Assessment of predictive models for characterizing the atomization process in a spray dryer’s bi-fluid nozzle

Chemical Engineering Science 180 (2018) 42–51

Contents lists available at ScienceDirect

Chemical Engineering Science journal homepage: www.elsevier.com/locate/ces

Assessment of predictive models for characterizing the atomization process in a spray dryer’s bi-fluid nozzle Sadegh Poozesh a,b, Nico Setiawan a, Nelson K. Akafuah c, Kozo Saito c, Patrick J. Marsac a,⇑ a

University of Kentucky College of Pharmacy, Lexington, KY 40536, United States Mechanical Engineering Department, Tuskegee University, Tuskegee, AL 36088, United States c Mechanical Engineering Department, University of Kentucky, Lexington, KY, 40506, United States b

h i g h l i g h t s

g r a p h i c a l a b s t r a c t


A spray dryer’s nozzle with different

Droplet size distribution which is a pivotal factor affecting both size and morphology distribution of the final dried particle is investigated and characterized utilizing non-scaled parameters.

designs, able to atomize polymers dissolved in solvents under versatile conditions is investigated.
Scale analysis is utilized to relate nonscaled parameters to droplet size distribution.
Models are provided to describe both the droplet Sauter mean diameter and the span of the droplet size distribution.

a r t i c l e

i n f o

Article history: Received 30 April 2017 Received in revised form 24 November 2017 Accepted 22 January 2018

Keywords: Amorphous solid dispersion Spray drying Droplet size distribution Non-dimensional parameters Breakup phenomena

a b s t r a c t Spray drying as a commonly used process to produce amorphous solid dispersions of poorly water soluble active pharmaceutical ingredients (API) involves dissolution of the API and often a polymer, surfactant, and/or other functional excipient(s) into a volatile solvent. This feed solution is then pumped to an atomizing nozzle to produce droplets inside the drying chamber. The current paper aims to utilize non-scaled parameters to characterize the atomization process. A bi-fluid nozzle with two different designs commonly used in a lab-scale spray dryer was investigated under different operating conditions. The feed solutions were made of several excipients commonly used to produce amorphous solid dispersion composites. Atomization characterization is presented via both mean droplet size and size distribution. Various models are evaluated for predicting droplet mean diameter and span suitable for extrapolating atomization in the spray drying process. These approaches may be extended to other nozzles and across scales. Ó 2018 Published by Elsevier Ltd.

1. Introduction Poorly soluble APIs are often prepared as amorphous solid dispersion composites whereby the API (i.e. the drug) is combined with one or more excipients. Via this approach, the API may ⇑ Corresponding author. E-mail address: [email protected] (P.J. Marsac). https://doi.org/10.1016/j.ces.2018.01.033 0009-2509/Ó 2018 Published by Elsevier Ltd.

achieve improved thermodynamic activity in solution so as to increase the driving force for absorption and therefore the oral bioavailability. Pharmaceutical spray drying is commonly used to make amorphous solid dispersions (Beyerinck et al., 2010; Paudel et al., 2013) of the drug by combining with polymers and often other functional excipients such as surfactants. The steps in a pharmaceutical spray drying process include the following (Masters,

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51

1979): (i) Preparation of a feed solution using a solvent or solvent system that can readily dissolve the drug, polymer and other functional excipients, (ii) Atomization of the feed solution into the drying chamber where it is mixed with the heated drying gas, (iii) Evaporation of the fine spray droplets to produce solid particles, and (iv) Separation of the processing gas and the particles in a cyclone placed downstream of the main spray drying chamber. The formation of droplets via atomization is a critical step directly influencing the size and porosity of the resulting particles (Sander and Penovic´, 2014), the droplet size distribution may impact residence time and drying rates which may impact particulate physical structure and chemical homogeneity in space (Poozesh et al., 2017). Understanding the relationship between the processing conditions and resultant droplet size distribution and evaporation rate is therefore essential to identifying acceptable process ranges for various instrument scales. The purpose of this paper is to highlight those properties of the feed solution that impact the droplet size distribution for a given nozzle design and generate some guidance for understanding associated process sensitivity. Although the pharmaceutical literature has limited focus on the atomization process, many pharmaceutical unit operations rely heavily on the associated theory and application. Atomization is considered as one of the key steps in several pharmaceutical unit operations such as spray drying (Vicente et al., 2013), sprayfreeze drying (Wanning et al., 2015), tablet coating for elegance, modified, and controlled release formulations (Dennison et al., 2016), and fluidized bed granulation (Burggraeve et al., 2013) among others. For all applications, reproducibility must be achieved if consistent functional outcomes are expected. The atomization process must be well controlled, efficient, and reproducible across the controlled operational space. That is, performance measures must not be greatly influenced by expected fluctuations in the process conditions. As an example, it was shown that for a given batch of spray dried felodipine and poly(vinyl) pyrrolidone, large particles (presumably originating from large droplets) showed amorphous-amorphous phase separation while small droplets (presumably originating from small droplets) existing as a single homogeneous phase (Poozesh et al., 2017). Further, it was shown in the same publication that the morphology can be impacted by the distribution in droplet size. Clearly, a fundamental understanding of the spray drying process must start with an understanding of atomization phenomena. The process of droplet inception (atomization) to particle birth (solidification of the dissolved species) involves interacted fluid

