Discussion

This section aims to explain the results recorded in the preceding pages and understand the relative importance of the variety of processes which influence particle movement through the separator. The modelled results demonstrate the behaviour of particles under the forces described in section 5.2.1 and also based on assumptions regarding the 1-dimensional nature of the acoustic field and fluid flow profile, not dissimilar to the assumptions adopted for other simulation work discussed in chapter 2. Therefore, disagreement between the experimental and simulated separation results can be attributed to a more complex system of forces and/or significant 2 or 3-d variations with the acoustic and fluid flow fields, some of which have been investigated in detail in chapters 3 and 4 and therefore the discussion also draws from these chapters.

Figure 6.11: Influence of outlet flow proportions on particle separation. Experiment operated usingC0= 2.2e8 particles/ml andQt= 0.005ml/s.

Figure 6.12: Influence of outlet flow proportions on particle separation based on both experimental (black) and predicted (red) data. Experiment operated usingC0 = 2.2e8 particles/ml andQt = 0.005ml/s.

Structure of particle stream

The initial set of experiments investigating the influence of concentrationC0represents the closest comparison between modelled and measured data (figure 6.8) and, indeed, separation of particles is demonstrated. At the lower voltage values experimental error and fluid lift forces explain why the predicted data underestimates the separation slightly, although the particle model incorporating the acoustic impedance transfer model gives a good description of particle movement. However, at increased voltage there is a significant discrepancy between the data sets where the model over- estimates the level of separation.

To begin to explain this discrepancy, it is initially assumed that at high voltages a higher radiation force is experienced by the particles and all particles move into a high concentration particle stream. Whilst the prediction assumes that all particles within this stream pass through outlet 2, the experimental results suggest that part of the particle stream passes through outlet 1. For this to be possible, the particle stream cannot consist of particles all accurately aligned along a uniform pressure node plane, but instead be distributed within a band extending a certain distance either side of the pressure node plane.

Two explanations are put forward. Firstly, simulation of the acoustic field covered in chapter 4 suggests that the occurrence of enclosure modes with the fluid layer will prevent a uniform pressure node plane from forming, the vertical positionyof the pressure node a function of the lateral positionzacross the device. The movement of particles towards zero pressure amplitude locations would result in them being distributed at various y positions. Also, the strength of the acoustic field will influence the rate at which a particle moves towards the nodal plane. For example, if a weaker acoustic field exists towards the lateral edges of the fluid chamber, as shown in chapter 4, figure 4.7 for mode 2, the resulting particle stream may be more dispersed within these areas as compared to the centre of the chamber.

The second explanation is that disturbance of the particle stream will cause particles to become more dispersed about the nodal plane. Such disturbance is likely to originate from acoustic stream- ing which has been observed during subsequent experimental work on the separator. Also, the lit- erature records the occurrence of streaming within similar resonators which influences the move- ment of 1µm and similar sized particles and, as streaming is related to the acoustic energy losses, streaming velocities will increase as a function of voltage.

not disturb the movement of larger particles. This is supported by the separation of larger yeast particles as reported by Harris et al. (appendix 5) where almost total clearance is measured, which indicates that for larger particles the particle stream is not significantly dispersed.

Diffusion of the particles may also limit the maximum concentration of the particle stream as predicted by Higashitani [72] where for a long residence time, diffusion becomes more significant relative to the radiation forces. However, the flow rates used in experiment do not result in a sufficiently long residence time to support the significant influence of diffusion upon the maximum concentration of the particle stream. Even without the influence of the acoustic radiation forces, the movement of particles due to diffusion is negligible (see section 5.2.1).

Separation at high voltage levels

At high transducer voltages the radiation forces also should become high and the formation of a concentrated particle stream more likely. However, as the voltage is increased further other forces or phenomena due to the presence of the acoustic field become more significant; for example, the streaming flow velocities and secondary radiation forces (particle interaction) will increase.

Some experimental results indicate that at 450mVpkpk, particle separation becomes less efficient than at 350mVpkpk. It is possible that at the higher voltage particles begin to agglomerate and disturb the flow, the agglomeration being a combined effect of lateral and secondary radiation forces. This would be consistent with some observations, during preliminary tests with larger particles, e.g. yeast.

The formation of bubbles within the fluid chamber has also been observed. This is associated with the heating of the PZT and separator device which causes expansion of gas trapped in the fluid sample. These bubbles are large enough to significantly disturb the flow and increase the flow velocity through the fluid chamber, thus damaging the separation process.

The presence of a highly concentrated particle stream due to a high voltage and located at the nodal plane may influence the acoustic field, as another layer is effectively present in the device. As mentioned in chapter 2, particles of an acoustically soft material may not significantly alter the acoustic impedance and therefore characteristics of the device, similarly for a low concentration of particles. In the experimental results above there is no evidence that the particle stream is tightly packed with a high concentration, which does not suggest that the particle stream formed at high voltages will influence the acoustic field.

Verticalylocation of particle stream

The third set of experimental results investigating the influence of relative outlet flow rates pro- vides information regarding the approximate location of the particle stream relative to the position which represents the division of flow between the two outlets. For example, if the position where the flows divide was located at y = h0, all fluid passing along the channel within the bounds 0 < y < h0 will pass through outlet 1 and the remaining fluid withinh0 < y < h(where h is the height of the channel) will pass through outlet 2. This is illustrated in chapter 5 figure 5.10. It is reasonable to state that asQ1increases relative toQ2,h0 moves up towards the reflector layer. Therefore, at a certain flow condition,h0will be located above the acoustic node and the majority of particles will instead pass through outlet 1. This is seen in figure 6.11 where forQ1/Qt=0.25 and 0.35 the majority of particles pass through outlet 2, however, this changes forQ1/Qt =0.45 where the majority of particles pass through outlet 1.

