3D - Velocity measurements in microscopic two-phase flows by means of micro PIV

Transcription

1 3D - Velocity measurements in microscopic two-phase flows by means of micro PIV Ulrich Miessner 1, Ralph Lindken 2, Jerry Westerweel 3 1: Laboratory for Aero- and Hydrodynamics, Technical University of Delft, Netherlands, u.p.miessner@tudelft.nl 2: Laboratory for Aero- and Hydrodynamics, Technical University of Delft, Netherlands, r.h.lindken@ tudelft.nl 3: Laboratory for Aero- and Hydrodynamics, Technical University of Delft, Netherlands, j.westerweel@tudelft.nl Abstract This work focuses on the investigation of three dimensional velocity distributions in microfluidic dispersed oil/ water two-phase flows. As part of the development of a real-time quality management system a DNA-analysis process is integrated on a micro-fluidic lab-on-a-chip device. To move sample through the channel, droplets are used as sample containers. In order to successfully operate the processes, the internal convection patterns are identified as the key parameter to manipulate the samples. Therefore, we investigate the spatial velocity distribution inside and outside the liquid slugs by means of micro particle image velocimetry (micro-piv). A continuous droplet train is established in a microfluidic PDMS (polydimethylsiloxane) device featuring rectangular channel walls (height: 100µm, width: 100µm). The capillary number of the pressure-driven flow is Ca ~ 1, while the Reynolds number, based on the channel geometry and the bulk velocity, is Re << 1. Octanol is used as bulk phase, while the droplets consist of a mixture of water and glycerol. Both, separator compartments and droplets are of the same length (length: 290µm). Micro particle image velocimetry is used to measure velocity fields in both phases simultaneously. Both phases are seeded with fluorescent tracer particles and refractive index matching ensures optical access without distortions caused by the liquid-liquid interface. A 3D scan of the volume is performed over 15 planes with a separation distance of 5µm each. For each measurement plane a set of 400 image pairs has been recorded, sorted and aligned with respect to each other. Due to the sparse tracer distribution ensemble averaging is used to calculate the velocity fields for the planes. The internal circulation is represented by streamlines calculated from relative velocity fields. The interface velocity is used as reference. Two main circulation patterns can be observed in the center plane: one inside the droplet and one outside in the separator compartment. These co-rotating main structures are linked with each other by smaller secondary vortices close to the interface. The secondary vortices are observed inside and outside the droplet. 1. Introduction Real-time quality management systems scanning for specific germs or bacteria in e.g. dairy products heavily depend on time consumption to be functionally effective. Although, conventional DNA-analysis offers the specificity needed, it is not suitable for real-time processes due to its long process time. However, a lab-on-a-chip device for fast DNA analyses will effectively intensify the involved process steps by means of miniaturization. Such processes, like heat and mass transfer, mixing, reaction and separation, directly depend on convection patterns in the micro fluidic flow. To manipulate these unit operations the involved velocity fields need to be understood. The use of single-phase flows to process samples offers well known and easily manipulated velocity distributions: parabolic profiles in tubes and hyperbolic profiles in rectangular channels. However, a given sample will suffer from band broadening due to the effects of Taylor-Aris dispersion (Taylor 1953; Aris 1956; Garstecki et al. 2006; Günther and Jensen 2006). Additionally, cross contamination of different samples might occur without a flushing sequence. In contrast, the application of a micro fluidic two-phase flow results in multiple confined compartments (Song et al. 2003; Tice et al. 2003; Zeng et al. 2004), which are neither subject to band broadening nor to cross contamination. However, the circulation pattern inside these droplets is less understood and thus not as easy to manipulate. Microscopic two-phase flow has been studied, described and simulated - 1 -