43

dynamic, heat and mass transfer phenomena, mandates step by step research throughout the spray dryer. First and foremost, the atomization process must be understood. Atomization is the process by which a liquid jet disintegrates into unstable sheets, then ligaments and finally droplets. Focusing on the left panel of Fig. 1, sheet formation exists immediately adjacent to the nozzle tip and is difficult to capture – even via high speed imaging. Nevertheless, a continuous liquid is observed at the regions close to the nozzle tip; then by gradually going downward, the ligaments are more easily visualized. Finally, droplets are more easily identified as shown in this figure. The position and timescale of these breakup processes (i.e. sheet formation, ligament formation and droplet generation) are functions of nozzle design, operating condition, viscosity, density, and surface tension of the feed solution, among other variables. Obtaining meaningful droplet size distribution measurements must include consideration of the location of each of these transition points. Specifically, if measurements are focused on the region where sheet formation or ligaments dominate, the meaning of ‘‘droplet size” is unclear. Therefore, measurements should be made as a function of distance from the nozzle tip as will be shown below. Further, high speed imaging (the topic of a subsequent paper) may be coupled with droplet size measurements to ensure appropriate interpretation. According to Babinsky and Sojka (2002) there exist three available methods for modeling droplet size distribution: the maximum entropy method, the discrete probability function method, and the empirical method. Based on maximum entropy method the most likely droplet size distribution is the one that maximizes the entropy function under a set of physical constraints (e.g. conservation of spray mass, minimization of surface energy, etc.). These constraints are relied on at least two representative diameters of the droplet size distribution calculated via instability analysis by which only one diameter can be obtained. This shortcoming makes this method less favorable for practical applications. On the other hand, in discrete probability function method, a droplet size distribution is obtained from applying deterministic linear or non-linear breakup models on non-deterministic initial conditions that depend on variety of factors (such as turbulence, surface roughness, vortex shedding, mixture composition, etc.). This method can be used only for the initial stage (i.e. primary breakup) in atomization process; so its application is limited particularly for our bi-fluid nozzle whereby the secondary breakup controls the droplet size. Finally, there is classical method of modeling droplet size distributions: a curve is fit to data collected for a wide range of

Fig. 1. Left: breakup mechanism, right: configuration of the bi-fluid nozzle.

44

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51

atomizer nozzles and operating conditions. The problem with this method is the difficulty of extrapolating the data to other nozzle design or processing conditions. To overcome this weakness, realizing breakup controlling forces and translating them into nondimensional parameters pave the road to make cross-scale analysis to correlate droplet size distribution with underlying breakup forces. The current study focuses on the bi-fluid nozzle design whereby the atomization process is dictated by the balance between several forces. Specifically, the mass flow rate between the liquid and the gas (alternatively, often captured in terms of the relative velocity between the gas and the liquid) which promotes the atomization process and the capillary and viscous forces which act as a resistance to the atomization process. Ideally, the sensitivity of the droplet size on process conditions can be deduced from the relative magnitude of these factors/forces. For example, increasing the difference between the gas and liquid velocity at the nozzle outlet, either by lowering the liquid flow rate or raising the air pressure, results in smaller droplets. One approach to this end is to develop correlations between these forces and the associated measured droplet size. This may be accomplished using non-dimensional analysis capturing both the physical properties of the feed solution and operation conditions to droplet size distribution at an appropriate distance downstream of the nozzle tip. This approach captures the critical operational conditions and formulation physicochemical properties into a dimensional parameters that describe the underlying physics behind the breakup phenomena (Hassan et al., 2012). Significant efforts have been made to correlate droplet size (defined by the measures such as the Sauter Mean Diameter (SMD)) with operational and formulation conditions for various applications using bi-fluid nozzles. For example, Dennison et al. (2016) implemented a design of experiments approach to assess the impact of gas atomization pressure, liquid flow rate and polymer concentration on droplet size produced via a bi-fluid external mixing nozzle for a film coat application. By controlling these variables, the authors were able to control the mean droplet size and consequently manipulate characteristics of the resulting film coat to achieve a thinner, more uniform and less porous coating. In yet another study, Vicente et al. (2013) assessed and quantified the impact of solution viscosity, gas and solution flow rates on droplet size and consequently particle size and morphology . In a more comprehensive study, Petit et al. (2015) presented a non-scaled expression to describe the mean droplet size in terms of the physical characteristics of the inlet solution and the operating conditions of a bi-fluid atomizer on various solutions and skimmed milk concentrates. The characteristic length of the non-scaling droplet size, which is often a matter of debate in the scientific community, was chosen to be the vertical distance between the nozzle tip and the measurement location. Although various semi-empirical correlations are proposed in the literature, they are most often customized to a specific bi-fluid nozzle design, feed solution, and operating condition. Herein, these correlations cannot be applied to systems of pharmaceutical interest. Further, existing literature most often focuses on mean droplet diameter – largely ignoring the impact of droplet size distribution (Hede et al., 2008). Although modern bi-fluid nozzles render narrow droplet size distributions, the width or the span of the distribution may dramatically influence the attributes of the final particle including homogeneity, yield, bulk powder moisture content, etc. The current study takes into consideration the sensitivity of the droplet size distribution on the properties of the feed solution and the operating conditions. The aim of the current study is to correlate process conditions and formulation physiochemical parameters to the droplet size distribution for two externally mixed bi-fluid nozzles. Various feed solutions were evaluated. Since methanol and acetone are com-

