Scaling-up ultrasound standing wave enhanced sedimentation filters

Ultrasonics 56 (2015) 260–270

Contents lists available at ScienceDirect

Ultrasonics journal homepage: www.elsevier.com/locate/ultras

Scaling-up ultrasound standing wave enhanced sedimentation filters Jeff E. Prest b, Bernard J. Treves Brown a, Peter R. Fielden b, Stephen J. Wilkinson c, Jeremy J. Hawkes a,⇑ a

Manchester Institute of Biotechnology, The University of Manchester, Oxford Road, Manchester M13 9PL, UK Department of Chemistry Faraday Building, Lancaster University, Bailrigg, Lancaster LA1 4YB, UK c University of Chester Faculty of Science and Engineering, Thornton Science Park, CH2 4NU, UK b

a r t i c l e

i n f o

Article history: Received 21 March 2014 Received in revised form 4 August 2014 Accepted 5 August 2014 Available online 21 August 2014 Keywords: Filtration Resonant chambers Acoustic radiation force Acoustofluidic Scale-up

a b s t r a c t Particle concentration and filtration is a key stage in a wide range of processing industries and also one that can be present challenges for high throughput, continuous operation. Here we demonstrate some features which increase the efficiency of ultrasound enhanced sedimentation and could enable the technology the potential to be scaled up. In this work, 20 mm piezoelectric plates were used to drive 100 mm high chambers formed from single structural elements. The coherent structural resonances were able to drive particles (yeast cells) in the water to nodes throughout the chamber. Ultrasound enhanced sedimentation was used to demonstrate the efficiency of the system (>99% particle clearance). Subwavelength pin protrusions were used for the contacts between the resonant chamber and other elements. The pins provided support and transferred power, replacing glue which is inefficient for power transfer. Filtration energies of 4 J/ml of suspension were measured. A calculation of thermal convection indicates that the circulation could disrupt cell alignment in ducts >35 mm high when a 1 K temperature gradient is present; we predict higher efficiencies when this maximum height is observed. For the acoustic design, although modelling was minimal before construction, the very simple construction allowed us to form 3D models of the nodal patterns in the fluid and the duct structure. The models were compared with visual observations of particle movement, Chladni figures and scanning laser vibrometer mapping. This demonstrates that nodal planes in the fluid can be controlled by the position of clamping points and that the contacts could be positioned to increase the efficiency and reliability of particle manipulations in standing waves. Ó 2014 Elsevier B.V. All rights reserved.

1. Introduction 1.1. Physical particle filtration and concentration Particle concentration and filtration stages are used by almost every laboratory and industry that processes particle suspensions. These stages could be accomplished by a wide variety of devices but are usually performed by centrifuges, settling devices, filtration screens or chemical flocculation. These few devices are very successful for handling fluid volumes scales from cm3 to m3 but when very small scale, very large scale, high flow rates or continuous flow are needed these standard devices have limited development potential. At the sub-millilitre scale there are several particle manipulation processes based on physical characteristics which promise to increase the scope of microfluidic applications, for example electrostatic [1] and magnetic attraction [2], thermophoresis [3], microthermal field-flow fractionation (micro-TFFF), shear-induced ⇑ Corresponding author. Tel.: +44 151 2818068. E-mail address: [email protected] (J.J. Hawkes). http://dx.doi.org/10.1016/j.ultras.2014.08.003 0041-624X/Ó 2014 Elsevier B.V. All rights reserved.

particle migration [4], sedimentation based field-flow-fractionation [5], dean-flow inertial-focusing [6], electrowetting [7,8], optical traps [9], dielectrophoresis [10] and ultrasound-standing-wave particle filtration [11–13]. This last process, the subject of this paper, can in principle also be scaled up and it can also operate with: gas or liquid suspension phases; high or low media conductivity; and opaque or transparent samples. Its channels can be more than 10 mm across for filtering 5 lm diameter particles (therefore contaminating debris is unlikely to cause blockages). Despite its seemingly huge potential ultrasound filtration is currently not widely used, although commercial devices have existed for nearly 20 years. An inability to achieve significant scale-up and reliability have been two factors holding back development. Both challenges are addressed by the system described in this paper. 1.2. Scale-up limited by heating The largest commercial ultrasound-enhanced-sedimentationfilter (from AppliSens [14]) has a processing volume of just 290 ml and its ultrasonic path length is <0.055 m. Current ultrasound

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

filtration systems operate in the low MHz region and a quick calculation shows that sound absorption by the fluid is not limiting scaleup for most current chambers: At 1 MHz, 50% of sound wave energy is absorbed over a distance of 27 m in deionised water and, 0.4 m in whole blood. These distances are calculated from d = ln(A0/A1)/a, where A0 and A1 are the initial and transmitted sound amplitudes, and their ratio, A0/A1, is equal to 2 for 50% absorption. The absorption coefficients a are 0.03 Np m1 for water [15] and 1.4 Np m1 for blood [16]. However although sound absorption within the fluid is low, device heating is a problem and therefore for almost every apparatus with more than 1 ml of fluid in the resonant chamber; air, water or Peltier cooling is used [13,17–20]. In one exception to this, heating is permitted and a thermal equalization period is introduced [21]. We assume a significant amount of the heat comes from destructive interference arising from the geometries and interfaces of the PZT and chamber. The 36 ml chamber described here does not have a cooling system. Instead minimally-damped-coherent-waves are encouraged throughout the duct. This is achieved through by directly coupling the PZT to the chamber rather than using a glue interlayer; minimizing contact points, so damping at suboptimal contacts is reduced, and by using a duct formed from a single element as the resonant chamber, so reducing losses at joints. We give some proof-of-principle experiments which show that in the current system, filtration is possible with only a low level of heat production and we discuss ways to further reduce heat production. Heat disrupts ultrasound filtration because it changes the resonant frequency [22] and in addition thermal convection disrupts the acoustic organisation of particles [11]. While resonant frequencies can be tracked, in large systems thermal convection presents the greatest problem, we quantify the problems and suggest a reduced chamber height as an amelioration method.

1.3. Advantages of ultrasound enhanced sedimentation for scale-up There are two main approaches used for ultrasound filtration: ultrasound enhanced sedimentation, (UES) [11,17,18,23–25] and hydrodynamic acoustic filtration (HAF) [22,26,27]. UES filters are usually multi-wavelength devices with a channel 10– 60 mm wide and do not require high precision in their construction. HAF devices require much higher manufacturing precision: they are usually half wavelength devices, the acoustic path length is <1 mm and downstream from the sound the channel is divided to direct each band of focused particles to a separate outlet. Scale-up of HAF devices is complex, as multi-wavelength sound paths require additional outlet-ducts. By contrast, UES systems can be scaled up in any dimension without adding more ducts. Therefore we selected UES filtration to demonstrate our scale-up of the resonant section. The approaches described such as glue free contacts and control of the nodal plane direction could however also be introduced to HAF and small-scale devices.

1.4. Wave interference between surfaces The wave-shape on the surface of the PZT and the duct wall (or any two parts consisting of different materials or geometries) are not usually the same at any frequency (as in Fig. 1), therefore any interface with face-to-face contact will have some out of phase regions (marked 180° in Fig. 1) where interference is destructive, and even regions moving in phase will move with different magnitudes when impedances differ. Therefore transferring vibrations from one structure to another is always accompanied by some conversion of ordered vibration into thermal energy.

