Manufacture of poly(methyl methacrylate) microspheres using membrane emulsification

Abstract

Accurate control of particle size at relatively narrow polydispersity remains a key challenge in the production of synthetic polymer particles at scale. A cross-flow membrane emulsification (XME) technique was used here in the preparation of poly(methyl methacrylate) microspheres at a 1–10 l h⁻¹ scale, to demonstrate its application for such a manufacturing challenge. XME technology has previously been shown to provide good control over emulsion droplet sizes with careful choice of the operating conditions. We demonstrate here that, for an appropriate formulation, equivalent control can be gained for a precursor emulsion in a batch suspension polymerization process. We report here the influence of key parameters on the emulsification process; we also demonstrate the close correlation in size between the precursor emulsion and the final polymer particles. Two types of polymer particle were produced in this work: a solid microsphere and an oil-filled matrix microcapsule.

This article is part of the themed issue ‘Soft interfacial materials: from fundamentals to formulation’.

1. Introduction

The range of applications for polymeric microspheres continues to grow across a wide scope of technology areas. In many cases, particles with controlled size characteristics are required [1]. For applications such as particle sizing standards this is self-evident, but it has also long been a requirement for chromatography applications where the particles act as the absorbent stationary phase and variations in surface area affect the retention time in the column.

Over the last few decades, polymer microspheres have found increasing use in biomedical applications [2]. The incorporation of pharmaceutical actives into inert carrier particles is frequently described as a route for novel drug delivery materials; and the link between drug availability and particle size is now well established, providing further impetus for managing particle sizes accurately. Passage into and through the body is also known to be dependent on particle size, for example inhaler applications have a very tight particle size requirement [3]. Finally, we note that there is an ever-growing interest in microparticles for use in medical imaging applications, where again good size control is a key requirement [4].

Suspension polymerization is used industrially for the production of a wide range of important commercial polymers such as polystyrene (PS), poly(methyl methacrylate) (PMMA) and poly(vinyl chloride) (PVC). It is also used to produce a range of polymer particles such as ion exchange resins and chromatography beads. For particle synthesis, the size distribution of the final particles is often very similar to that of the original emulsion, and so good control over the emulsion may be expected to provide good control over the final particulate product [5]. Industrial production of emulsions is usually achieved using rotor–stator systems, high-pressure homogenization or ultrasound. In general, these approaches use large quantities of energy and produce emulsions with a wide distribution of sizes [6]. During the emulsification, there will consequently be a large degree of drop break-up and drop coalescence during the process. Where the incorporation of another material into the particles is intended, for example an active drug, this complex process may cause difficulties in ensuring equal dosing in all particles. In a further limitation, these processes typically use very high shear conditions that may be detrimental to sensitive active molecules in a pharmaceutical formulation, for example.

The production of emulsions using drop-by-drop approaches is of growing interest for a wide range of possible applications as a result of the generally excellent control over droplet size and size distribution available [2]. Examples of such approaches using microfluidic devices as well as controlled pore size membranes have been reported. The advantages of such approaches are many: they are typically low energy; the shear fields are generally weak; exquisite size control is available; almost monodisperse product can be attained; and, through careful use of multiple systems in sequence, complex droplet-in-droplet architectures can be produced [6,7]. Despite this, commercial applications based on this technology remain very few: one disadvantage is the generally limited production rates, especially for microfluidic devices. For membrane emulsification (ME) systems, as an example, the oil flux for a 0.5 μm Shirasu porous glass (SPG) membrane has typical values ranging from 10 to 45 dm³ m⁻² h⁻¹ [8]. Higher fluxes have been reported using ceramic membranes with possibly up to one order of magnitude increase available. Manufacturing approaches that use premix routines and recycling of emulsions can also be used to improve operational efficiency and improve emulsion quality [9]. It is worth noting here that cross-flow membrane emulsification (XME) systems of the type used in this work can be easily parallelized to increase production volumes, and it is entirely feasible to match the output of an industrial homogenizer with a similar-sized unit [10]. Membrane or channel fouling is another potential issue for these drop-by-drop approaches that needs to be carefully managed if they are to be successfully scaled up for manufacturing applications.