monly used solvents in pharmaceutical spray dry operations, these were studied in detail. Further, two commonly used polymers, Poly (vinylpyrrolidone-co-vinylacetate) (PVPVA) 64 and hydroxypropylmethylcellulose acetate succinate H-grade (HPMCAS-H) were evaluated – each at different concentrations in each solvent. In the studies conducted by Konno et al. (2008) and Hugo et al. (2013) to identify the impact of polymer selection on dissolution rate of amorphous solid dispersions in presence of API, these two polymers exhibited superior bio-availability in terms of both dissolution rate, and storage stability. Besides, as will be discussed in the next section, in the present study attempt has been made to choose these polymers showing completely different viscosity behaviors to cover a wider spectrum of viscosity. Droplet size distribution in terms of Sauter mean diameter as well as the span of the distribution is described through original non-dimensional parameters for two sizes of the liquid orifice, dL, (Fig. 1, right). Multivariable non-linear regression was used to obtain the best fit out of the data. 2. Materials and methods 2.1. Materials PVPVA and HPMCAS were selected as model polymers as they are often used for the preparation of amorphous solid dispersion systems with poorly soluble APIs. Methanol and acetone serve as model solvents for the same reason. PVPVA was obtained from International Speciality Products (Wayne, NJ, USA) and HPMCAS was purchased from Shin-Etsu Chemical Co., Ltd (Tokyo, Japan). Methanol and acetone were obtained from Sigma-Aldrich, St. Louis, MO. PVPVA and HPMCAS were added to Methanol and Acetone, and mixed using a magnetic stir bar until completely uniform. The following solutions were prepared: (1) solutions of 5%, 10%, and 20% (w/w) PVPVA in methanol, (2) solutions of 5%, 10%, and 20% (w/w) PVPVA in acetone, (3) solution of 5% (w/w) HPMCAS in methanol, and (4) solution of 5% (w/w) HPMCAS in acetone. 2.2. Methods Viscosity, surface tension, and density are fundamental properties of the solvent/solution that may be used to characterize the atomization process (Rizkalla and Lefebvre, 1975). Viscosity measurements of the solutions were performed on a Brookfield DV-III + Rheometer (Massachusetts, USA) using spindle CPA-40z at different rotation speeds. All of the examined formulations except the solutions with HPMCAS showed a Newtonian behavior (linearity between shear rate versus shear stress). The solutions with HPMCAS showed a pseudoelastic behavior; and the reported viscosities were obtained at 30 (rpm). Important to note that since the liquid experiences a high shear stress at the nozzle head, shear rates and consequently reported viscosities are prone to change during atomization. Surface tension was measured using Surface Tensiometer (Fisher Scientific, PA, USA). All the measurements are done at room temperature and the results are shown in Table 1. The experimental setup is shown schematically in Fig. 2 and consists of a RTS 5114 Malvern Spraytec (Malvern Instruments, Worcestershire, UK), an adjustable holder for the nozzle, a ventilation unit for collection of the spray, a feed solution and associated pumping system, and a bi-fluid atomizing nozzle. Full cone sprays were produced with an external mixing bi-fluid nozzle with liquid orifice diameters, dL, of 0.7, 0.5 mm. These are standard nozzles used in a pilot-scale B-290 spray dryer (Buchi, Flawil, Switzerland) having a cap diameter of 1.5 mm (see Fig. 1(b)). A PHD 4400 Hpsi Programmable syringe pump (Harvard Apparatus, MA, USA) with a 50 ml syringe capacity is used to control the liquid flow rate. The

45

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51 Table 1 Physical properties of the atomized solutions. Property

DI Pure Pure 5-PVPVA5-PVPVA10-PVPVA10-PVPVA- 20-PVPVA20-PVPVA- 5-HPMCAS- 5-HPMCASwater (a) methanol (b) acetone (c) Methnaol (d) Acetone (e) Methanol (f) Acetone (g) Methanol (h) Acetone (i) Methanol (j) Acetone (k)

Density (kg/m3) 999 Surface tension 72.0 (mN/m) Viscosity 1.05 (mPas)

792 22.7

785 25.5

793 22.8

785 25.5

796 22.8

795 25.7

805 22.8

806 25.7

793 22.8

786 26.0

0.6

0.32

1.66

0.71

3.3

1.4

10.6

4.27

9.35

5.57

Fig. 2. Schematic of the experimental setup.