261

1.5. Introducing point contacts to give vibrational freedom In traditional UES filter design the wave shape on interacting surfaces is not considered, they are constructed with a fluid separating a driven wall and a reflecting wall; the driven wall is either a PZT or a wall with a PZT glued behind it; the side walls and watertight seal are not designed to couple structural vibrations to the reflector [17,24,27]. Models are usually one dimensional [28–32], this means they only consider the compression waves along the axis, from the transducer, through any glue and coupling layer, into the fluid and the reflector. These 1D models suggest maximum energy in the water layer is achieved when the whole sound path acts as a coupled resonant unit and like travelling wave systems [33], matching layers increase the coupling efficiency between layers. However in resonant systems with out of phase connections face-to-face as shown in Fig. 1 matching layers increase both constructive and destructive interference between the layers, and are therefore unhelpful [27]. The design in this paper is based on the concept of allowing individual parts to vibrate freely where possible by using sub-wavelength contact points as shown in Fig. 2b. Multiple small contacts are used to give more structural stability than a single contact. These contacts are not aligned at points of matched displacement on two surfaces, because alignment is currently not practical due to the large number of potential resonances at MHz frequencies. However our 3D models indicate that making contact only at in-phase regions is a feasible aim.

2. Materials and methods 2.1. Chambers and sound transmission from the PZT to the resonance chambers Four springs placed at the corners of a 100  20  3 mm aluminium clamping plate pressed it onto the back face of the 2 mm thick (1 MHz nominal resonance thickness) 20  20 mm PZT (Pz26, Meggitt A/S, Kvistgård, Denmark). The PZT is pressed with a force of 10 N against the resonant duct (see Fig. 2a and c). For the reasons given in Section 1.4, the contact face of the aluminium plate is milled with an array of pin protrusions 0.2  0.2  0.2 mm at 2.5 mm spacing in a square grid, (the same pin-grid is used for all face-to-face contacts with vibrating parts). In addition to mechanical support for the PZT, the aluminium plate carries the live electrical signal to the outer PZT face. A second aluminium plate, with the pin-grid restrains the opposite face of the chamber. One of these plates has a BNC connector fitted at one end, and acts as a ground plate, the other plate is live with the signal transferred by a long spring-loaded-screw to the central BNC pin. When a glass (non-conducting) chamber is used, a second long spring loaded screw makes the ground connection to the chamber side of the PZT. Three types of chamber were used, each cut to 100 mm length from extruded duct: Chamber I was a rectangular borosilicate glass duct with an internal section 30  3 mm, wall thickness 1.5 mm (Vitrocom, Mountain Lakes, NJ, USA) volume 9 ml (Fig. 2a and c). Chamber II was a square section aluminium duct internal section 19  19 mm, wall thickness 3.25 mm (RS Components, Corby, UK) (Fig. 2b) volume 36.1 ml. Chamber III was an aluminium duct similar to chamber II but with a wall thickness reduced (milled) uniformly to 1.5 mm. The PZT was pressed against the lower sections of all ducts, 20 mm above the lower end. To increase vibration freedom of the aluminium duct and PZT, the grid of pin protrusions was cut into the duct at the contact area. Chamber I, the glass duct, was used in two configurations: (a) The PZT was pressed against the 30 mm wide flat wall; (b) The PZT was pressed against the 6 mm curved wall.

262

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

Fig. 1. Visualisation by finite element simulation of wave-shapes on freely-suspended aluminium and PZT plates vibrating at the same frequency. (a) Full view of each plate. (b) Section view in near contact, arrows indicate regions of, mismatched (180°) and matched (0°) displacements. Contacts at these points would lead respectively to destructive and constructive interference. In this example the PZT is vibrating in a compression mode and the aluminium is vibrating in a flexural plate mode. The diagram is not a quantitative representation, displacements are exaggerated. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

2.2. Fluid connections to the chamber The seal between the chamber and the support housing is formed with an ‘O’ ring over the end of the duct (24.5 mm ID, 30.5 mm OD on the aluminium duct), and confined by a ridge on the Perspex housing. To ensure the flow leaving the duct is not distorted in any direction, it travels first through a 1 mm gap moving radially outwards to a pressure equalization volume before entering the outlet port. Fig. 2d shows the equalization volume as a 2  2 mm, 50 mm diameter ring shaped duct in the upper block. A 4 mm outlet connects to the ring. The top Perspex layer provides a clear window into the resonating duct. In the lower connecting block another ring brings fluid in (almost equally) to all sides of the duct from a 4 mm inlet. Fluid moves radially inwards through a 1 mm gap to the glass duct, entering on all sides. A funnel cut in the lower block carries concentrated sample ‘sediment’ from the base of the resonant duct to a 6 mm outlet. 2.3. Fluid circuit

Fig. 2. Glass and aluminium filtration ducts and manifold (a) glass duct in operation, viewed from above at an angle of 30°. (A) Glass duct with lines of yeast cells along its length positioned by the sound field; (B) PZT pressed against the left hand duct wall; (C) Grid pins on the horizontal aluminium plate press against right hand duct wall; (D) BNC connector; (E) connecting screw taking power to right hand plate; (F) Clamping springs. (b) Aluminium duct. Square section aluminium duct outer dimensions 25.4  25.4  100 mm, the grid of milled pin protrusions at the lower end act as mini horns connecting with the PZT driver. (c) and (d) Manifold for the resonant glass duct indicating inlet and outlets, clamps against wide glass faces. (c) Schematic of duct inlet and outlets (front clamping plate not shown). (d) Drawing of Perspex housing for the glass duct and aluminium clamping plates. (Glass duct and screws not shown).

The inlet and clarified outlet were connected to 1.5 mm ID PVC tubing. The concentrated outlet was connected to 4 mm ID silicone tubing. Tubing length from sample beaker to chamber was <300 mm. Tubing length from chamber to spectrometer <200 mm. The sample beaker was constantly stirred with a magnetic stirrer. Two channels of a peristaltic pump (MINIPULS 3, Gilson, Middleton, USA) were used to draw fluid from both outlets at a fixed Clarified: Concentrated ratio of 1:1.5. The flow rate from the upper (clarified outlet) used in these experiments was 0.33 ml s1, therefore the average flow velocity was 0.9 mm s1 through the aluminium chambers and 3.7 mm s1 through the glass chamber. Peristaltic pulses to the clarified fluid outlet were smoothed with a plenum chamber (a bottle containing 80 ml air and 10 ml fluid) with narrow tubing on the duct side of the bottle [34], these act like a capacitor and resistor forming a low-pass filter to smooth pulses. 2.4. Drive electronics A sine wave was generated by a function generator (33120A, Agilent, Santa Clara, USA), driving an amplifier (ENI 240L, Rochester, NY, USA), which was connected to the PZT. The peak-peak voltage was measured close to the PZT with an oscilloscope (54621A,