Our work here focuses on the production of PMMA microspheres via a suspension polymerization process, using a stable pre-prepared emulsion of monomer and initiator dispersed in an aqueous continuous phase. For ME, the basic method is very simple in concept: a pressure is applied to one liquid, forcing it to permeate through a porous membrane (figure 1) into the continuous phase. As droplets are extruded from the membrane pores, they are detached by a combination of the hydrodynamic forces generated by the cross-flow of the continuous phase, the transmembrane pressure driving the flux of liquid through the membrane and the wetting properties of the droplet and continuous phase on the membrane. For membranes with tight pore size distributions, it is possible to produce essentially monodisperse emulsions [6]. Empirical relationships between pore size, liquid viscosity, transmembrane pressure and membrane type allow control over the droplet size [8]. A range of possible materials have been tested for use as membranes, including glass, ceramics, polycarbonates, microlaser-drilled steel and Teflon [11]. The most widely reported system is based on the SPG membranes. These membranes are appealing as the surface wettability can be easily modified and a wide range of pore sizes are available (typically from 0.05 to 30 μm) [12]. Regardless of the membrane type, it is known that the droplet sizes are usually 2–10× the pore diameter for stable emulsion production [13].

Figure 1.

Images of a typical ceramic membrane used here show the seven star-shaped internal channels through which the continuous phase flows.

Given that it has been previously established that particle sizes from suspension polymerization closely match those of the initial emulsion droplets, we have elected to explore XME as a route to produce the emulsion. XME uses a membrane material, in our case a sintered ceramic, with a narrow pore size distribution through which the dispersed phase can be expressed. In principle, a droplet can grow from each of the pores directly into the continuous phase. The stability of a droplet on the membrane is controlled by a balance of the various forces acting on it; and the detachment point is usually controlled by an imbalance between the Laplace pressure and the shear stress caused by the cross-flowing continuous phase [11,14]. Earlier work has highlighted the importance of factors such as cross-flow rate, driving pressure, surface tension and droplet stabilization by surfactant adsorption during growth and detachment as important for controlling the ultimate droplet size and polydispersity of the end product [15].

The first demonstrations of ME for the production of polymer microspheres were reported by Omi and co-workers in the late 1990s [13,16,17]. In this pioneering work, both PS and PMMA microparticles were synthesized from emulsions produced using a porous alumina-silicate glass membrane. Owing to difficulties producing a stable MMA emulsion, a more complex two-stage process with initial emulsification and then subsequent droplet swelling was employed. Despite this, stable dispersions of PMMA particles with diameters ranging from 3 to 35 μm were reported with a typical coefficient of variation (CV) of 10%. Later work from the same group describes a single-stage process for the emulsification of MMA prior to the subsequent polymerization reaction, although details of what had been modified to achieve this were not provided [18,19]. We assume here that the wettability of the membrane had been adjusted although this was not explicitly stated. Since these early studies, a number of other reports concerning other polymer materials and/or other polymerization or particle formation approaches have been published [2,9]. Despite the breadth of this earlier work, there are only limited studies on the preparation of polymer microspheres and these typically use emulsions at relatively low disperse-phase volume fractions (5–7%), with the polymerization being undertaken on batch sizes of 300 ml or less. Wolska & Bryjak [20] have reported a process to produce PS microspheres using ME where the emulsion was produced at 15% w/w styrene.

In our earlier work [21], we demonstrated the potential of ME for microcapsule manufacture. Here, we extend our earlier work to produce dispersions of solid PMMA particles at a relatively high particle concentration and in considerably larger volumes than has been previously demonstrated in the open literature. Using a sintered ceramic membrane, we aim to develop a simpler formulation and production process that will ultimately facilitate easier scale-up to industrial levels. Two types of particle were examined in this work: a solid polymer microsphere and an oil-filled matrix microcapsule type of particle.

2. Material and methods

(a). Chemicals

Methyl methacrylate monomer (Sigma Aldrich; 30 ppm) was passed through an alumina column to remove inhibitors and then stored in a refrigerator prior to use. Benzoyl peroxide (BPO) (Alfa Aesar; 25% moisture content), an oil-soluble initiator; lauryl alcohol (LA) (Sigma Aldrich; 98%); sodium dodecyl sulfate (SDS) (Sigma Aldrich; 98%); and Mowiol (PVA) (Sigma Aldrich; degree of polymerization approx. 1400, 86.7–87.7 mol% hydrolysis, molecular weight approximately 67 000) were all used as supplied. Hydroquinone (Alfa Aesar; 99%) was used as an inhibitor to control secondary nucleation and was also used as supplied. Castor oil (Fluka; 100%) was used as received. All water used here was of Millipore Milli-Q grade.