refractive index of N2 as the atomizing gas was chosen to be of 1.00 + 0.00i in all calculations. The formulations’ refractive indexes were assumed to be the same as the pure solvents. In particular, standard values of: 1.33 + 0.00i, 1.327 + 0.00i and 1.3586 + 0.00i were used for water, methanol, and acetone, respectively. The Spraytec results are reported in terms of SMD and the mean diameter (D50). The span is used for characterizing the droplet size distribution’s width. It is worth mentioning that during atomizing the solutions comprising acetone, volatile components caused a scattering response to be observed at small angles. This effect which is known as ‘‘Beam steering” shows peaks of apparent large material. The causes lie in the fact that evaporation of volatile components (here acetone) and temperature gradients caused by different drive gases result in inner detector scattering. This can be corrected for with the Spraytec software and this should give correct data. To obtain realistic data, the inner detectors are eliminated, and, as such the low-angle data channels affected by beam steering are removed before calculating the droplet size distribution of the spray. Fig. 3 shows an example of undesired detectors elimination process and its influence on droplet size distribution. As illustrated in the panel (a) of this figure, initial full detectors configuration renders two picks and consequent high reported droplet size. After detector treatment, panel (b), the distribution peak is reduced to one with more realistic mean droplet size.

3. Results and discussion 3.1. Sauter mean diameter The Sauter mean diameter (SMD or D[3, 2]) is the diameter of a sphere with the same volume to surface area ratio of the entire

spray and may be considered as a measure of the fineness of the spray (Filkova and Mujumdar, 1995). In this paper, we use this diameter to describe the atomization process since it captures the total area for heat and mass transfer – important for understanding the particle formation process. Nevertheless, it is first necessary to determine the appropriate geometry and position of the measurement device relative to the spray nozzle. The axial distance between the nozzle tip and the laser beam should include consideration of both the droplet formation process, droplet coalescence, and the evaporation process. In particular, if the laser is too close to the nozzle tip, the results will inappropriately reflect dimensions characteristic of the ligaments prior to droplet formation. Alternatively, if the laser is too far from the nozzle tip, it is conceivable that some evaporation may have occurred prior to the measurement. Of equal importance, the appropriate distance between the transmitter and receiver of the Spraytec should be selected so as to maximize intensity while avoiding droplet collision with measurement components. The influence of the downstream distance of the nozzle tip on SMD is shown in Fig. 4 for gas flow rate of 7.2 E4 kg/s. It was observed that for liquid flow rates of 30, 50 and 80 cc/min, 6 cm gives the best results. In particular, as shown in Fig. 4, a similar descending and then ascending trend is observed for all flow rates. Lasheras et al. generated the same general trend when evaluating the near- and far-field breakup and atomization of a water jet by a high-speed annular air jet (Lasheras et al., 1998). Recalling Fig. 1(a), the breakup is not completed at the regions close to the nozzle tip, and that is why at 2 cm, the values of SMD are artificially high. At 6 cm from the tip, dramatic changes in SMD subside suggesting that the main breakup (atomization) process is complete. Also, it can be perceived from this figure that for 30 ccm flow rate, since the atomization is completed in closer downstream distance (about

46

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51

Fig. 3. Beam steering issue for acetone. Droplet size distribution before and after correction (a), detectors’ configuration after elimination process (b).

45 40

SMD(µm)

35 30 25 20

30 ccm

15

50 ccm

10

80 ccm

5

0

2

4

6

8

10

12

14

Distance from the nozzle tip (cm) Fig. 4. Impact of the downstream distance between the nozzle tip and laser beam _ g ¼ 0:72 mkg=s. on SMD for pure water and m

3–4 cm), the turning point for this cure is different from the others. Further downstream of the spray cone, the dynamic pressure of the gas can no longer provide sufficient inertial forces to overcome the surface tension forces. From this location downward, droplets may coalescence and show some increase in the overall measured size. Meanwhile, evaporation may occur and eventually dominate the droplet size measurement as the axial distance from the nozzle continues to increase. Moreover, when making these types of measurements in a spray drying chamber, the impact of the processing gas must also be considered – perhaps these effects become more dramatic with increased difference between the velocity of the processing gas and the velocity of the droplets. Finally, the distance between the transmitter and the receiver was chosen to be 60 cm. This standard distance is selected to both ensure the highest laser intensity and avoid droplet collision with the lasers. The results of changing the size of the liquid orifice are shown in Fig. 5 with the solid lines corresponding to measurements using the 0.7 mm liquid orifice and the dashed lines corresponding to measurements using the 0.5 mm liquid orifice. As noted in Fig. 1, the inside diameter of the outer orifice remains constant throughout all of the experiments but the size of the liquid needle is altered – both 0.5 mm and 0.7 mm liquid needles are studied. Several observations are noted in Fig. 5. For both nozzles, the