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

Agilent). The oscilloscope was also used to measure the analogue temperature output from the IR probe. Software written in LabVIEW (National Instruments, Austin, TX, USA) was used to: control the function generator (GPIB); take measurements from the oscilloscope (GPIB) and the spectrophotometer (RS232C serial); and construct frequency spectra and time course graphs for the filtration efficiency. 2.5. Sample preparation Degassed deionised water was used to produce samples (degassed and deionised water is not essential but was always used to provides a standard medium and avoid the possibility of bubbles forming on the side of the duct which would disrupt the resonance and filtration). Water was degassed by boiling for 3 min. Degassed water was stored in full airtight bottles and used within two weeks. Yeast suspensions, with a concentration of 0.004 g ml1, were produced using dried baker’s yeast (Fast action, Sainsbury’s Supermarkets Ltd., London). Samples were produced by mixing dried yeast with degassed deionised water which was then placed in an ultrasonic bath for 2 min to break up clumps. The number of cells present in such a sample was found to be around 1  108 cells ml1. The average cell radius of these yeast in deionised water is 2.3 lm [18]. A 150 ml sample was recirculated during all experiments except for the temperature measurements. Red water/glycerol solution used for observing Chladni figures. This coloured solution consisted of 1% detergent (Decon 90, Hove, UK), 0.1% red dye (New Coccine, Sigma–Aldrich), 30% glycerol (Sigma–Aldrich) and 68.9% deionised water. The liquid was spread to an even depth of 1 mm before the ultrasound was turned on. Detergent is used to reduce surface tension and allow an extensive flat surface, glycerol increases the viscosity allowing a greater depth to be used without draining away. 2.6. Cell concentration and filtration efficiency The absorbance of the yeast suspensions was measured at 470 nm with a spectrophotometer (6305 UV–Vis, Jencons, Franklin, USA) and a flow-through 0.5 mm light path, quartz cuvette (170.700-QS, Hellma Analytics, Müllheim, Germany). There is a slight peak in the absorbance spectrum at 470 nm, however, the yeast absorbance spectrum is essentially flat over a wide range of wavelengths so this value is not critical. Absorbance measurements made at the filter’s clarified outlet were converted to cell concentration with the following calibration procedure: The cell concentration of an initial sample was measured by counting, using a microscope and counting chamber (Neubauer Haemocytometer, Marienfeld, Lauda-Königshofen, Germany). The absorbance of this sample was then measured by filling the cuvette by syringe. The initial sample was then subjected to a series of serial dilutions. The absorbance of each of these diluted samples was measured and these were used to construct a calibration curve that was used to convert the absorbance measurements to cell concentrations. The absorbance was found to correspond to concentration in a linear manner up to a cell concentration of 1  108 cell ml1 (the initial sample). A correlation coefficient of 0.997 was determined using linear regression. The slope of 0.41 absorbance units per 1  108 cell ml1 was used to calculate all cell concentrations in this paper. Filtration efficiency Filtereff was calculated using the following equation:

Filtereff ¼



1

 cout  100 cin

ð1Þ

263

where Filtereff = Filtration Efficiency; cin = concentration of cells flowing into system; cout = concentration of cells flowing out of the upper (cleaned) outlet. 2.7. Frequency spectra The process of running the system at one frequency, measuring a parameter and stepping up to the next frequency was automated with LabVIEW to produce frequency spectra. Spectra were produced for: voltage on the PZT; filtration efficiency; and temperature. Filtration efficiency measurements were made at the chamber’s clarified output. It should be noted that the measured frequency for maximum efficiency is not actually the frequency at which the clumping takes place due to the time difference between the filtration and the sample reaching the detector. No correction for the time lag has been made in the results presented here. 2.8. Temperature measurements Temperature was measured with an IR probe with a 0.8 mm sensing spot (CT-SF15-C3 with close-focus lens CF15, MicroEpsilon). The sensor was fixed 10 mm from the surface being measured. PZT temperature was measured on the surface facing away from the duct. Chamber temperatures were measured on the face opposite the PZT, 10 mm higher than the PZT top edge. To retain some of the radiated heat from the duct the chamber was wrapped in one layer of aluminium foil with a 5  5 mm gap at the measurement area. The sample was not recycled to ensure the input fluid temperature remained constant. 2.9. Modelling USW structures The ducts were modelled using the finite-element package, (Abaqus CAE, Simulia, Dassault Systèmes, Paris), which integrates modules providing drawing, abstraction as a mesh of finite elements, calculation and visualization. The solver used was Abaqus Standard, and this together with the material values in Table 1 were used to find the frequency and shape of the model’s vibrational modes. Water was modelled as an acoustic medium. Within the duct the water is simulated using finite elements, and beyond it as an infinite sphere of an acoustic medium, centred on the centroid of the water within the duct. 2.10. Laser vibrometer measurements A scanning laser vibrometer was constructed, consisting of a detector head and controller (OFV534 and OFV 2570 HF, Polytec, Waldbronn, Germany), connected to an oscilloscope (WaveRunner 64xi, LeCroy, Teledyne, NY, USA), together with our own LabVIEW code that was developed for: controlling two motorised stages (NT58-568, Edmund Optics, Barrington, USA) to form an X–Y stage; controlling an arbitrary waveform generator (33220A, Agilent) for the excitation; reading the oscilloscope output; and calculating the Fast Fourier Transform to find the magnitude and phase of the vibrations at each point. Using the X–Y stage, a map was built from an array of measurements (20  50 mm with a resolution of 1 mm) made over a period of 45 min. The software calculated a 50–300 kHz linear chirp which the 33220A generated in burst mode, to ensure that each sweep started at the same phase angle. The chirp period was 14.6 ms long, limited by the function generator’s 216 points per waveform, and a decision that the wave should have a precision of 15 points per cycle at the maximum frequency (300 kHz). The chirp was generated repeatedly with a period of 100 ms. The chirp duration

264

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

Table 1 Material values used for the model.

Aluminium Glass Water

Density kg m3

Elastic modulus GPa

2700 2300 1000

70.76 64.00

Bulk modulus GPa

Poisson’s ratio 0.3375 0.2000

2.20

is long enough for resonances to reach maximum amplitude and the interval is long enough for the excited vibrations to decay (ring-down). The rise and ring-down time to 1/e (37%) is given by Q-factor/p [11,35]. If we assume the Q-factor for the water filled duct is 100 then at 50 kHz, the decay (and rise) time is 0.64 ms, and at 300 kHz the decay time is 0.11 ms. The chirp was amplified (ENI 240L) and applied to the PZT on the duct. Each measurement is the average of the response to 10 chirps. 3. Results In preliminary experiments it was found that the pressure contact method with non-specific pin positions introduced large variability and adding water to the pin contact area stabilized the output. Therefore to ensure consistency, water was placed on the contacts for all the results presented here. It was later found, after running the aluminium ducts for several hours with water on the pin contacts a high efficiency filtration was always obtained without water. When these systems were dismantled there was a pattern suggesting corrosion may have removed out of phase contacts, experiments have not been carried out to further demonstrate that selective corrosion is actually the process which stabilises the system. 3.1. Resonance the full length of the chamber Although the PZT was only in contact with a 20 mm section near the lower end of the glass duct Figs. 2a and 3b show bands of particles the full length of the chamber when the PZT is pressing against the narrow side, and also when the PZT is pressing on the wide face of the chamber the clump formation seen in Fig. 3c occurred throughout the chamber. These particle patterns are seen with or without flow in the chamber, indicating that the whole