(b). Cross-flow membrane emulsification

Emulsions were prepared using an XME method. The membranes used in the work reported here were multi-channel ceramic tubes, as shown in figure 1. These alumina-coated membranes were specially commissioned (Fairey Industrial Ceramics Ltd, UK) and had average pore sizes of 0.2, 3 and 10 μm, respectively, on the detachment surfaces. The tubular membranes have a 600 mm length and an external diameter of 20 mm; in each case, they have seven star-shaped channels into which the disperse phase is expressed. Surface porosity has previously been reported as 10% [22] with a pore size span of 0.3–0.8 (see below).

The ceramic membrane of choice is housed in a stainless-steel module that allows the continuous phase to be pumped along the inner surfaces of the membrane, creating a shear field on those surfaces. The disperse phase is then pushed under pressure through the membrane from the external surface into the inner channels where oil droplets form at the membrane pores. The droplets will then grow and eventually detach under the action of the shear force caused by the cross-flowing continuous phase. The emulsion is usually re-circulated through the system using a pump until the desired emulsion concentration is achieved. The conditions used in the XME during production, such as the cross-flow rate and the pressure used to force the disperse phase through the membrane, can be tightly controlled.

All emulsions were produced at room temperature (20±1°C) and, in each experimental sample, the continuous phase was re-circulated through the system, whereas the disperse phase was expressed, resulting in a final emulsion with a droplet concentration of approximately 30% w/w. Emulsion samples were analysed using laser light scattering for droplet size and size distribution.

(c). Suspension polymerization

Suspension polymerization protocols were initially based on those of Omi et al. [16,17]. Prior to running a large-scale batch polymerization capable of accepting the full volume produced by the XME system, a series of small batch polymerizations were performed in a Carousel-6 reactor with a Tornado overhead stirrer (Radleys Ltd, UK) to optimize the formulation for use here. The formulation recipes tested are given in table 1. For each formulation, a batch of emulsion was pre-prepared using the XME system, with a 3 μm pore size membrane. After loading small batches (200 g) into the reactor, all samples were purged with nitrogen, and the temperature was raised to 65°C, which is sufficient to cause the breakdown of BPO and the formation of free radicals to initiate the polymerization. For scale up of the suspension polymerization, an approximately 2.4 kg batch of the preferred emulsion formulation was prepared using XME at 30% w/w MMA. Details of the formulation recipe are given in table 2. A 900 g sample batch of this emulsion was transferred to a jacketed reactor vessel equipped with an overhead stirrer and a condenser. Prior to initiation of the polymerization reaction, the emulsion was diluted with water and PVA (12.2 g in 1823 g) to give a final emulsion weight of 1.8 kg (i.e. approx. 16% w/w MMA). This was done primarily to manage the viscosity in the reactor and to prevent the formation of excess coagulum around the stirrer. Once again, the emulsion was purged with nitrogen and then heated to 65°C to initiate the reaction.

Table 1.

Formulation details for the different small batch suspension polymerizations undertaken in this work.

no.a	hydroquinone concentration (wt.% to total emulsion)	NaSO4 concentration (wt.% to total emulsion)	polymerization duration (h)	final polymer concentration (wt.%)b	dispersion characteristics
1	0	0	3	30	white, thick, magnetic stirrer eventually stopped, solid lumps of coagulum forming around it
2	0	0	20	35	white, very thick, magnetic stirrer eventually stopped, solid lumps of coagulum forming around it
3	0.02	0	8	3	yellow, very oily, separated
4	0.03	0	24	16	yellow, oily blobs, separated
5	0.03	0.02	24	27	beige, very thick, magnetic stirrer eventually stopped, solid lumps of coagulum forming around it
6	0.15	0	24	12	yellow, oily blobs, separated

^aEach formulation had the same base water/SDS/MMA/LA content at 1553 g/0.7 g/750 g/37.5 g.

^bPolymerization was initiated at 65°C using 1.5 g BPO.

Table 2.

Base polymerization formulation used in large-scale tests.