2 by (Bretherton 1961; Sarazin et al 2006). Gas-liquid flow has been investigated quantitatively by (Waelchi and Rohr 2006; van Steijn et al. 2007; van Steijn et al. 2008) using micro-piv. However, given the high viscosity ratio between the phases it can be assumed that hydrodynamic coupling via the gas liquid interface is unidirectional: from the liquid into the gas phase. Additionally, it is not possible to confirm the influence due to a lack of suitable measurement principles. A 3D velocity measurement of microscopic liquid-liquid two-phase flow in square channel geometry has been conducted by (Kinoshita et al. 2006) applying high-speed confocal scanning microscopy. In this case hydrodynamic coupling is bidirectional because of the lower viscosity ratio. However, due to the limitations of the measurement technique the speed of the investigated droplet is low and only the convection pattern inside the droplets has been investigated. Several authors investigated velocity distributions in microscopic liquid-liquid two-phase flow qualitatively using micro-piv. CFD-simulations were compared to the results of micro-piv measurements in circular capillaries (Kashid et al. 2005). (Malsch et al. 2008) also reported experiments in round tubes. Here, the flow in a separator compartment of gas-liquid flow is investigated and complemented with a separate liquid-liquid experiment to show the flow inside the droplet. In this study, we simultaneously measure the 3D velocity distribution inside and outside the droplets of a continuous droplet train in square channel geometry by means of micro particle image velocimetry (micro-piv) in high magnification and resolution. A 3D scan of the channel volume is performed over 15 planes with a separation distance of 5µm each. 2. Experimental Setup In order to create a liquid-liquid two-phase flow, we designed a PDMS (polydimethylsiloxane) device featuring rectangular channel walls (height: 100µm, width: 100µm) (see Fig. 1). On a silicon wafer a) a layer of photo-resist (SU , MicroChem, USA) with a thickness of 100µm is deposited by spin coating b). The photo-resist is covered with a mask, which carries the layout for the channel. The photo-resist is exposed to UV-light through the mask c). This leads to cross linking of the polymer chains where it is exposed d). After dissolving the non-cross-linked resist e), the positive channel structures remain on the wafer f). This structured wafer serves as master form for the PDMS casting. Then the PDMS is mixed in a ratio of polymer and curing agent of 10:1 and cast in a thick layer (1cm) onto the master structure g). The semi-cured channel is removed from the master structure h). To form a thin channel lid, we use a glass wafer i) and spin-coat thin layer of PDMS on it j). Joining both semi-cured parts carefully and successively completing the curing process perform the bonding of the channel parts k). a) b) c) d) e) f) g) h) i) j) k) Fig. 1: Photolithography process steps: SU-8 is used to deposit microstructures on a silicon wafer. A PDMS-layer is casted on top to form rectangular channel cavities. A thin non-structured PDMS-layer forms the lid. The semi-cured parts are brought in contact and cured together to establish a durable bond

3 The two-phase flow is created as a continuous droplet train at a T-junction of the PDMS device (see Fig. 2). Octanol is used as bulk liquid and the droplet consist of an aqueous solution (water/ glycerol). Surfactants are not applied in this set up. A pulsation-free programmable syringe pump (Cetoni nemesys) is used to establish a stationary flow. Both phases enter the channel with the same volume flow rate (1µl/min). The connecting tubing consists of flexible glass capillaries (150µm inner diameter, fused silica) to keep the dead volume low. Water/ Glycerol Octanol Syringe Pump PDMS-Chip Waste PIV Raw Image Fig. 2: A syringe pump establishes continuous flows of octanol and water/ glycerol. The two immiscible liquids enter the channel through fused silica capillaries. A continuous train of droplet forms at the T-junction of the channel. Octanol wets the walls and serves as carrying bulk phase, while the water/ glycerol mix forms aqueous droplets. The refractive index of the liquids is matched. The chosen measurement method is micro particle image velocimetry (micro-piv) (Santiago et al. 1998; Meinhart et a. 1999) (see Fig. 3). A dual cavity laser (New Wave Pegasus, frequencydoubled Nd:YLF, λ=527nm) is used as light source in combination with a 12bit-CCD camera (PCO SensiCam QE Imager intense, Resolution: 1376 x 1040 pixel) to record the particle images through an inverted epi-fluorescent microscope (Zeiss Axiovert 200). Rhodamine-B coated fluorescent particles of about 1µm diameter are immersed in the aqueous phase (microparticles: PEG-RhB, d p =1µm) as well as in the oil phase (microparticles: MF-RhB, d p =1.3µm). A diffuser and a lens system shape the laser light after it enters the microscope. It is reflected at the dichroic mirror through the objective (Zeiss, M 25x, LD LCI Apochromat, multi-immersion) into the transparent channel. The flow is volume illuminated. The tracer particles absorb the light and emit fluorescent light. Thus, light with two different wavelengths re-enter the microscope through the objective. While the mirror reflects the laser light, the fluorescent light of the particles passes the mirror and is recorded with the camera. The measurement plane is represented by the focal plane of the objective and the depth of correlation (z corr = 5µm) determines the contribution of out-of-focus particle to the measurement result (Olsen and Adrian 2000). At a distance of a equivalent channel diameters downstream of the junction the velocity fields are measured as a scan of 15 positions in height. The distance between the planes is Δz = 5µm. A piezo stepper (Piezo Jena) is used to precisely control the vertical position of the focal plane with respect to the channel. For each measurement plane a set of 400 image pairs has been recorded. The time delay between the double frames is Δt = 500µs, which results in a maximum particle image displacement of 20 pixel. The pixel resolution is 0,363µm/pixel