glass duct is resonating and not just the region near the PZT. The shape of the bands varies slightly with frequency, but at most frequencies the bands of particles collected at the nodal planes parallel to, the face of the PZT and the opposing clamped wall. This parallel band arrangement is clear in Fig. 3b, but is less obvious in 3c where stacks of clumps have formed across the narrow axis of the chamber and only the closest clump is seen from the viewing direction (the model, Fig. 3d and e formed similar patterns and are discussed in Section 4.1.2). The bands form within 1 s of the sound being turned on, the diameter of the clumps increases over the following 30 s as particles within the bands move towards the clump locations. 3.2. Ultrasound enhanced sedimentation The banded appearance seen after 1 s in Fig. 3b is the result of horizontal particle movement into the nodal planes (maximum distance 0.375 mm). Clumps form with particles moving parallel to the nodal planes (maximum distance 15 mm). The acoustic traps responsible for the migration into clumps are thought to be created by variations in field intensity (the movement may be a combination of lateral forces and acoustic streaming of the fluid [27,36]). At the upward flow velocity used here (3.7 mm s1) most clumps were trapped so their diameter increased as new cells joined from suspensions flowing into the chamber. The clumps begin to sediment when they reach a size where the force due to gravitational acceleration exceeds the acoustic potential well and the upward drag force from the fluid [11,18]. In Fig. 4b the cell clumps can be seen ‘‘raining down’’, this view of the sedimentation at the base of the duct below the PZT was seen shortly after sedimentation started (before the view became obscured with falling clumps). The typical sedimentation velocity measured by checking video playback is 6.7 mm s1 (3.7 + 3 mm s1). This compares well with the calculated, maximum sedimentation possible of 10 mm s1 at 1 MHz [11]. 3.3. Flow paths through the chamber No turbulence due to the flow through the glass chamber was observed at any of the flow rates used, however within 1 min of the sound being turned on, a convection current starts in the glass duct, as seen in Fig. 4a. The flow appears to be driven by heat transferred from the PZT. Initially band disruption was seen along the

Fig. 3. Glass duct filled with yeast cells, sound near 1 MHz. (a) Sound off. (b) PZT on left side. Sound on. Narrow lines of cells formed (throughout the full height of the chamber) (c) PZT against duct’s wide face. Sound on, large clumps of cells formed. (a) and (b) Oblique view from above the chamber. (c) Horizontal view facing the chamber. (d) Model of system (b) with PZT and fixed boundary against narrow sides near the duct base. (e) Model of system (c) with PZT and fixed boundary against wide faces near duct base. At high sound levels particles collect at the pressure node regions.

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

Fig. 4. (a) Thermal convection (indicated by the arrow) in the glass duct seen 30 s after the sound was turned on. The disruption of particle alignment indicates the location of the circulation-cell which draws fluid from the centre of the duct to the rising fluid at the left wall. (b) View at lower end of the glass duct. 30 s after switching sound on. PZT against duct face. Clumps sediment against the upward flow.

wall adjacent to the PZT, a few seconds later a horizontal stream appeared, moving towards the lower edge of the PZT to feed the additional upward current and destroying lines of cells as it passed. Similar convection is assumed to occur in the aluminium duct. Preventative measures for this thermal convection are given in Section 4.2.2. Initially, filling the glass duct was difficult since fluid took the short flow path from the narrow duct edges closest to the flow equalisation ring (shown in Fig. 2d), leaving large air bubbles trapped in the gap above the longer glass walls in the outlet block. These bubbles were dislodged by tilting and adding detergent to a filling fluid which was replaced by the sample fluid before use. This problem was overcome by using a rectangular flow equalization channel at the same distance from the duct wall at all points. 3.4. Frequency spectra When a PZT is driven by the 50 X output impedance of the ENI amplifier, a voltage spectrum measured with an oscilloscope across

265

the PZT is nearly proportional to the PZT conductance and reveals resonances as minima and maxima pairs (these are close to the admittance series and parallel resonances [35]). The PZT’s spectrum, shown in Fig. 5 has two strong voltage maxima at 1.13 and 3.41 MHz, these correspond with the fundamental and third harmonic compression modes for the PZT [37,38]. The duct’s filtration efficiency spectrum in Fig. 5a indicates that filtration only occurs in the region of these two PZT resonances. We note that the filtration spectrum in Fig. 5a has three features which should be attributed to the system configuration and not the filtration efficiency of the duct under test, these are: (1) The base line rises across the spectrum, this is because cells and clumps settle in regions of low flow velocity (such as the flow equalization ring) when filtration efficiency is low. Therefore the concentration of the re-circulating sample was gradually depleting. (2) The width of the filtration peaks are wider than the resonance indicated by the voltage spectrum. (3) The peaks in filtration do not occur when the filter is at resonance but a short time later when the frequency is higher. Both 2 and 3 occur because the scan rate was too high for samples to leave the chamber before the following measurement. The frequency range is divided by ten in Fig. 5b, this allows a lower scanning rate so that more peaks are resolved without increasing the base line drift. In this finer resolution spectrum filtration efficiency reaches a higher level and follows the voltage variations more closely. In addition to following the approximate profile of the voltage scan, this filtration spectrum has peaks separated by 38 kHz which corresponds well with a 39 kHz fundamental of the 19 mm wide water filled duct. By progressively decreasing the frequency range of the measured spectrum, 1.115 MHz was selected as the optimum operating frequency. This frequency was used for all subsequent fixed frequency measurements. 3.5. Time course for filtration efficiency When the ultrasound was turned on, the output from the filter cleared progressively up to a stable level. The time course of this clearing is plotted in terms of filtration efficiency in Fig. 6 for both 0.35 and 0.6 Vp–p applied to the amplifier. The time to remove a fluid volume equal to one chamber volume from the clarified outlet is 110 s at the flow rate of 0.33 ml s1. A useful calculation would be the time required to displace all the current fluid in the duct (this is the time to reach maximum efficiency) but this has not been calculated since it is complicated by the parabolic flow profile

Fig. 5. Frequency dependence. 3.25 mm wall aluminium duct. Flow rate 0.33 ml s1 = 1 chamber volume replaced every 109.4 s, 0.27 Vp–p at amplifier input. Thick solid line = voltage measured at the PZT. Dotted line = filtration efficiency. Thin solid line = temperature change of piezoelectric element. (a) Scan frequencies 0.9–4 MHz, 1800 steps, 2 s per step 94.2 kHz for every chamber volume of fluid replaced. (b) Frequencies 0.9–1.3 MHz, voltage data expanded from shaded region in of (a); filtration efficiency spectrum 500 steps, 2 s per step (note the higher peak than in (a) due to the finer scanning resolution as discussed in the text), 32.8 kHz for every chamber volume of fluid replaced; temperature spectrum of the PZT 400 steps 0.2 s per step.