This switch between outlets has not been predicted by the modelled results, although the following paragraphs explain these observations. As shown in section 3.6, thisy position representing the division of the outlet flows is not necessarily uniform across the width of the channel due to the geometry of the outlet region. In the experimental system it is instead likely that the region of fluid drawn through outlet 1 extends further in the y-direction (see figure 3.24) than that assumed in the simulation, which is based on a uniform flow pattern. Therefore, whilst the simulation may predict the extraction of the particle stream through outlet 2 under certain operating conditions, in the experimental system the particle stream may instead be contained within outlet 1 flow, as reflected by the data plotted in figure 6.12. (Other flow effects may explain why a greater degradation in separation than expected is seen with an increase in flow rateQt, although no specific explanations can yet be presented.)

Another explanation is that the location of the pressure node has not been accurately predicted and in the experimental system is closer to the silicon matching layer. It is important for the simulation to predict the level of acoustic energy at resonance rather than accurately predict the frequency at which this resonance occurs, therefore the simulation uses the predicted resonant frequency rather than the experimental operating frequency. However, a small error in frequency will influence the predicted acoustic pressure profile and nodal position. Based on simulated data, figure 6.13 illustrates the normalised acoustic pressure profiles at both the experimental operating frequency and the predicted resonant frequency. This figure suggests that the simulation places the node approximately5µm above the true node location and therefore overestimates the minimum outlet

Figure 6.13: 1-d simulation of acoustic pressure profile within fluid chamber at frequencies used for experimental measurements (solid line) and simulated results (dotted line).

1 flow rate (Q1/Qt) at which particles will pass through outlet 1. In general, ’alignment’ error between the acoustic pressure profile and parabolic flow profile will alter particle trajectories.

What can also be seen in this figure is a pressure anti-node near the silicon layer boundary at y= 0where the pressure amplitude reaches a maximum. This suggests that a small proportion of particles will be forced towards this surface and, if the particles do not adhere to the surface, will be carried to outlet 1 with the clarified flow. This explains why the simulated concentration results do not predict complete clearance (figures 6.8, 6.10 and 6.12).

The position of the pressure anti-node based on the experimental frequency also shown in figure 6.13 is even closer to the silicon surface than for the frequency used in simulation. Therefore, the anti-node cannot be expected to cause a significant increase in concentration within the clarified flow and consequently is not the explanation as to why the experimental results do not show complete clearance. There is more evidence to suggest that streaming is the cause.

Forces experienced by particles

The discussion above has explained a significant number of observations in terms of non-uniformities in the acoustic and fluid flow fields simulated in previous chapters, and other phenomena such as acoustic streaming. As other processes and degrees-of-freedom are present in the experimental system, it is difficult to judge the accuracy of the predicted separation using the particle model and based on the 1-d simulation of the acoustic and fluid flow and associated particle forces. Regard- ing the acoustic radiation force, the literature discussed in section 2.3.1 is confident about how the acoustic radiation force acting on a particle relates to a known acoustic field, therefore the ability of the model to accurately predict the nature of the acoustic field becomes important.

The predicted and measured device impedance has already been discussed in section 6.2.1. This showed that the acoustic simulation of the device can predict the presence of resonant modes and the general trend of the impedance spectrum, although some error is present in the predicted magnitude of impedance. When simulating the performance of the device at a certain resonance, it is important to use thepredictedresonant frequency in the simulation in order to more accurately predict the acoustic energy in the fluid layer. However, this does incorporate an error in frequency.

To illustrate this point, figure 6.14 shows the measured and predicted impedance spectra of the printed device and there is an error of at least 100kHz between the two resonant frequencies. For the purposes of experiment the location of the voltage minimum is typically assumed to give a reasonable estimate of the resonant frequency, for which in this case a frequency of 4.378MHz was consistently used. Similarly for the simulation, the location of the predicted voltage mini- mum was used for simulated concentration results, highlighted in the figure at 4.515MHz and also corresponding to the impedance minimum.

The third graph in the figure describes the predicted acoustic energy density within the system, which is directly related to the magnitude of the acoustic radiation force (equation (2.8)). The peak in this energy is associated with resonance and it is assumed that a similar peak exists for the experimental device at a slightly lower frequency, although this currently cannot be measured. It is assumed that the experimental operating frequency and frequency used for simulation both correspond to similar positions on the respective energy peaks, therefore providing a reasonable estimate of the radiation forces. However, it is noted that the peak appears reasonably sharp and therefore a small change or error in frequency will significantly influence the resulting radiation force.

Figure 6.14: Characteristics of impedance, transducer voltage,Vt, and energy density and over region of 4.5MHz resonance and indicating frequency used for simulated results. Transducer voltage and energy density normalised relative to the peak values.

It can also be seen from the figure that the voltage (or impedance) minimum does not necessarily represent the energy maximum and the best frequency for separation. This frequency is located at a slightly higher frequency, typically between the minima and maxima of the voltage or im- pedance spectra. This suggests that the optimum operating frequency has not been used in the experimental work reported here. It also has implications on micro-processor control of the device and the ability to locate the optimum frequency based on a measurable characteristic of the device, although it is possible that a measure of electrical power could be used.