	MMA emulsion (XME) (g)
disperse phase
MMA	750.0
LA	37.5
BPO	15.0
continuous phase
water	1552.5
SDS	10.4
PVA	15.0
total mass	2380.4

In an alternative formulation, the MMA was partially replaced with castor oil to explore the potential of this approach for producing oil-filled microcapsules. Samples ranging from 6 to 24 wt.% castor oil, within the oil phase, were prepared.

For all polymerization reactions, final conversions were determined gravimetrically. The sample weight was recorded for the wet suspension. Subsequently, it was dried in an oven at 70°C for 8 h and reweighed.

(d). Emulsion and particle characterization

A Malvern Mastersizer 2000 (Malvern Instruments Ltd, UK) equipped with a Hydro S dispersion cell was used here for size analysis of particle and emulsion samples. Typically, a small sample was dispersed into a bath of water within the dispersion cell and then circulated through the measurement cell. Sample concentration was adjusted until an appropriate obscuration was achieved. The Mastersizer produces a particle size distribution from the low-angle scattering pattern using a conversion algorithm based on Mie theory. Each sample run consists of 10 measurements, and the reported size and size distribution data are an average of these multiple measurements.

Data are reported from the equipment as the D₁₀, D₅₀ and D₉₀, which correspond to the droplet diameters at 10%, 50% and 90% of the total volume, respectively. Size distribution is reported as the span, where

Zeta potential data as a function of pH, for the PMMA latex particles, were analysed using a Malvern ZetaSizer nanoinstrument.

Polymer particles were observed with a Carl Zeiss EVO MA15 scanning electron microscope (Carl Zeiss Ltd, UK). Care was taken to ensure that the beam intensity did not lead to particle melting.

3. Results and discussion

(a). Formulation optimization

In earlier work [17], hydroquinone was added to the aqueous phase to prevent the formation of particles from soluble monomer in the continuous phase. Other salts, such as sodium sulfate tested here, were also used to control the pH. In the formulation tests conducted here, the small batch polymerizations were always unsuccessful in the presence of hydroquinone. Interestingly, the simplest formulations tested (table 1: no. 1 and no. 2) gave the most reliable polymerization results and were therefore chosen as the base for further investigation with ME. During the scale-up, it was decided to increase the SDS concentration (table 2) to ensure that the formulation was well above the critical micelle concentration. We also added PVA to enhance the stability of the final particle sample.

(b). Emulsion formation

Table 3 summarizes the experimental conditions and the associated size information for the emulsions produced here by ME.

Table 3.

Details of the experimental test conditions and resultant emulsion droplet properties for the cross-flow emulsification tests. CO, castor oil.

		experimental conditions				droplet properties
sample run	formulation	membrane pore (μm)	membrane pressure (MPa)	disperse phase flux (l m−2 h−1)	continuous phase cross-flow (m s−1)	D50 (μm)a	span	Weber numberb (disperse)	capillary numberb (continuous)
A	MMA	0.2	0.05	11	2	0.9, 0.1		1.5×10−6	1.79
B	MMA	0.2	0.1	33	2	2.4	1.2
C	MMA	0.2	0.3	66	2	3.1	1.4
D	MMA	0.2	0.3	60	1.5	2.4	1.5		1.34
E	MMA	0.2	0.3	57	1	1.0, 5.1			0.89
F	MMA	0.2	0.3	56	0.5	1.1, 7.0			0.45
G	MMA/CO 6%	0.2	0.3	65	2	0.9	0.6
H	MMA/CO 9%	0.2	0.3	65	2	1.2	1.7
I	MMA/CO 12%	0.2	0.3	64	2	1.4	1.3
J	MMA/CO 24%	0.2	0.3	66	2	2.5	1.5
K	MMA	3	0.15	273	2	0.9, 5.8, 46
L	MMA	3	0.20	684	2	1.3, 91
M	MMA	3	0.23	3650	2	2.0	2.0	2.3×10−3	1.79
N	MMA	10	0.23	10945	2	3.1	2.7	6.8×10−2	1.79

^aTrue values of D₅₀ along with the span are given only for samples with a monomodal size distribution. Approximate values of the mean size for each peak are given for multi-modal samples.

^bAssumes an interfacial tension of 1 mN m⁻¹ based on SDS data from [29].