266

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

Fig. 6. Cleanup efficiency of 1.5 mm thick wall aluminium ultrasound filter. The input sample was yeast at a concentration of 0.004 g ml1. Voltages supplied to the amplifier 0.6 Vp–p (continuous line) and 0.35 Vp–p (dotted line), frequency 1.115 MHz, flow rate 0.33 ml s1.

and the development of the sedimentation profile in the chamber. However, Fig. 6 shows that at a flow rate of 0.33 ml s1 filtration reached stable levels within 6 min, with efficiencies of 93% at 0.35 Vp–p and >99% at 0.6 Vp–p. Therefore to ensure this stable condition was reached subsequent measurements of filtration efficiency were made 15 min or more after the sound was turned on. 3.6. Temperature and energy efficiency of filtration The operation of the ultrasonic filter causes some heating of the fluids within the duct. This temperature rise is used here to find the energy dissipated during the filtration process. This represents the minimum energy needed for filtration assuming all the electronic control uses no energy itself. In our system the electronic control, the signal generator and amplifier probably use more than 100 times the energy supplied to the filter but we know this can be reduced substantially by matching the drive circuit to a particular filter. We are aiming to measure the transformation of ultrasound energy to heat to calculate the efficiency of the chamber (and not the drive system). Since measured temperature changes are small (<1 K) most losses will be from the fluid leaving the duct. The suspension was not recycled during temperature measurements, in initial measurements recycling retained considerable heat which increased the input temperature, these results were discarded and an open-ended flow circuit was introduced. The temperature time-course plots in Fig. 7 are the temperatures from three consecutive tests. A much larger temperature rise of around 1 K was recorded with the first runs (open circles in the figure). The reason for this could be due to the aluminium duct warming up during the experiments and retaining some of this heat. For the calculation below we use the first run value of a 1 K temperature rise after which the temperature stabilized and a steady state energy balance applies. Measurements of the PZT temperature showed the same first run trend. To calculate the energy dissipated we assume that heat loss to the air is small compared to heat loss to the water and therefore, when the temperature reaches a stable value

heat generated ¼ heat removed by water flowing through the system ð2Þ At the stable equilibrium temperature, power consumption of the filter P (W), can be calculated from:

P ¼ Q qcDT

ð3Þ

Fig. 7. Time course of temperature change in the 1.5 mm wall aluminium duct filter; frequency 1.115 MHz; input sample, 0.004 g ml1 yeast; flow rate out from the clarified outlet 0.33 ml s1; voltage applied to the amplifier 0.35 Vp–p. Data represents consecutive ‘‘runs’’.

where Q = flow rate (3.3  107 m3 s1); q = density (998 kg m3); c = specific heat capacity (4183 J kg1 K1); DT = Duct temperature – input temperature (K). By taking the thermal equilibrium level to be 1 (±0.2) K above ambient (level after 1000 s in Fig. 7), the filtration energy requirements of the duct have been calculated and are given in Table 2. Much of the heat is generated within the PZT (see below) and conducted through the pin contacts to the chamber. Therefore the true efficiency of the duct alone is higher than the values in Table 2. There is scope for efficiency improvements to the duct but also for the system as a whole which we have not made any attempt to optimise. In the current system the hierarchy of heat generators is the electric drive (a broadband amplifier) followed by the mechanical drive (a PZT) with the chamber contributing the least. The theoretical minimum energy requirement has not been calculated for any stage but reductions of >90% for each stage can probably be attained. 3.7. PZT resonance and temperature The temperature of the PZT was measured by focusing the IR probe directly onto the PZT. Temperatures recorded on the piezoelectric element are shown in Fig. 5b. These measurements can be compared with the voltage and filtration efficiency spectra, this gives a rough indication of filtration efficiency and power used, but the spectra were made with no off period between each frequency and therefore some cumulative temperature will be present. The spectrum was measured in steps from low to high frequencies, therefore the temperatures at higher frequencies are influenced by the immediately preceding lower ones. The maximum temperature change measured is 1.5 K within 20 s which is significantly more than the 1 K change seen on the duct with a higher driving voltage over 15 min. This suggests that the PZT generates most of the heat in the system. The small connection points with the duct will reduce but do not eliminate heat transfer to the duct.

Table 2 Duct energy requirements. 1.5 mm wall aluminium duct filter; measured after 15 min; frequency 1.115 MHz; input sample, 0.004 g ml1 yeast; flow rate out from the clarified outlet 0.33 ml s1; voltage applied to the amplifier 0.35 Vp–p. SD = standard deviation based on three replicate runs.

DTemperature K (±SD)

Power W (±SD)

Energy consumption J ml1 (±SD)

Cleanup efficiency % (±SD)

1 (±0.2)

1.4 (±0.28)

4.18 (±0.84)

82.4 (±5.8)

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

267

The temperature maxima are clearly associated with the regions of filtration efficiency but they are not an exact match. So there is scope for identifying an energy-efficient system where the ratio (energy loss/filtration efficiency) is minimized. A spectrum for the duct temperature was not made since duct temperatures are lower and therefore measurement noise from draughts and lighting would make interpretation difficult. 3.8. Wall mode visualization A future development stage for this system will be to make contacts between the PZT and the duct wall only at contact points which lead to constructive interference patterns within the duct. These are the 0° phase difference regions in Fig. 1. To achieve this aim the mode shape on the duct and the PZT must be identified. Although an experimental system with contact only at in-phase locations has not yet been developed, a combination of experimental and modelling mode-visualisation methods are described here. Low frequencies are used for these proof-of-principle experiments since patterns at higher frequencies are extremely complex, and hence hard to interpret; our aim for this stage is to identify methods to experimentally validate and tune the models before advancing to higher frequencies. 3.8.1. Chladni figures and experiment and models Initially the empty 3 mm wall aluminium duct and manifold was placed horizontally and 100 lm diameter glass beads were placed on the duct’s upper surface. When driven at frequencies between 100 and 300 kHz the beads moved to regions of minimal displacement. However, after the first movement a change in frequency produced patterns with gaps where a displacement antinode was replaced by a node as the frequency changed. To observe patterns over a series of frequency steps the glass beads were replaced with a liquid, which gathers in heaps at displacement antinodes. Patterns can be produced with any liquid spread over the surface to a depth 1–2 mm, however to obtain clear patterns we used a red coloured water/glycerol solution (described in Section 2.5). Many patterns were produced as the vibration frequency was changed. A simple pattern, formed at 203 kHz, is shown in Fig. 8a. The same pattern was identified at 199 kHz in the FE model results. Dimensional (fabrication) tolerances on the duct are sufficient to account for this small frequency difference. This agreement between model and experiment is impressive since it was achieved with a model where the only inputs were the material density, bulk modulus, Poisson’s ratio and the geometry. The position of the pin contacts is included in the model (below the grey area in Fig. 8b). The pins do not introduce significant interference for this mode. 3.8.2. Water filled duct When the duct was filled with water, the displacements were reduced and Chladni figures could only be produced at high power levels. Our long term aim is to identify modes (including weak modes) and to select and strengthen a particular mode using pin contacts at in-phase locations between the drive and the duct. Since Chladni figures could not be used effectively for identifying weak modes, laser vibrometer surface displacement maps were selected for identifying modes. Although more expensive, the method gives greater resolution and sensitivity. 2D maps were easily obtained of the water filled ducts for modes inside and outside the main frequency range of the PZT. Maps of many modes are produced by sweeping the frequency, an example of one mode is shown in Fig. 9a. Introducing water into the duct modelled in Fig. 8b increases the number of variables which may explain the 10% frequency