droplet size decreases with increasing pressure drop. Of course, increasing the gas flow rate creates a larger pressure drop at the orifice and therefore smaller droplets are produced. Moreover, it is observed that for low liquid flow rates, the smaller liquid orifice produces smaller droplet sizes over the entire range of atomizing gas pressure studied. Once the liquid flow rates are increased to 80 ccm, both nozzles show a similar droplet size as a function of gas pressure. Further increases in the liquid flow rate leads to larger droplet sizes for the smaller orifice. This behavior can be interpreted by considering two competing effects. First, at a given pressure drop, the gas flow rate will be higher for the small nozzle since it has a larger annulus; and, therefore, the gas to liquid ratio will be higher for the small liquid nozzle orifice relative to the gas to liquid ratio for the larger liquid nozzle orifice1. Second, at any given liquid flow rate, the smaller liquid orifice will have a higher liquid flux (i.e., the liquid must be faster for the 0.5 mm orifice as compared to the 0.7 mm since the flow is constant but the crosssectional area is reduced). At lower liquid flow rates, the former factor dominates and the smaller nozzle produces smaller droplets; however, for higher liquid flow rates, the effects of reduction in relative velocity as a result of higher liquid velocity for the smaller nozzle overwhelms the former and counters the trend. Each nozzle will produce droplets that decrease in size as the gas to liquid ratio increases. However, the rate of change in droplet size is unique for each nozzle. Clearly, the 0.7 mm nozzle shows a stronger droplet size dependence on liquid flow rate as compared to the 0.5 mm nozzle and at large liquid flow rates the 0.5 mm liquid orifice produces larger droplets. Finally, it is noted that the differences in droplet size between the two nozzles diminishes as the atomizing gas pressure increases suggesting that the atomizing gas dominates the droplet size at high gas velocity. To have a rigorous estimation of the resultant droplet size range, the SMD in Fig. 5 ranges between 10 and 90 lm. If this range is considered for 20% PVPVA in methanol from Table 1, based on the continuity equation, produced droplets will have an SMD in the range of 1.1–2.3 lm. After establishing experimental parameters and demonstrating results consistent with expectation, additional systems of pharmaceutical relevance were studied next. Commonly used solvents in 1 According to Darcy–Weisbach equation, and assuming same head loss and friction factor for both the nozzles, then, volumetric flow rate(gas annular area)1.25.

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51

90

SMD (µm)

80

30 ccm

70

50 ccm

60

80 ccm 110 ccm

50 40 30 20 10 50

70

90

110

130

150

170

190

210

Gas pressure (kPa) Fig. 5. Impact of liquid orifice size on SMD for two sizes: 0.7 (solid lines) and 0.5 mm (dashed lines) in different pure water flow rates.

the spray drying unit operation include methanol and acetone while commonly used polymers include HPMCAS and PVPVA. As shown in Table 1, the density, viscosity, and surface tension of several systems were measured with the intention of developing relationships between the physicochemical properties of feed solutions and the measured droplet size. With an understanding of the functional dependence of droplet size on feed solution prop-

47

erties, product sensitivities to scaling, inherent process fluctuations, and process/equipment changes may be explored in more detail. Further, critical process parameters and formulation variables can be interrogated based on fundamental physicochemical properties as opposed to strictly relying on empirical design of experiments. As shown in Table 1, the physical properties of representative feed solutions show a strong dependence on solvent type, polymer type, and concentration of dissolved species. These differences should translate to differences in droplet size and ultimately particle size, morphology, and performance. In particular, the SMD measurements of solvent and solution systems reported in Table 1 are shown in Fig. 6 for liquid flow rates of 30 and 80 ccm and gas flow rates up to 1.25 g/s. Worth noting, the viscosity is markedly sensitive to the amount of polymer dissolved in the solvent system – perhaps an anticipated result. In contrast, the surface tension and density show a less pronounced dependence on the amount of dissolved polymer. Careful review of the entire dataset uncovers some general conclusions. First, for the representative systems studied here, the atomizing gas flow rate plays the most influential role in determining the SMD. Second, an asymptotic trend of the SMD is observed with increasing air volume flow rate – especially for lower liquid flow rates. Third, larger droplets are produced with increasing concentration of dissolved polymer which, of course, follows the trend of increased viscosity. Fourth, solutions prepared in methanol yield larger droplet size as compared to identical solutions prepared in acetone. Worthy of note that in the present

Fig. 6. Impact of formulation selection on SMD as a function of feed solution composition and atomizing gas flow rate.

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51

manuscript, the investigation has been done in the absence of API. The addition of API, depending on the API to polymer ratio, may change the physiochemical properties of prepared solution. As noted previously, the solution surface tension, and with lower extent its density, are independent of polymer type. Analogously, the changes in these non-dimensional parameters are the factors dictating the droplet size distribution, regardless of the type of API present. A closer look at Fig. 6 panel an and panel c reveals that representative feed solutions with similar physical properties give similar droplet size across all gas flow rates studied. For instance, a solution of 20% PVPVA yields a droplet size very similar to a solution of 5% HPMCAS. However, in panel b, the solution with 20% PVPVA has slightly higher SMD at lower gas flow rates despite its lower viscosity in comparison to the ones with 5% HPMCAS. This may be the result of the fact that PVPVA solutions display a higher density. As the gas flow rate is increased, the impact of density appears to be less relevant. Next, panel d has a higher liquid to gas ratio as compared to panel b and the viscous forces play a large role in determining the resulting SMD. Also shown in Fig. 5, acetone based solutions have a more compact SMD profile and this can be explained by the narrower range of physicochemical properties with the most viscous solution in this category measured at l = 5.6 mPas. General trends between processing variables, physicochemical properties of the feed solution, and the resulting droplet size for a given nozzle were explored as outlined above. Although most observed trends are consistent with expectation, complex phenomena were identified and explained in terms of nozzle design, feed solution properties, and processing conditions. For instance, the 0.5 mm and 0.7 mm nozzle produced dramatically different droplet sizes as a function of flow rate at several atomization pressures. Moreover, the 0.7 mm nozzle produced larger droplets at lower liquid flow rates while the 0.5 mm nozzle produced larger droplets at the higher liquid flow rates. Nevertheless, it would be desirable to develop quantitative relationships relating the droplet size to physical properties of the feed solution and characteristics properties of the nozzle as a function of processing conditions. This may be accomplished via describing and capturing the various forces in the atomization process in terms of reduced or dimensionless parameters. This approach is meant to compare the relative magnitude of various forces in terms of fundamental properties. The Weber number (Weg) compares the dynamic gas pressure acting on the liquid sheet to the liquid capillary pressure of the liquid sheet (Hede et al., 2008) as shown in Eq. (1).