4 Fig. 3: A dual cavity laser is used as light source in combination with CCD camera to record particle images of the fluorescent tracers through an inverted epi-fluorescent microscope. A diffuser and a lens system shape the laser light after it enters the microscope. The light is reflected at the dichroic mirror through the objective into the transparent channel. The flow volume is entirely illuminated. The tracer particles absorb the light and emit fluorescent light. Thus, light with two different wavelengths re-enter the microscope through the objective. While the mirror reflects the laser light, the fluorescent light of the particles passes the mirror and is recorded with the camera. The measurement plane is the focal plane of the objective. 3. Experimental Methods In order to successfully apply micro-piv optical access has to be assured. The transparency of the PDMS channel provides good optical access to the flow. The straight surface of the glass wafer and its thin PMDS coating are positioned perpendicular to the optical axis of the laser light. Negligible optical distortions are caused by the solid-liquid interface of the channel lid. However, the liquidliquid interface between the oil and water is curved and introduces optical distortion due to a mismatch in refractive index. Consequently the index of refraction has to be matched. Refractive Index Matching We used the method for refractive index matching as described in (Miessner et al. 2008). A detailed description of the physics and methods to match the index of refraction is given by (Budwig et al. 1994; Narrow et al. 2000; Nguyen et al. 2004; Miller et al. 2006). Since only the refractive index needs to be matched we introduce a single degree of freedom by using a water/ glycerol mixture as aqueous dispersed phase. The oil phase is not altered. The mass fraction x B of glycerol, which is added to the water, determines the refractive index and the viscosity of the droplet. The change of index of refraction with increasing mass fraction of glycerol is shown in Fig. 4. Based on the mass fraction of glycerol in the aqueous mix the index of refraction can be chosen between n=1.333 (pure water) and n=1.473 (pure glycerol). The refractive index of octanol at room temperature is n= Thus, the mass fraction has to be chosen as follows: water x = 0.2 and glycerol x=

5 Fig. 4: The change of index of refraction with increasing mass fraction of glycerol. Based on the mass fraction of glycerol in the aqueous mix, the index of refraction can be chosen between n=1.333 (pure water) and n=1.473 (pure glycerol). The refractive index of octanol at room temperature is n= Thus, the mass fraction has to be chosen as follows: water x = 0.2 and glycerol x=0.8. Micro-PIV Evaluation Although, a set of 400 image pairs has been recorded for each of the 15 measurement planes, not every image pair can contribute to the evaluation with ensemble averaging. Since the recording frequency of the micro-piv system has not been synchronized to the droplet generation frequency of the droplets, the entire droplet is recorded in only 15% of the images. These image pairs have to be pre-processed to shift them to a reference point at the droplet tip. In order to align the images within a set, a manual sifting, combined with a position determination of prominent droplet image features, has been performed. The prominent features are the front and backside of the droplet. Since the droplet length varies slightly (Δl~10µm) over time, a single reference point is not sufficient to align the image pairs. Alignment of the raw images to the front results in inaccurate ensemble averaging at the back and vice versa. Therefore, two groups of image pairs are created from one set; one is manually aligned to the interface on the backside of the droplet and one to the front side. Both alignments are performed with respect to the interface position in the first image. In order to increase accuracy, the manually pre-aligned groups are then scanned with a cross-correlation algorithm that accurately shifts the prominent features on top of each other. Successively, the image groups are divided into front and back part and reorganized in a single double image (see Fig. 5a). The group aligned to the front of the droplets provides the front parts of the reorganized set, while the group aligned to the backside interface of the droplet provides the back parts. Finally, the reorganized sets are manually shifted with respect to each other in order to create a 3-dimensional representation of the measurement volume. Background noise is removed by a sliding minimum image pre-processing algorithm (see Fig. 5b)and a mask is applied prior to the ensemble averaging of the images (see Fig. 5c). The mask covers the channel walls and visually divides the reorganized parts from each other. The ensemble averaging is done with a decreasing window size from 128pixel down to 16pixel with a 50% overlap. To give an example of the according results the relative velocity field is presented (see Fig. 5d) along with a field of streamlines (see Fig. 5e) and the distribution of the cross-channel velocity plotted as grey values (see Fig. 5f)

