culation zones numerically, e.g. by Jayanti and Hewitt (1997). 117 In early work, local velocities in multiphase flows were characterized and 118 measured by intrusive techniques, i.e. hot wire or film anemometry (Ueda 119 and Tanaka, 1975). It was, however, noted that the local velocity measure-120 ments are sensitive to the disturbances caused by the presence of the probes, 121 and their intermittent alternating exposure to the gas and liquid at the in-122 terface Kawaji, 1998, 2000). Therefore, there has been a recent 123 desire to move towards advanced, non-intrusive optical measurement meth-  The paper is structured as follows. The flow facility, which features a 141 specially designed test section and the laser measurement system, is intro-142 duced briefly in Section 2.1 (details can be found in Zadrazil et al. (2014)).

143
Following this, the data analysis methodology is described in Section 2.3.

144
Experimental results are then presented and discussed in Section 3. Specif-  Finally, concluding remarks from this work can be found in Section 4.  The precision of the alignment was 0.2 mm/1 m, giving a maximum deviation 162 from the true vertical position of 0.6 mm or 0 • 0 41 .

163
The investigated Re ranges for the liquid (i.e. water) and the gas (i.e. 164 air) phases were: Re L = 306 − 1 532 and Re G = 0 − 84 600, respectively. 165 It is noted that Re L = U L δ /ν L is defined in the present work based upon 166 the bulk-mean film velocity U L and corresponding mean film thickness δ , 167 such that it is four times smaller than that based on the superficial liquid   The thickness of the laser sheet at the measurement point was approximately 189 0.1 mm.

190
The flow was seeded with 10 µm mean-diameter, monodisperse (< 3%) 191 and highly uniform in shape melamine resin particles doped with Rhodamine-  Table 1). The laser measurement configuration was identical to that used in  The raw images were first processed in order to obtain liquid-phase informa-210 tion from the PLIF measurement (of the Rhodamine-B dye in the water).

211
The procedure by which this was done is described in detail in Zadrazil et al.  The pre-processed image-pairs (an example pre-processed image is shown 228 in Figure 2(b)) were then processed using a cross-correlation function that 229 was applied to the images based on a multi-pass approach. During the first 230 pass (i.e. the initial estimation of the velocity vectors) the PIV interrogation 231 window was set to 64 × 64 pixels with 50% overlap of adjacent areas. Based 232 on this window size, each interrogation window contained 7 particles on av-233 erage. Adrian (1990, 1992) showed that the probability of valid  increases to approximately 0.01 pixels (10%) for a particle image shift of 244 0.1 pixels (corresponding to 56.0 ± 5.6 mm/s and 14.9 ± 1.5 mm/s for a dt 245 of 40 and 150 µs, respectively). These estimates fall within the expected 246 range of the uncertainty in the particle displacement, which is typically less 247 than 5 − 10% of the particle diameter (Prasad et al., 1992). For the second 248 and third passes the PIV interrogation window was reduced to 16 × 16 pixels 249 with 50% overlap, while employing the information from the PIV window 250 displacement from the first pass that was retained. reinserted if a difference was < 1.9 × to the average of the rms of the same 268 vectors, (iv) groups containing less than 10 vectors were removed.

269
An example of a resulting PIV velocity vector-map can be seen in Fig-270 ure 2(c), where the PIV vector-to-vector spatial resolution is 179 µm. Finally, 271 based on the PIV results, a PTV algorithm was also employed in which indi-272 vidual particles were tracked within 8 × 8 pixels PTV interrogation windows.

273
A particle size in the range 1 − 3 pixels was applied, allowed vector range 274 of ±2 pixels relative to the reference vector from PIV, as well as a mini-275 mum particle scattering intensity during the PTV calculation. Figure 2(d) 276 shows an example of the instantaneous PTV velocity vector-map that was 277 generated from a single PIV image-pair.

278
The instantaneous velocity profiles within the waves (see Figure 3(c)) 279 were constructed from the instantaneous PIV velocity vector-maps described 280 in the previous paragraphs (see Figure 3(a)). The profiles were obtained by 281 spatial averaging over ≤ 100 pixels (≤ 2.6 mm, or 7.8% of the image width) 282 in the streamwise direction, which corresponds to a maximum of 12 vectors.

283
The averaging was only performed in manually selected areas with a flat gas-284 liquid interface. The topology of the gas-liquid interface was obtained from 285 the PLIF measurement as described previously and shown in Figure 3(b).

286
Finally, for each given experimental run (i.e. set of independent flow con-287 ditions; see Table 1), the set of 500 instantaneous PTV velocity vector-maps where u x and u y are the instantaneous axial and radial velocity components, 301 respectively. The instantaneous temporal fluctuations of velocity u x and u y 302 are defined from Reynolds decompositions: where u x and u y are the time-mean axial and radial velocities, respec-304 16 tively. The axial and radial velocity fluctuation rms are defined as: where n is the number of instantaneous velocity field images and u x,i and u y,i 306 are the instantaneous axial and radial velocity components, respectively. In 307 addition, the x-y Reynolds stress (RS) component is given by: Finally, the 2-D kinetic energy (KE) associated with temporal axial and 309 radial velocity fluctuations in the liquid film flows is defined as: In this work all reported statistical velocity measures, including the mean 311 and rms velocities, the Reynolds stresses and kinetic energies, were evaluated 312 from at least 75 000 velocity vectors.  suggest that with increasing δ p /δ s values the waves become more complex and 337 begin to feature multiple recirculation zones and hence stagnation points.

338
The presence of multiple recirculation zones within a wave is not surpris-339 ing, since as pointed out by Karimi and Kawaji (1999), the length scales of    and specifically for Re L > 600 in our investigated conditions (Re G < 21 100).

532
In summary, it is found that a partial collapse of the mean axial veloc-  Re L . On the other hand, the KE is clearly higher in the higher Re G flow.

607
Importantly, when normalized by the local (axial) flow speed u x (y/R) 608 (see Figure 9(e-f)) and not the bulk flow speed U LF (see Figure 9(c-d)), the 609 profiles collapse approximately onto a common shape, although they retain 610 some distinction; at increasingly higher values of Re L the profiles appear to 611 shift to shorter distances y/R, which may scale with the film thickness δ .        Table 1: Experimental matrix of investigated flow conditions. Here, Q is the measured flow rate, Re * L denotes the liquid Reynolds number based on the superficial liquid velocity U SL and the pipe diameter D, and dt is the time interval between the PIV image-pairs. Q L ·10 −5 Q G ·10 −3 U SL ·10 −2 U SG Re L Re *