Looking first at the emulsions produced with the smallest pore size membrane (0.2 μm), we can examine the effects of various operational parameters. (X)ME is known to be sensitive to the interplay between a wide range of operational and materials parameters, including: the cross-flow shear field; the cross-membrane pressure; the wettability (of both phases) on the membrane; the viscosity of both phases; densities; and the interfacial tension [11,23]. Furthermore, the shape, spatial location and spacing of the pores can also have a significant influence [22]. Despite many empirical studies using a wide variety of each of these parameters, and various computational investigations [24,25], no fully predictive rules for this process are available. It is pertinent then to optimize the process using the preferred formulation for the polymerization reactions.

In a drop-by-drop emulsification process, the growth and detachment of a droplet from the mouth of a pore is controlled (primarily) by the balance between the interfacial tension force (resistance) and the drag force from the cross-flowing continuous phase (detachment). Any buoyancy forces are deemed to be negligible [25]. Should the inertial force owing to the disperse phase emerging from the pore become larger than the interfacial tension force then we would transition from ‘dripping’ to ‘jetting’ behaviour [26,27]. In practice, for most ME equipment, this transition point is never reached, and ‘dripping’ behaviour is dominant. A simple force balance model can be used to show that [28]

where R_d and R_p are the droplet and pore radii, respectively, and Ca is the capillary number that defines the ratio between the interfacial tension and drag forces, where

and γ is the interfacial tension, η is the viscosity and ν_c is the continuous-phase velocity (taken as the average cross-flow velocity). The transition from dripping to jetting is frequently described through a regime map that plots the continuous-phase capillary number against a disperse-phase Weber number [27], where

where v_d is the disperse-phase injection velocity, d is the fluid density and r is the pore radius. The injection velocity was calculated from the measured disperse-phase flux, where it was estimated that 5% of the total pore surface area was accessible. Given the relatively low pressures and small pore sizes, this active area was an average value from previous literature [29], although it was not measured directly and may not be constant across the different pore sizes used.

In figure 2, data for the droplet size distributions as a function of the cross-membrane pressure are given for the 0.2 μm pore membrane. A cross-flow velocity in the continuous phase of 2 m s⁻¹ is used in each case. We see here a general shift to a larger mean size as the membrane pressure is increased, consistent with previous investigations. This increase in size with an increase in membrane pressure has previously been rationalized on the basis of dynamic interfacial tensions. It is postulated that, at the higher pressures, the rapidly expanding interface of a growing drop is insufficiently coated with surfactant, and so the dynamic surface tension is higher [30]. This leads to increased stability of the droplet at the membrane surface and a corresponding increased growth prior to detachment. The measured span values at 0.3 and 0.1 MPa are 1.4 and 1.2, respectively. This small increase in the span at higher pressure may be owing to the increased number of active pores at the higher pressure: at higher cross-membrane pressures, more smaller pores will be active as their capillary pressures are exceeded [29].

Figure 2.

Droplet size distribution data for three emulsion samples (table 3: A–C) produced using a 0.2 μm pore membrane with a fixed velocity for the cross-flowing continuous phase of 2 m s⁻¹, and at different cross-membrane pressures, as shown.

Interestingly, at the lowest membrane pressure tested here, 0.05 MPa, we see a bimodal distribution in the size data. This is consistent with the formation of satellite droplets during droplet detachment from the membrane and suggests a more complex process than is often described for the droplet formation/detachment [30,31]. Various computational fluid dynamics and lattice Boltzmann (LB) simulations have shown how droplets growing at a membrane pore become distorted in the direction of the cross-flow [28,32]. In this case, the slower growth of the interface should result in a lower dynamic interfacial tension with an associated smaller mean drop size, as is seen. However, the reduced interfacial tension will also allow a greater distortion of the growing drop within the cross-flow shear field. This distortion may lead to an elongation of the droplet and, under certain conditions, the formation of a capillary neck that eventually breaks, releasing the main drop. The breaking of the neck and the associated snap back to the pore may also give rise to a satellite drop, a well-understood hydrodynamic phenomenon. The relatively narrow distribution of drop and satellite sizes suggests that, in this case, the formation process is still highly reproducible.