Weg ¼

V 2g qg dL

r

ð1Þ

Here, vg, qg, dL and r are gas velocity, gas density, liquid orifice diameter, and liquid surface tension, respectively. When Weg is increased, the atomization mechanism becomes more intense, and the resulting droplets are smaller (Chigier and Farago, 1992). When a gas stream is directed onto a liquid surface, oscillations and waves on the surface of the liquid are produced and these promote the fragmentation process of bulk liquid. This phenomenon is captured in the numerator of Eq. (1). The resisting force to these oscillations is liquid viscous force – captured in the denominator of Eq. (1). Another dimensionless number that relates gas inertia to the liquid viscous force is the Reynold’s number (Re) (Poozesh et al., 2016; Saito and Williams, 2015). However, while working with both We and Re number, to isolate inertia effects, another pi-number, the Ohnesorge number (Oh), can be derived via combination of We and Re, Oh = We0.5/Re. This dimensionless number is purely dependent on formulation properties and geometry of liquid orifice. The Oh number is defined as shown in Eq. (2).

l

Oh ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi ql rdL

ð2Þ

The liquid density is represented by the variable ql and other variables were defined previously. The ratio of the magnitude of the involved forces associated with the atomization process may be determined with consideration of both the Weg and Oh numbers. In addition to the involved forces, the information on phases mass flow rates are needed to analyze the dynamics of the _ g =m _ l , is atomization system. The gas-to-liquid ratio, GLR ¼ m defined as the ratio between air and liquid mass flow rates and has often been cited as a dimensionless quantity which captures the atomization operating conditions (Lefebvre, 1980). Considering these parameters, Groom et al. (2005) obtained the following semiempirical relationship to predict SMD:

" SMD ¼ dL C 1 

#C2

Weg

ð1 þ GLRÞ2

ð1 þ C 3  OhÞ

ð3Þ

In the Groom study, commercial nozzle were used and consisted of an external liquid capillary outlet with a co-flow gas stream at comparatively high relative velocity. The experiments were performed with water and aqueous glycerol solutions with a viscosity ranging from 1 to 100 mPas and an atomizing air pressure ranging from 50 to 300 kPa. The liquid orifice diameter was 1 mm. Due to the nozzle design and operation conditions proximity to our experimentations, Eq. (3) serves as the benchmark for modeling our SMD dataset. Fitting the obtained droplet size distribution data obtained using the Spraytec to the general form of Eq. (3), the following relationship was obtained. That is, the measured SMD for all the tested formulations under different operational conditions was correlated to the dimensionless parameters Weg, GLR, and Oh as shown in Eq. (4).

SMD ¼ 0:106  We0:188  GLR0:45  ð1 þ 1:5  OhÞ g dL

ð4Þ

Fig. 7 compares the predicted and measured droplet size using Eq. (4). The p-Value for all of the variables are significantly low and suggest that the identified variables capture the fundamental process variables and feed solution properties as well. The response contour plots shown in Fig. 8 provide a visual representation of the changes in droplet size across the operational ranges studied here. This representation of the data allows for optimization of the process conditions. For instance, the plots indicate that for very fine droplet formation the major critical factors are GLR and Weg (i.e., remarkable change in droplet size are shown as a function of these dimensionless numbers).

71

SMD (μm)-predicted

48

61 51 41 31 21 11 1

1

11

21

31

41

51

61

71

SMD (μm)-measured Fig. 7. Calculated SMD versus experimental measured SMD. The dash is the trendline with R-Squared = 0.965.

49

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51

Fig. 8. Response contour plot of SMD/dL based on Eq. (4).

3.2. Distribution of spray mist diameter

span ¼

The droplet size distribution of the spray may be described in terms of the dimensionless span or width defined as span = (D90  D10)/D50. Note that Da (i.e. a = 90, 10, 50 here) is a representative diameter where a% of the total volume of the liquid sprayed is made up of droplets with diameters smaller than the stated value and the rest is made up of droplets with diameters bigger than the stated value (Chaker et al., 2002). In Fig. 9 the span is shown as a _ g for different formulation compositions and processfunction of m ing conditions and no obvious trend is observed. Another approach to describe the span based on formulation and operation parameters is the widely used expressions of droplet size distribution based on Rosin-Rammler distribution as described by Lefebvre (Lefebvre, 1989) and shown in Eq. (5).