Fig. 8. (a) Liquid Chladni figure. Red coloured water/glycerol solution moves to the vibration displacement antinodes on the surface of the aluminium duct vibrating at 203 kHz. (b) An Abaqus model of the duct shows vibration at 199 kHz produce antinodes in the same pattern as those observed in the experiment. Rectangles indicate regions covered by the manifold and PZT clamp. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

difference between the model and the laser vibrometer measurements 55.60–60.53 kHz in Fig. 9a and b. and their only partially similar wave shapes. However development of this will give the significant advantage of predicting the pressure distribution inside the opaque duct (Fig. 9d) from a corresponding vibrational pattern on the duct surface. 4. Discussion 4.1. Sound propagation 4.1.1. Coherent single-element structures The single-element ducts used here are a variation on the glass capillaries which have been used for many years to form small acoustic chambers [39–46]. The duct resonates throughout its whole length, as also achieved over the shorter lengths of some of the small capillaries [39,40,43,44,46]. The lower harmonics of some modes in a near-one or two dimensional structure (such as a guitar string and a glockenspiel plate) can be predicted from analytical equations [47], these predictions become more difficult with higher frequencies, higher harmonics or more dimensions but Araz et al. [40] have successfully predicted the node positions at MHz frequencies in an almost one dimensional system (a narrow capillary). Although 3D models are still too complex for forming designs with explicit structural-modes, by using simple single element ducts, some agreement between models and experiments is seen in Figs. 3 and 9 indicating that selection of the structural mode shape may be possible, even in large systems at MHz frequencies. 4.1.2. Fixed contacts determine the axis of propagation In the absence of any interface contacts, travelling wave vibration energy falls with distance from the source, whereas in

268

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

Fig. 9. Measured and calculated vibrations of the water filled 1.5 mm wall aluminium duct. (a) Scanning laser vibrometer map of duct surface displacements at 55.60 kHz. (b) Abaqus finite element model of duct surface displacements at 60.53 kHz. The model predicts that this mode gives the greatest displacements over the 45–65 kHz range and it is also the best fit to the measured mode shape (see (a)) in the range. Grey horizontal bands indicate the position of the PZT and clamp pins. (c) Duct end-view with contours plotted on deformed model (deformation not to scale). (d) Pressure in the water at the centre plane. Scales in c and d indicate relative values not true quantities.

resonant systems the source is not necessarily the region of highest energy density. However in continuously driven resonant systems the direction of wave propagation from the source is usually the axis of the strongest nodal planes giving the illusion that energy falls with distance from the source. The illusion arises because in most systems there are multiple interface and contact regions (as in Fig. 1) that cause losses from destructive interference. There are only a few contact regions in our ducts (rubber seals and pin connectors) and Figs. 2–4, 8 and 9 show vibrations extending far beyond the region of the PZT. However in the glass duct, where bands can be seen, Figs. 2–4 still give the inaccurate impression that the propagation direction is important because bands of yeast cells formed lengthways in the duct, parallel to the PZT and the clamping plate. The same direction for the nodal planes was seen in both the experiment Fig. 3b, and the model (Fig. 3e). When the plane of the PZT and the non-moving clamp was turned through 90° the plane of the bands followed the rotation (Fig. 3c and e). Although this appears to indicate that nodal planes are stacked along the axis of initial sound propagation, our explanation for the node alignment is that the clamping partially fixes one wall reducing its displacements and this allows duct breathing [48] but reduces flexing. Breathing transfers compression waves into the fluid, parallel to the fixed wall. This conclusion is supported by another model (not shown) where, the driving wall is the large face and the fixed wall is the narrow wall at right angles, in this case nodes form parallel to the fixed wall, similar to Fig. 3b and d. (The plane of the nodes has no importance for UES systems).

4.1.3. Acoustic energy is transferred regardless of the mode in the wall and the fluid UES filtration increases with the vibration energy in the fluid. Vibration energy is produced by the PZT at its fundamental and odd harmonic (1 and 3 MHz), and these are the only frequencies where filtration is seen in the rough filtration spectrum Fig. 5a. Closer inspection of the 1 MHz region (Fig 5b) reveals some fine detail in both the voltage and filtration efficiency spectra. This detail appears to be resonances from three superimposed sources; the pzt (series and parallel resonance), structural (duct wall) modes, and the fluid cavity. The step between resonant frequencies of the cavity width is 39 kHz (=c/k = 1500/(2  0.019)). At 1 MHz, many structural mode resonances are present in freely suspended

ducts of this size but our duct is not completely free and energy is also lost more readily from resonances in the large surface to volume ratio of the wall than the compact fluid. Therefore wall resonances are the smallest of the three sources and only resonances corresponding with the PZT and cavity resonances have been identified from the spectra in Fig. 5b. 4.2. Flow paths Flow passing parallel to a clump or band of yeast cells will not easily break its structure. However, cross-flow as low as 200 lm s1 will move yeast cells away from the nodal regions and destroy the bands [11,49,50]. Since most bands of yeast cells form parallel to the duct wall it is important to minimise crossflow within the duct. The two main potential causes of cross-flow are: turbulence due to flow and circulation due to thermal convection. The probability of either disruption is estimated below. 4.2.1. Turbulence due to flow Turbulence due to flow will only occur in the central region of long straight ducts when the Reynolds number is large, usually 500–2000 [11,51]. At the flow rate used, 0.33 ml s1 the Reynolds numbers were 6.2 for the glass duct and 19.3 for the aluminium duct, therefore flow-induced turbulence is unlikely in the central region of the duct. At the ends of the duct flow lines will not be parallel and there could be some attached vortices. The maximum entrance length [11,52,53] for these distortions to become fully formed laminar flow is calculated to be 1.4 mm for the glass duct and 15.3 mm for the aluminium duct. In our duct, parallel flow lines are expected to develop in a shorter distance since flow enters and leaves from all sides of our ducts and this should produce only minimal initial cross-flow. Therefore it is unlikely that acoustic formations are disrupted by turbulence arising from fluid flow. 4.2.2. Turbulence due to thermal convection Turbulence due to thermal convection in the glass duct, Fig. 4a and Section 3.3, usually forms a rotating cell of fluid within 1 min of the sound being turned on and appears to be driven by heat from the PZT. The direction of the flow is towards the driven wall and runs up this face, consistent with thermal convection, whereas Eckart streaming [36,54,55], which is seen in similar systems