6 a) d) b) e) c) f) Fig. 5: Image pre-processing prior to the ensemble averaging of a set. a) First image of a reorganized image set. The rest of the images of the set have been aligned with respect to the front interface of the droplet (right half of the image) and with respect to the interface at the backside of the droplet (left side of the image). b) Enhanced image: the sliding average of the grey values is subtracted to remove background noise. c) A mask ensures an evaluation only inside the channel walls and creates a visible gap between the reorganized parts of the image set. d) Relative velocity field: a result of the ensemble averaging. The reference velocity is the movement of the front interface at the channel centre. e) Streamlines calculated from the example velocity field. f) Cross channel velocity printed as grey value. The line represents the interface position. 4. Results In the following, the results of the 3 dimensional micro-piv scan through a liquid-liquid two-phase flow is presented. All results will be displayed as a range of six measurement planes beginning in the center plane of the flow (z = 50µm), stepping in 10µm steps towards the wall (z = 0µm). An automated interface recognition could not be used, since the refractive indices of the liquids were matched. Nevertheless, we manually inserted a line to indicate the position of the interface, based on the raw images and the recirculation patterns. At first, the relative velocity field in and outside of an average droplet in a continuous slug flow is presented. The interface velocity in the center plane serves as reference velocity. Secondly, the streamlines, derived from the velocity fields, are used to interpret the recirculation pattern in the droplet, outside the droplet and their interdependency. The velocity measurement planes are presented in Fig. 6. The droplet passes the field of view from left to right. In the center plane (z = 50µm) hyperbolic velocity profiles can be seen in the oil compartments of the carrier liquid as well as in the very center of the aqueous droplet. The main relative flow direction is indicated with an arrow. The rear and the front of the droplet are areas with low velocities. Along the centerline of the droplet liquid is accelerated from the back of towards the middle and decelerated at the front. In this region, the internal flow is divided and accelerates along the wall towards the back and where it decelerates. A similar velocity distribution can be found in the separation compartment between two successive droplets. The magnitude of velocities in the center of the droplet decreases in depth direction towards the wall. An almost stagnant layer can be found at a distance of about 10µm from the lower channel wall. Only near the wall liquid is flowing backwards. In the closer proximity of the lower wall the main flow direction has reversed its direction with respect to the absolute velocity of the droplet. The shape of the interface differs from the front to the back of the droplet. This is reflected in the Capillary number of the system Ca ~ 1. Thus, the interfacial forces do not dominate viscous forces and the interface may be deformed by the flow. In depth direction the size of the intersected interface of the droplet decreases. Thus, the droplet does not fill the channel completely and a communicating octanol flow can be observed between the separating compartments. Additionally a decrease in channel width can be seen with along the z-axis of the channel. Consequently, the channel geometry is not as rectangular as previously assumed. The shape of the cross section appears to be trapezoidal

7 z = 50µm z = 40µm z = 30µm z = 20µm z = 10µm z = 0µm Fig. 6: Velocity measurements on certain planes of the continuous liquid-liquid slug flow. The displayed measurements have been performed on 6 planes from the centre plane (z = 50µm) downwards to the wall (z = 0µm) with a distance of 10µm. The line indicates the interface position. The flow velocities are displayed as relative vector fields. The interface velocity at the front of the droplet in the centre plane (z = 50µm) is the reference velocity. The absolute motion of the droplet is directed from left to the right. Generally, a parabolic velocity profile can be observed in the central region of the droplet as well as in the centre of the separation compartment. On plane z = 10µm a flow reversion with respect to the absolute motion direction can be observed