8 D10 ¼ 0:1521=q > > < D50 D90 ¼ 3:3221=q D50 > > : D50 ¼ b1=q Cð1  1qÞ SMD

ð5Þ

Here, q is Rosin–Rammler parameter, C is the gamma function, and b is a constant with value of 0.693. From the first two expressions of Eq. (5), one can find the span as a function of q as shown in Eq. (6).

D90  D10 ¼ 3:3221=q  0:1521=q D50

The SMD and D50 may be measured via the Spraytec and the q values may be calculated from Eq. (6) which allows for calculation of the last expression in Eq. (5). After adjusting b in the expression, Fig. 10 illustrates the ratio of measured D50/SMD versus the modified last expression from Eq. (5). The value of 0.726 for b shows good agreement between the modified expression and the experimental results. These results clearly show the possibility of modeling span based on the q values drawn from the modified expression of Eq. (6). The problem is then reduced to how one can predict D50/ SMD to back calculate q then and finally get the span values. Previously a model was presented and successfully tested to predict SMD in this paper; as such, the solution is to look for another correlation for D50 based on the operating conditions, as well as formulation physical properties. Fortunately, such a correlation was presented by Kim and Marshall (1971). They found the following correlation for a convergent external mixing type of nozzle operated on melts of wax mixtures over a range of liquid viscosity from 0.001 to 0.050 Pas, relative velocities from 75 to 393 m/s, GLRs of 0.06–40, liquid density of 800–960 kg/m3 and gas densities of 0.93–2.4 kg/m3.

1.50

1.40

Span

1.30

1.20

Pure water Pure methanol

1.10

Pure acetone 1.00

5%HPMCAS-methanol 5% HPMCAS-acetone

0.90 0.0004

0.0005

0.0006

0.0007

0.0008

ð6Þ

0.0009

0.001

0.0011

0.0012

Gas flow rate (kg/s) _ g on the span measured at 40 ccm liquid flow rate. Fig. 9. The effect of formulation selection and m

0.0013

50

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51

1.8

The sensitivity of both SMD and span of the mist on design, operational conditions and formulation (in the absence of API) were evaluated and analyzed. For SMD, a predictive model based on the literature was presented and customized to the understudied variables. Regarding the span of the distribution, which has been neglected in previous studies, a new approach was introduced and successfully tested against measured span data. These predictive models can be leveraged to estimate both of the critical spray attributes, SMD and span. Then this data may serve as the input for further computational or theoretical fluid dynamic simulations which is encompassed as the future work of this study.

D50/SMD-predicted

1.7 1.6 1.5 1.4 1.3 1.2 1.1 1

1

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

D50/SMD-measured Fig. 10. Modified expression based on q values extracted from measured span data for predicting D50/SMD. The dash is the trendline and R-Squared of the fitting is 0.855.

81

D50(μm)-predicted

71

51 41 31 21 11 1

11

21

31

41

51

61

71

81

D50(μm)-measured Fig. 11. Calculated D50 based on Eq. (7) versus experimental measured data. The dash is the trendline with R-Squared = 0.932.



D50 ¼ C 1

We are grateful to the National Institute for Pharmaceutical Technology and Education (NIPTE) and the U.S. Food and Drug Administration (FDA) for providing funds for this research. This study was funded by the FDA Grant to NIPTE titled ‘‘CRITICAL PROCESS PARAMETERS FOR SPRAY DRYING PHARMACEUTICALS‘‘; with grant number of U01FD004275. References

61

1

Acknowledgment

lc2 rC3 l þ C6 rql mCr 4 qCl 5

C 7

GLRC 8

ð7Þ

Here, Cis are constants to be determined, and vr is the relative gas and liquid velocities. Worth noting, Eq. (7) is the result of modification and adaptation of the original empirical equation according to the variables studied in this paper (e.g. gas density in the original equation was variable, and in this study is treated as a constant). Employing multivariable non-linear regression C1-8 are found as to be: 21,335, 0.18, 0.22, 1.0, 0.16, 25.659, 0.014, and 0.7, respectively, using MATLAB programming software. Fig. 11 shows the degree of agreement between the governing empirical equation and the measured data for all of the examined variables in the design, formulation, and operational conditions. In this figure, for droplet sizes more than 50 lm, deviation from the trendline is higher compared to smaller ones. Typically for the understudied nozzle, bigger droplets are resulted from low atomization gas flow or/and high liquid flow rates (generally low GLR). In such cases, the atomization is poor and the nozzle operation conditions are not at the optimum level. Besides, as can be seen, the size data in that region is scattered showing less conducted measurements (approximately 3% of data), which also can add more uncertainty in the presented model. 4. Conclusions This study investigated the characterization of the spray issuing from a bi-fluid nozzle adopted in a laboratory-scale spray dryer.