In figure 3, data for the droplet size distributions as a function of the continuous-phase cross-flow rate are given, with all other operational parameters being held constant and a membrane pressure of 0.3 MPa being used. We see here a reduction in the mean size as a function of increased cross-flow rate that is consistent with many earlier studies. Again, from a simple force balance model, we may expect that as the drag force increases the droplets will detach sooner and at a smaller mean size, as is seen. At the lower cross-flow rates tested here, we also see essentially bimodal size distributions. Once again, the possibility for the formation of satellite drops being produced during detachment needs to be considered. Here, the droplets are expected to grow to a larger mean size in the lower shear field; these larger droplets will be more deformable as a result of a lower Laplace pressure and may elongate more in the direction of the cross-flow. It is also noteworthy that the mean drop sizes (largest peak) at these low cross-flow rates are 20–30× the pore size, consistent with considerable distortion of the droplet [28]. Previous modelling has shown that the drag force is reduced as the droplet deforms, allowing for further growth of the droplet prior to detachment.

Figure 3.

Droplet size distribution data for four emulsion samples (table 3: C–F) produced using a 0.2 μm pore membrane with a fixed cross-membrane pressure of 0.3 MPa, and at four cross-flow velocities for the continuous phase, as shown.

The effects of increasing the viscosity of the disperse phase on the droplet size distributions are shown in figure 4. In this example, increased percentages of castor oil are added to the MMA in the disperse phase. The viscosities of the MMA and castor oil are 0.6 and 650 mPa s, respectively, at 20°C. Clearly, the addition of the castor oil will significantly increase the viscosity of the oil phase. In the data shown in figure 4, we see that an increase in the castor oil content leads to an increase in the droplet mean size; in all cases, the size distribution is monomodal with a relatively narrow CV. This increase in the mean size correlates with an increased stiffness of the droplets at the membrane surface. It has previously been shown that changing the ratio of disperse-phase to continuous-phase viscosities leads in some cases to an increase in droplet size, as seen here, whereas in other examples the opposite effect occurs [33]. An increase in drop stiffness might be expected to lead to a higher drag force and a resultant decrease in size; clearly, a complicating effect must be prevalent here although it is not clear what that might be. We note that the interfacial tension of MMA–water and castor oil–water are both around 15 mN m⁻¹, and so we do not expect a large change in interfacial tension with changing oil composition.

Figure 4.

Droplet size distribution data for four emulsion samples (table 3: G–J) with differing ratios of MMA to castor oil (as shown) produced at fixed cross-membrane pressure and continuous-phase cross-flow conditions.

Oil flux is obviously important for determination of the overall emulsion production rate, which is important for longer-term interest in this technology as a manufacturing route for large volumes of emulsions or particulates. Simple calculations suggest that, for oil flux rates of up to 50 l m⁻² h⁻¹ used here with a 0.2 μm membrane, we can expect to produce around 7.5 l of emulsion per hour. This should be compared with an industrial homogenizer that can produce around 1 m³ h⁻¹. It has previously been postulated that this production challenge could be solved by the design of an emulsifier with an array of ceramic tubes in parallel [33]. We note here that such a design has yet to be demonstrated.

Another route to increase the oil flux and improve the production rate is to use a membrane with a larger pore size. Data for the size distribution as a function of the cross-membrane pressure are shown in figure 5 for a membrane with a 3 μm pore size. In this case, low oil flux leads to an emulsion with a very polydisperse droplet size. An essentially monomodal distribution with a low span is seen at a membrane pressure of 0.23 MPa. Interestingly, the median size for this sample is D₅₀=2.0 μm, which is less than, but close to, the mean pore size. In contrast with the data for the small pore membrane (above), the broadening of the size distribution for this membrane is towards larger droplets as the pressure is decreased. The most plausible explanation for this would seem to be that it is the result of coalescence between droplets from neighbouring pores on the membrane.

Figure 5.

Droplet size distribution data for three emulsion samples (table 3: K–M) produced using a 3 μm pore membrane with a fixed velocity for the cross-flowing continuous phase of 2 m s⁻¹, and at different cross-membrane pressures, as shown.

As the pore size is increased, the cross-membrane pressure needed to activate a given pore is reduced and so we may expect more pores to be active at all times. Previous research has suggested that less than 10% of pores are active at any time for the 0.2 μm membranes used here [29]. However, for the 3 μm pore size membranes, at a similar cross-membrane pressure and for a similar formulation, we can expect that a larger fraction of the pores are active. Obviously, if more pores are active, then the mean distance between two growing drops on the membrane surface at any given time will be reduced and the possibility of neighbouring drops interacting will be increased. Also, as the pore size increases, then the expected drop size will also increase and larger drops have a lower Laplace pressure and are more deformable. We can expect considerable deformation in the direction of the cross-flowing continuous phase with the associated reduction in drag force leading to larger drop sizes, as discussed above. In addition, larger drops, with the associated deformation, may facilitate coalescence between neighbouring drops. All of this is consistent with the polydisperse droplet size distributions recorded for the lower pressures in this case.