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

[21] moves away from the driven face and is considered an unlikely mechanism for this streaming. Heat from the PZT could be removed by air cooling (a fan), however two design improvements which do not use additional energy also appear possible. One is to reduce the temperature of the duct by minimizing contact between the PZT and the duct. This could also improve resonance as discussed in Section 1.4. Another possibility is to increase the temperature gradient threshold for both chaotic and stable convection which can be achieved by reducing the duct height. Convection is usually considered to begin when the convection to conduction ratio exceeds 1, the ratio is known as the Nusselt number (Nu). A single convective stream is not very disruptive for UES but horizontal movement caused by multiple circulations or chaotic convection is disruptive. The breakup of the single convective stream has been observed in systems with Nusselt numbers as low as 10 [56]. A method to calculate Nu is available for a channel such as ours where one side wall is held at a higher temperature than the opposite wall [11,57]. For our 100 mm high chambers this calculation gives Nu = 10 for a temperature difference of only 0.05 K across the aluminium duct and 0.1 K across the glass duct. Temperatures measured here (Section 3.6) suggest the gradients across these channels will exceed 0.1 K, therefore there is a risk of chaotic convection, and a certainty of the orderly convection as seen in the glass duct Fig. 4. Temperature gradients below 0.1 K would be difficult to achieve experimentally but 1 K is an achievable gradient with a high acoustic power level (<0.5 K variation rise is seen in Fig. 7 after the initial 15 min). By reducing the height of the aluminium duct to 35 mm, a Nu below 10 is obtained with a 1 K temperature gradient. It is thought that a 35 mm sound field is sufficient to form clumps for UES, since most UES filters are less than 50 mm high so this height reduction is unlikely to bring adverse consequences. 4.3. Wall thickness Two wall thicknesses were tested with the aluminium duct, 3.25 and 1.5 mm. These were selected because at 1 MHz in the compression mode these are ½ and 1=4 wavelengths. For travelling waves at normal incidence, transmission through a thin plate is highest through ½ wavelength plates while 1=4 wavelength plates form the best reflectors [33,34]. The 1=4 wavelength wall duct was found to work best so we only report the filtration results for this duct. The wall thickness is clearly influential, but this comparison of these two wall thicknesses was not carried out rigorously and it is probable that in our system many other factors are more significant than reflection at normal incidence. 5. Conclusion The ducts used for this work were standard parts. Contact pins were milled into the aluminium duct but for the glass duct the slight curve allowed face to face contact with the PZT without any milled pin contacts. These non-optimised parts demonstrate that cooling is not essential for scale up giving 99% filtration efficiency with the aluminium duct (Fig. 6). The need for water on the contacts during an initial bedding-in period does however indicate that more work is needed to develop the contact method. The 4 J ml1 energy dissipated, can almost certainly be reduced by increasing the ducts freedom of vibration (see Section 1.4.) and more simply by reducing the height of the duct to reduce convection (see Section 4.2.2). Energy efficiency and excess heat production can be both overlooked in small systems but they become vital considerations when systems are scaled up. Here the steps taken towards reducing excess heat are, the use of single element chambers to

269

reduce interfering wave patterns and the replacement of face-toface contacts by pin contacts (see Section 1.4 and Fig. 2) The extent of free movement achieved is confirmed with 3D models and wave-shape visualisation methods (see Sections 3.8.1 and 3.8.2). Although models and laser vibrometer measurements are initially both lengthy steps, they form a convenient pair for iterative development. The single element duct structures described here are clearly amenable to this high precision optimisation approach. The intention is to use models and node visualisation extensively in future developments. For example, one aim is for contact regions to be placed only at nodal lines of the vibrating structure. In addition, matching the structural and fluid mode shapes may be possible. The acoustic limit to scale-up is not known but since particles are required to move less than 1 mm in a sound field to achieve filtration (and other similar functions) the energy required per particle is low compared to systems such as centrifuges and membrane filters. With further optimisation standing wave systems should become the energy efficient option for many large scale filters and particle manipulators. Acknowledgement The experimental work reported here was funded by the Carbon Trust, UK. For the Algal Biofuel Challenge. Grant URN 033. References [1] H.J. Kim, B. Han, W.S. Hong, W.H. Shin, G.B. Cho, Y.K. Lee, Y.J. Kim, Development of electrostatic diesel particulate matter filtration systems combined with a metallic flow-through filter and electrostatic methods, Int. J. Automot. Technol. 11 (2010) 447–453. [2] D. Wild, The Immunoassay Handbook: Theory and Applications of Ligand Binding ELISA and Related Techniques, fourth ed., Elsevier Science, 2013 (4th Revised edition). [3] D. Vigolo, R. Rusconi, H.A. Stone, R. Piazza, Thermophoresis: microfluidics characterization and separation, Soft Matter 6 (2010) 3489–3493. [4] H. Zhao, E.S.G. Shaqfeh, V. Narsimhan, Shear-induced particle migration and margination in a cellular suspension, Phys. Fluids 24 (2012). 011902-011901– 011902. [5] T. Chiane, N.E. Assidjo, P.J.P. Cardot, Sedimentation field-flow-fractionation: emergence of a new cell separation methodology, Talanta 51 (2000) 835–847. [6] I.D. Johnston, M.B. McDonnell, C.K.L. Tan, D.K. McCluskey, M.J. Davies, M.C. Tracey, Dean flow focusing and separation of small microspheres within a narrow size range, Microfluid Nanofluid 17 (2014) 509–518. [7] R.W. Barber, D.R. Emerson, Recent advances in electrowetting microdroplet technologies, in: P. Day, A. Manz, Y. Zhag (Eds.), Microdroplet Technology, Springer, New York, 2012. [8] D. Chakraborty, G.S. Sudha, S. Chakraborty, S. DasGupta, Effect of submicron particles on electrowetting on dielectrics (EWOD) of sessile droplets, J. Colloid Interface Sci. 363 (2011) 640–645. [9] A. Barron, A. Kar, T. Aspray, A. Waddie, M. Taghizadeh, H.T. Booke, Two dimensional interferometric optical trapping of multiple particles and Escherichia coli bacterial cells using a lensed multicore fiber, Opt. Express 21 (2013) 13199–13207. [10] S. Patel, D. Showers, P. Vedantam, T.-R. Tzeng, S. Qian, X. Xuan, Microfluidic separation of live and dead yeast cells using reservoir-based dielectrophoresis, Biomicrofluidics 6 (2012). 034102-034101–034102-034112. [11] J.J. Hawkes, S. Radel, Acoustofluidics 22: multi-wavelength resonators, applications and considerations, Lab Chip 13 (2012) 610–627. [12] M. Ohlin, A.E. Christakou, T. Frisk, B. Önfelt, M. Wiklund, Influence of acoustic streaming on ultrasonic particle manipulation in a 100-well ring-transducer microplate, J. Micromech. Microeng. 23 (2013) 11. [13] D.G. Sadikova, T.N. Pashovkin, Cell concentration and separation in the field of a standing ultrasonic wave for medicine and biotechnology, Open J. Biophys. 3 (2013) 70–75. [14] STS90 Technical data sheet, BioSep: the advanced acoustic cell retention device, in: AppliSens (Ed.), Applikon Dependable Instruments bv, Schiedam, The Netherlands. [15] L.E. Kinsler, A.R. Frey, A.B. Coppens, J.V. sanders, Fundamentals of Acoustics, fourth ed., John Wiley and Sons Inc., New York Chichester, 2000. [16] Tables of Physical & Chemical Constants (16th ed., 1995), Kaye & Laby Online, Version 1.0, in, National Physics Laboratory, 2005. [17] F. Trampler, S.A. Sonderhoff, P.W.S. Pui, D.G. Kilburn, J.M. Piret, Acoustic filter for high density perfusion culture of hybridoma cells, Biotechnology 12 (1994) 218–284. [18] J.J. Hawkes, M.S. Limaye, W.T. Coakley, Filtration of bacteria and yeast by ultrasound enhanced sedimentation, J. Appl. Microbiol. (Formerly J. Appl. Bacteriol.) 82 (1997) 39–47.