Babinsky, E., Sojka, P., 2002. Modeling drop size distributions. Prog. Energy Combust. Sci. 28, 303–329. Beyerinck, R.A., Ray, R.J., Dobry, D.E., Settell, D.M., 2010. Method for making homogeneous spray-dried solid amorphous drug dispersions using pressure nozzles. Google Patents. Burggraeve, A., Monteyne, T., Vervaet, C., Remon, J.P., De Beer, T., 2013. Process analytical tools for monitoring, understanding, and control of pharmaceutical fluidized bed granulation: a review. Eur. J. Pharm. Biopharm. 83, 2–15. Chaker, M., Meher-Homji, C.B., Mee, T., 2002. Inlet fogging of gas turbine engines: Part B—Fog droplet sizing analysis, nozzle types, measurement and testing, ASME turbo expo 2002: power for land, sea, and air. Am. Soc. Mech. Eng., 429– 441 Chigier, N., Farago, Z., 1992. Morphological classification of disintegration of round liquid jets in a coaxial air stream. Atomization and Sprays 2. Dennison, T.J., Smith, J., Hofmann, M.P., Bland, C.E., Badhan, R.K., Al-Khattawi, A., Mohammed, A.R., 2016. Design of experiments to study the impact of process parameters on droplet size and development of non-invasive imaging techniques in tablet coating. PloS one 11, e0157267. Filkova, I., Mujumdar, A.S., 1995. Industrial spray drying systems. Handbook of industrial drying 1, 263–308. Groom, S., Schaldach, G., Ulmer, M., Walzel, P., Berndt, H., 2005. Adaptation of a new pneumatic nebulizer for sample introduction in ICP spectrometry. J. Anal. At. Spectrom. 20, 169–175. Hassan, R., Loubiere, K., Legrand, J., Delaplace, G., 2012. A consistent dimensional analysis of gas–liquid mass transfer in an aerated stirred tank containing purely viscous fluids with shear-thinning properties. Chem. Eng. J. 184, 42–56. Hede, P.D., Bach, P., Jensen, A.D., 2008. Two-fluid spray atomisation and pneumatic nozzles for fluid bed coating/agglomeration purposes: a review. Chem. Eng. Sci. 63, 3821–3842. Hugo, M., Kunath, K., Dressman, J., 2013. Selection of excipient, solvent and packaging to optimize the performance of spray-dried formulations: case example fenofibrate. Drug Develop. Ind. Pharm. 39, 402–412. Kim, K., Marshall, W., 1971. Drop-size distributions from pneumatic atomizers. AIChE J. 17, 575–584. Konno, H., Handa, T., Alonzo, D.E., Taylor, L.S., 2008. Effect of polymer type on the dissolution profile of amorphous solid dispersions containing felodipine. Eur. J. Pharm. Biopharm. 70, 493–499. Lasheras, J., Villermaux, E., Hopfinger, E., 1998. Break-up and atomization of a round water jet by a high-speed annular air jet. J. Fluid Mech. 357, 351–379. Lefebvre, A.H., 1980. Airblast atomization. Prog. Energy Combust. Sci. 6, 233–261. Lefebvre, A.H., 1989. Properties of sprays. Part. Part. Syst. Char. 6, 176–186. Masters, K., 1979. Spray drying handbook. Spray drying handbook. Paudel, A., Worku, Z.A., Meeus, J., Guns, S., Van den Mooter, G., 2013. Manufacturing of solid dispersions of poorly water soluble drugs by spray drying: formulation and process considerations. Int. J. Pharmaceut. 453, 253–284. Petit, J., Méjean, S., Accart, P., Galet, L., Schuck, P., Le Floch-Fouéré, C., Delaplace, G., Jeantet, R., 2015. A dimensional analysis approach for modelling the size of droplets formed by bi-fluid atomisation. J. Food Eng. 149, 237–247. Poozesh, S., Akafuah, N., Saito, K., 2016. New criteria for filament breakup in droplet-on-demand inkjet printing using volume of fluid (VOF) method. Korean J. Chem. Eng. 33, 775–781. Poozesh, S., Setiawan, N., Arce, F., Sundararajan, P., Della Rocca, J., Rumondor, A., Wei, D., Wenslow, R., Xi, H., Zhang, S., 2017. Understanding the processproduct-performance interplay of spray dried drug-polymer systems through complete structural and chemical characterization of single spray dried particles. Powder Technol. 320, 685–695.

S. Poozesh et al. / Chemical Engineering Science 180 (2018) 42–51 Rizkalla, A., Lefebvre, A., 1975. The influence of air and liquid properties on airblast atomization. J. Fluids Eng. 97, 316–320. Saito, K., Williams, F., 2015. Scale modeling in the age of high-speed computation. In: Volume, I.I. (Ed.), Progress in Scale Modeling. Springer, pp. 1–18. Sander, A., Penovic´, T., 2014. Droplet size distribution obtained by atomization with two-fluid nozzles in a spray dryer. Chem. Eng. Technol. 37, 2073–2084.

51

Vicente, J., Pinto, J., Menezes, J., Gaspar, F., 2013. Fundamental analysis of particle formation in spray drying. Powder Technol. 247, 1–7. Wanning, S., Süverkrüp, R., Lamprecht, A., 2015. Pharmaceutical spray freeze drying. Int. J. Pharmaceut. 488, 136–153.