These arguments do not, however, explain the relatively small droplet size and monomodal size distribution seen for the highest cross-membrane pressure tested here. When the number of active pores is very high but where coalescence is somehow prevented, possibly by surfactant adsorption, an additional detachment mechanism has been suggested by Holdich et al. [34]. This so-called push-off force relies on the deformation of growing drops at adjacent pores. This deformed ellipsoidal shape increases the free energy of the droplet; detachment leads to the drops reverting to a spherical shape and a reduction in the free energy. The steric hindrance of neighbouring drops also prevents deformation in the direction of flow, further reducing the growth of the droplets. The addition of a push-off force when added to the drag force from the cross-flow results in the detachment of droplets from the membrane surface at smaller than predicted sizes. While we hypothesize that this effect may be occurring here, it is worth noting that previous reports of this effect have been with straight-through drilled steel membranes where the pore activity is typically very high and can approach 100%. It has been suggested that this effect would be much less likely on a glass or ceramic membrane where pore activity is usually lower [35].

Another possible explanation is that we are operating in a jetting regime rather than the classical dripping one more common with these systems. It has been shown previously that controlled jetting can give rise to droplets with narrow polydispersity in size [36]. Furthermore, these droplets typically have a mean size that is smaller than the pore dimensions. Previously, it has been claimed that an oil flux of more than 100 l m⁻² h⁻¹ through these ceramic membranes leads to unstable conditions and a possible loss of control in the droplet polydispersity [6]. Looking at the disperse phase flux values used here (table 3), we see that, for the 3 μm membrane example, we are either close to or significantly exceed this value. Interestingly, under the conditions where we significantly exceed this flux value we see the highest quality of droplet formation, i.e. lowest polydispersity; a similar outcome, at equivalent flux rates, was previously reported by Yuan et al. [21] when using the same experimental set-up as we employed here. The conditions for a transition from dripping to jetting behaviour have previously been defined in terms of dimensionless hydrodynamic numbers; specifically, the Weber number for the disperse phase and the capillary number for the continuous phase [25,27,35]. Looking at these predictions, Meyer & Crocker [27] suggest that for Ca greater than 10⁻² we can enter a jetting regime even for We less than 10⁻² (and a clear jetting regime for We above this value). Comparing our conditions with these dimensionless numbers for the experiments run here (table 3), we see that all data for the 0.2 μm pore membrane are within the dripping regime, whereas the data for the 3 μm pore membrane are close to or beyond the boundary with the jetting regime. It seems entirely feasible here that we are operating under conditions where jetting is dominant and where most of the pores are active. In the case of jetting, droplets will form downstream of the pore through Rayleigh instabilities within the jet that cause its break-up. This results in very rapid emulsification but, importantly, while still retaining reasonable control over the droplet properties. Further work is required to better understand these observations. Data for a 10 μm pore membrane are also given in table 3 for a single emulsification run. Here, we expect to be very strongly within a jetting regime. The size distribution data (not shown here) are again monomodal with a relatively narrow distribution; in this case, however, the median drop size is approximately one-third of the pore dimension. It is interesting that we retain a relatively good control on the emulsion properties even under these conditions.

Looking at the oil flux data for the 3 μm membrane, we can calculate an emulsion production rate in excess of 3500 l m⁻² h⁻¹ or around 500 l h⁻¹ on a single membrane channel. This compares much more favourably with the output of a homogenizer than the examples for the 0.2 μm membrane above. As noted before, a set of membranes in parallel could, in principle, increase the production rate very simply [21]. Obviously, issues associated with consistent pressures across all the membranes and equalized cross-flow conditions would need careful design. To date, very little work on this upscaling has been done, and it would seem timely that more attention is now given to this subject although this is beyond the scope of this work.