8 Even though 2D streamlines only represent flows correctly on symmetry planes, we use them for visualization of the whole measurement volume. Taking into account the loss of volume into the third dimension due to velocity components in the z-direction, qualitative assessment of the inplane circulation patterns is still possible. The streamline plots simultaneously represent the circulation pattern of the internal droplet flow and the separating bulk phase (see Fig 7). Two main circulation patterns can be observed in the center plane (z = 50µm): one inside the droplet and one outside in the separator compartment. The wall friction and the pressure gradient over the channel length create the circulation pattern outside the droplet. Since no shear occurs on an ideal liquid-liquid interface, flow is induced inside the droplet. The two main circulation patterns show the same rotational direction. These co-rotating main structures are linked with each other by smaller secondary vortices close to the interface. The combination of the refractive index matching with micro-piv applying a high numerical aperture and magnification allows the observation of the secondary vortex structures. The secondary vortices are observed inside the droplet at its front and outside the droplet at its rear interface of the droplet (at z = 50µm and at z = 20µm respectively). Five stagnation points can be observed in the centre plane. The first is converging and located on the interface at the front of the droplet, the second is diverging can be found in the droplet front, separating the secondary vortex from the main circulation, while the third is converging and is situated on the rear interface of the droplet. The last two converging stagnation points can be observed off the central axis on the interface at the droplet front, where the main circulation at the wall of the first oil compartment separates into the main circulation and the leakage flow into the next oil compartment. The secondary vortex structure at the front stays intact from the center of the channel to the measurement plane z = 10µm. Here, the flow at the droplet center stagnates and the droplet internal flow begins to reverse its direction with respect to the main absolute motion of the aqueous phase. In contrast, the secondary vortex structure at the rear interface of the droplet is located inside the oil compartment and develops from the channel centre towards the lower wall

9 z = 50µm z = 40µm z = 30µm z = 20µm z = 10µm z = 0µm Fig. 7: Streamlines of the relative velocity field of a droplet in a continuous droplet train. The streamlines are displayed from the channel center position (z = 50µm) towards the channel wall (z = 0µm). The line indicates the interface position. Two main circulation patterns can be observed: one inside the aqueous droplet and the other one in the oil compartment between two successive droplets. They show the same rotational direction. These co-rotating main structures are linked with each other by smaller secondary vortices

10 4. Discussion The PDMS production did not result in a perfectly square cross section of the microfluidic channel. Instead, we observed a trapezoidal shape. The channel is narrower at the bottom compared to the top of the channel. Due to extended exposure times during the manufacturing of the channel the top part of the photo resist experienced an over-exposure. An excess of acid is produced during the UVlight exposure, acid diffuses into the surrounding non exposed material and results in cross linking of the epoxy polymer without being exposed to UV light. This effect is referred to as T-topping. Since this structure is used as a master form for PDMS casting, a trapezoidal shape of the channel is produced. Thus, the flow pattern in the lower half and the upper part of the channel are not exactly symmetrical. Additionally, the wall roughness is not as smooth as one would expect. Small bumps in the wall may locally change the flow pattern close to the wall. However, this has a small impact on the measurements, since the evaluation does not always take place at the same position. The shifting procedure prior to ensemble averaging weakens the effect. Also related to the production of the PDMS channel is the fluctuation of the droplet size and droplet building frequency. The resolution of the mask (2400dpi) used for the UV- exposure is not sufficient to resolve sharp edges. However, a singularity at the T-junction is needed to create a stationary stream of mono-disperse droplets. Consequently, the shifting procedure during evaluation would be unnecessary and phase locked recording of the flow is possible. The converging stagnation points at the droplet front and back accumulate tracer particles at the surface of the droplet. This accumulation causes the interface to become ridged, which enables a shear stress transport onto the liquid-liquid interface. This might explain the missing vortex in the center plane at the back of the droplet. The position in z-direction is not exactly known, since the opposite channel wall could not be reached due to the limited working distance of the objective. The determination of the channel center has been done by evaluation of the streamline patterns. Since the flow pattern changes over the channel depth, the magnitude of the velocities is changing along with it. Thus, different delay times between the recorded frames can be used in order to optimize the particle image displacement for each measurement plane and increase the accuracy of the measurements. Furthermore, instead of using a standard ensemble-averaging algorithm, the interrogation window could be used with a weighting function to account for the main flow direction. This also increases the accuracy of the measurements. 5. Conclusion For the first time flow patterns have been measured simultaneously inside and outside the disperse phase of a microscopic liquid-liquid two-phase flow by means of micro-piv and refractive index matching. The fluid dynamical coupling of the flow in carrier phase with the droplet internal flow has been revealed. Now a parametrical study needs to be performed to discover manipulation possibilities for the vortex mechanisms. If these secondary vortex structures can be influenced, they could be used to break the symmetry of the flow inside the droplet and enhance convectional transport. 6. References Aris R (1956) On the Dispersion of a Solute in a Fluid Flowing through a Tube, Royal Society of London. Series A, Mathematical and Physical Sciences, Vol. 235, No. 1200, Bretherton FP (1961) The motion of long drops and bubbles in tubes. Journal of Fluid Mechanics