270

J.E. Prest et al. / Ultrasonics 56 (2015) 260–270

[19] M. Wiklund, Acoustofluidics 12: biocompatibility and cell viability in microfluidic acoustic resonators, Lab Chip 12 (2012) 2018–2028. [20] J.D. Adams, C.L. Ebbesen, R. Barnkob, A.H.J. Yang, H.T. Soh, H. Bruus, Highthroughput, temperature-controlled microchannel acoustophoresis device made with rapid prototyping, J. Micromech. Microeng. 22 (2012) 1–8. [21] B. Lipkens, J. Dionne, M. Costolo, A. Stevens, E. Rietman, Separation of bacterial spores from flowing water in macro-scale cavities by ultrasonic standing waves, (2010) arXiv:1006.5467v. [22] A. Lenshof, C. Magnusson, T. Laurell, Acoustofluidics 8: applications of acoustophoresis in continuous flow microsystems, Lab Chip 12 (2012) 1210– 1223. [23] V.M. Gorenflo, J.B. Ritter, D.S. Aeschliman, H. Drouin, B.D. Bowen, J.M. Piret, Characterization and optimization of acoustic filter performance by experimental design methodology, Biotechnol. Bioeng. 90 (2005) 746–753. [24] J.J. Hawkes, W.T. Coakley, A continuous flow ultrasonic cell-filtering method, Enzyme Microb. Technol. 19 (1996) 57–62. [25] S. Radel, Influence of biomass, throughput and true electric power input on the separation efficiency of a 60 ml acoustic filter, Elektrotech. Inftech. 126 (2009) 51–57. [26] P. Glynne-Jones, R.J. Boltryk, M. Hill, Acoustofluidics 9: modelling and applications of planar resonant devices for acoustic particle manipulation, Lab Chip 12 (2012) 1417–1426. [27] A. Lenshof, M. Evander, T. Laurell, J. Nilsson, Acoustofluidics 5: building microfluidic acoustic resonators, Lab Chip 12 (2012) 684–695. [28] M. Gröschl, Ultrasonic separation of suspended particles – Part II: design and operation of separation devices, Acta Acust. U. Acust. 84 (1998) 632–642. [29] M. Gröschl, Ultrasonic separation of suspended particles – Part I: fundamentals, Acta Acust. U. Acust. 84 (1998) 432–447. [30] M. Hill, R.J.K. Wood, Modelling in the design of a flow-through ultrasonic separator, Ultrasonics 38 (2000) 662–665. [31] Y.A. Ilinskii, B. Lipkens, E.A. Zabolotskaya, Energy losses in an acoustical resonator, J. Acoust. Soc. Am. 109 (2001) 1859–1870. [32] H. Cappon, K. Keesman, Numerical modeling, calibration, and validation of an ultrasonic separator, IEEE Trans. Ultrason. Ferroelectr. Freq. Control 60 (2013) 614–621. [33] J. Blitz, Fundamentals of Ultrasound, Butterworth, London, 1967. p. 27. [34] J.J. Hawkes, W.T. Coakley, Force field particle filter, combining ultrasound standing waves and laminar flow, Sens. Actuators, B 75 (2001) 213–222. [35] J. Dual, P. Hahn, I. Leibacher, D. Möller, T. Schwarz, Acoustofluidics 6: experimental characterization of ultrasonic particle manipulation devices, Lab Chip 12 (2012) 852–862. [36] M. Wiklund, R. Green, M. Ohlina, Acoustofluidics 14: applications of acoustic streaming in microfluidic devices, Lab Chip 12 (2012) 2438–2451. [37] J. Dual, T. Schwarz, Acoustofluidics 3: continuum mechanics for ultrasonic particle manipulation, Lab Chip 12 (2012) 244–252. [38] C.M. Cousins, P. Holownia, J.J. Hawkes, C.P. Price, P. Keay, W.T. Coakley, Clarification of plasma from whole human blood using ultrasound, Ultrasonics 38 (2000) 654–656.

[39] R.I. Jepras, D.J. Clarke, W.T. Coakley, Agglutination of Legionella pneumophila by antiserum is accelerated in an ultrasonic standing wave, J. Immunol. Methods 120 (1989) 201–205. [40] M.K. Araz, C.-H. Lee, A. Lal, Ultrasonic separation in microfluidic capillaries, IEEE Ultrason. Sympos. (2003) 1111–1114. [41] M. Wiklund, H.M. Hertz, Ultrasonic enhancement of bead-based bioaffinity assays, Lab Chip 6 (2006) 1279–1292. [42] B. Hammarström, T. Laurell, J. Nilsson, Seed particle-enabled acoustic trapping of bacteria and nanoparticles in continuous flow systems, Lab Chip 12 (2012) 4296–4304. [43] G. Goddard, G. Kaduchak, Ultrasonic particle concentration in a line-driven cylindrical tube, J. Acoust. Soc. Am. 117 (2005) 3440–3447. [44] A.C. Sorando, J.J. Hawkes, P.R. Fielden, I. González, Patterns of particles aggregation and streaming in resonating fluids, in: International Congress on Ultrasonics, AIP Conf. Proc., Gdansk, 2012, pp. 757–760. [45] J. Lei, P. Glynne-Jones, M. Hill, Acoustic streaming in the transducer plane in ultrasonic particle manipulation devices, Lab Chip 13 (2013) 2133–2143. [46] P.P.A. Suthanthirara, M.E. Piyasen, T.A. Woods, M.A. Naivar, G.P. Lopez, S.W. Graves, One-dimensional acoustic standing waves in rectangular channels for flow cytometry, Methods 57 (2012) 259–271. [47] R.D. Blevins, Formulas for Natural Frequency and Mode Shape, Krieger Publishing Company Malabar, Florida, 2001. [48] S. Kogan, G. Kaduchak, D.N. Sinha, Acoustic concentration of particles in piezoelectric tubes: theoretical modeling of the effect of cavity shape and symmetry breaking, J. Acoust. Soc. Am. 116 (2004) 1967–1974. [49] J.J. Hawkes, J.J. Cefai, D.A. Barrow, W.T. Coakley, L.G. Briarty, Ultrasonic manipulation of particles in microgravity, J. Phys. D: Appl. Phys. 31 (1998) 1673–1680. [50] M.A. Grundy, Ultrasonic concentration and manipulation of small particles and biological cells, PhD Thesis, Pure and Applied Biology, University of Wales Cardiff, Cardiff, 1994. [51] J.F. Douglas, J.M. Gsiorek, J.A. Swaffield, L.B. Jack, Fluid Mechanics, sixth ed., Prentice Hall, Harlow, England, New York, 2011. [52] H. Bruus, Acoustofluidics 10: scaling laws in acoustophoresis, Lab Chip 12 (2012) 1578–1586. [53] B. Atkinson, M.P. Brocklebank, C.C.H. Card, J.M. Smith, Low Reynolds number developing flows, AIChE J. 15 (1969) 548–553. [54] S.S. Sadhal, Acoustofluidics 13: analysis of acoustic streaming by perturbation methods, Lab Chip 12 (2012) 2292–2300. [55] C.P. Lee, T.G. Wang, Outer acoustic streaming, J. Acoust. Soc. Am. 88 (1990) 2367–2375. [56] N. Seki, S. Fukusako, H. Inaba, Free convective heat transfer with density inversion in a confined rectangular vessel, Wärme-und Stoffübertragung 11 (1978) 145–156. [57] N. Seki, S. Fukusako, H. Inaba, Visual observation of natural convective flow in a narrow vertical cavity, J. Fluid Mech. 84 (1978) 695–704.