(c). Particle synthesis

Size distribution data for particle samples and relevant precursor emulsions prepared with both the 0.2 and 3 μm membranes are given in figures 6 and 7, respectively. It is clear that the particle size distribution matches closely the precursor oil monomer droplet characteristics in both cases. This is consistent with a number of earlier studies concerning the use of ME technologies for the preparation of precursor droplets in a suspension polymerization process [16–20]. The absence of agglomeration between droplets/particles during the polymerization process suggests that the added PVA is as effective as a steric stabilizer. This control is very promising for the future and suggests that we can gain good control over the eventual particle sizes through control of the emulsion production. Interestingly, there does appear to be a small shoulder in the particle size distribution that is not seen in the emulsion precursor. It is not clear, at this time, what may have led to the appearance of this smaller size fraction.

Figure 6.

Size distribution data for PMMA/castor oil microcapsule particle samples and relevant precursor emulsions produced using a 0.2 μm pore membrane (table 3: I and J).

Figure 7.

Size distribution data for a solid PMMA particle sample and the corresponding precursor emulsion produced using a 3 μm pore membrane (table 3: M).

For both the solid polymer particles and the microcapsules, the conversion of monomer to polymer was calculated to be more than 85% w/w. Scanning electron microscopy (SEM) images of the solid polymer microspheres are given in figure 8. It is evident from these images that the surface of the microspheres for both the solid particles and the oil-filled capsules is essentially smooth. Interestingly, we see no evidence for the formation of small particulates in the continuous phase as a result of secondary nucleation reactions despite the removal of the hydroquinone from our formulation: in the earlier work of Omi et al. [16,17], small particles from such a secondary nucleation process were seen from SEM images to decorate the larger particles. Although not shown here, transmission electron microscopy analyses of the microcapsule samples confirmed the inclusion of the oil within the particles in a matrix-type structure.

Figure 8.

SEM micrographs of: (a) a PMMA/castor oil (CO) microcapsule sample at 12% w/w CO; and (b) a solid PMMA particle sample.

Zeta potential data (figure 9) for the solid PMMA particles provide additional information about their surface properties. As expected the particles are highly negatively charged and stable, primarily through the adsorption of the SDS, with the magnitude only reducing at very low pH near the isoelectric point. With the addition of increasing concentrations of salt, the zeta potential of the particles is increasingly reduced at each pH, destabilizing the dispersion almost completely at 10⁻² M (as the particle charge is almost completely suppressed here). While we expect some steric stabilization from the adsorption of the PVA, destabilization at the higher salt concentration suggests that this is inefficient.

Figure 9.

Zeta potential as a function of pH for the solid PMMA particle sample at three different background salt concentrations, as marked on the figure.

4. Conclusion

We have successfully demonstrated a two-stage approach for the manufacture of polymer microspheres with good control over the mean particle size and the associated size distribution. A thermally initiated suspension polymerization reaction followed the formation of a precursor emulsion of monomer (and oil-soluble initiator) in water produced using ME. A close correlation between the precursor emulsion droplets and resultant particle size distributions was found, suggesting that this approach could be used to gain very fine size control over polymer microsphere manufacturing. An analysis of the emulsification process highlighted how the complex interplay between materials and membrane properties as well as processing conditions must be understood on a system-by-system basis. It is also apparent from this work that much more still needs to be done to fully understand droplet formation and detachment mechanisms from ceramic membranes of the type used in this work.

Data accessibility

Supporting data for this article can be located at http://espace.library.uq.edu.au/view/UQ:376045.

Authors' contributions

J.B. is a PhD student and was primarily responsible for the collection of all data reported here and its initial analysis as well as aspects of the experimental design. J.B. also contributed by reading the manuscript and suggesting revisions during its preparation as well as giving final approval. M.S.M. contributed to the experimental design as well as data analysis. M.S.M. also contributed by reading the manuscript and suggesting revisions during its preparation as well as giving final approval. T.N.H. contributed to the analysis and interpretation of the data along with critical revisions to the draft manuscript. T.N.H. also gave approval to the final version of the paper. S.B. is group leader for the research and was a leading contributor to the conceptual design of the experimental programme. S.B. was also the primary author of the article responsible for various drafts and final approval.

Competing interests

We have no competing interests.

Funding

J.B. is primarily funded by an Engineering Physical Sciences Research Council (EPSRC)-Nuclear Decommissioning Authority Industrial CASE award managed by the National Nuclear Laboratory.