11 10: Budwig R (1994) Refractive index matching methods for liquid flow investigations, Exp Fluids, 17, Garstecki P, Fuerstman MJ, Stone HA, Whitesides GM (2006) Formation of droplets and bubbles in a microfluidic T-junction scaling and mechanism of break-up, Lab Chip, 6, Gunther A, Jensen KF (2006) Multiphase microfluidics: from flow characteristics to chemical and material synthesis. Lab on a Chip 6: Kashid MN, Gerlach I, Goetz S (2005) Internal circulation within the liquid slugs of a liquid liquid slug-flow capillary microreactor. Ind Eng Chem Res 44(14): Kinoshita H, Kaneda S, Fujii T, Oshima M (2006) Three-dimensional measurement and visualization of internal flow of a moving droplet using confocal micro-piv, Lab Chip, 7, Malsch D, Kielpinski M, Merthan R, Albert J, Mayera G, KÖhler JM, SÜße H, Stahl M, Henkel T (2008) PIV-Analysis of Taylor flow in micro channels, Chemical Engineering Journal 135S (2008) S166 S172 Meinhart CD, Wereley ST, Santiago JG (1999) PIV Measurements of a microchannel flow, Exp. in Fluids, Volume 27, 5, Miessner U, Lindken R, Westerweel J (2008) Velocity Measurements in Microscopic Two-Phase flows by Means of Micro-PIV, ASME Confrence Proceedings, Darmstadt, ICNMM Miller P, Danielson K, Moody G, Slifka A, Drexler E, Hertzberg J (2006) Matching index of refraction using a diethyl phthalate/ethanol solution for in vitro cardio-vascular models, Exp Fluids, 41, Narrow TL, Yoda M, Abdel-Khalik SI (2000) A simple model for the refractive index of sodium iodide aqueous solutions, Exp Fluids, 28, Nguyen TT, BiadillahY, Mongrain R, Brunette J, Tardif JC, Bertrand OF (2004) A method for matching the refractive index and kinematic viscosity of a blood analog for flow visualization in hydraulic cardiovascular models, J Biomech Eng, 126, Olsen MG, Adrian RJ (2000) Out-of-focus effects on particle image visibility and correlation in microscopic particle image velocimetry. Experiments in Fluids Suppl.:S166-S174. Santiago JG, Wereley ST, Meinhart CD, Beebe DJ, Adrian RJ (1998) A particle image velocimetry system for microfluidics, Exp. in Fluids, Volume 25, 4, Sarrazin F, Bonometti T, Loubie`re K, Prat L, Gourdon C, Magnaudet J (2006) Experimental and numerical study of droplets hydrodynamics in microchannels. AIChE 52(12): Song H, Tice JD, Ismagilov R (2003) A microfluidics system for controlling reaction networks in time. Angew Chem Int Ed 42: Taylor G (1953) Dispersion of Soluble Matter in Solvent Flowing Slowly through a Tube, Royal Society of London. Series A, Mathematical and Physical Sciences, Vol. 219, No. 1137, Tice JD, Song H, Lyon D, Ismagilov RF (2003) Formation of droplets and mixing in multiphase microfluidics at low values of the reynolds and the capillary numbers. Langmuir 19: van Steijn V, Kreutzer MT, Kleijn CR (2007) µ-piv study of the formation of segmented flow in microfluidic T-junctions, Chemical Engineering Science Volume 62, Issue 24, Pages van Steijn V, Kreutzer MT, Kleijn CR (2008) Velocity fluctuations of segmented flow in microchannels, Chemical Engineering Journal, Volume 135, Supplement 1, Pages S159-S165 Waelchli S, Rohr PR (2006) Two-phase flow characteristics in gas liquid microreactors, International Journal of Multiphase Flow, 32, Zheng B, Tice JD, Ismagilov RF (2004) Formation of Droplets of Alternating Composition in Micrfluidic Channels and Application to Indexing of Concentrations in Droplet-Based Assays, Anal Chem, 76,