High-Magnification Shadowgraphy Characterization of Rotary Atomizer Droplet Production for Pharmaceutical Applications

Transcription

1 IASS Americas, 19 th Annual Conference on iquid Atomization and Spray Systems, Toronto, Canada, May 06 High-Magnification Shadowgraphy Characterization of Rotary Atomizer Droplet Production for Pharmaceutical Applications C. M. Varga *, and H. E. Snyder Nektar Therapeutics San Carlos, CA 9 USA Abstract A rotary wheel atomizer used for powder production in spray drying processes has been characterized using highspeed, high-magnification shadowgraphy. Droplet sizes were measured over a range of wheel rotation speeds and liquid flow rates for both water and ethanol utilizing a commercially-available image-processing package. Dropletsize histograms were generated at each condition, and derived diameters were compared against predictions from classical rotary atomizer droplet size correlations from the literature. A multivariate regression analysis was performed using a genetic algorithm solver to produce a new predictive correlation demonstrating excellent agreement with the measured performance. The results of the present investigation indicate that over the range of conditions explored, the power-law dependencies of droplet-size on both rotary wheel speed and surface tension are stronger than those predicted by classical correlations, while the dependence on viscosity and liquid mass flow rate are weaker. * Corresponding author

2 Introduction Rotary wheel atomizers are utilized in a wide range of industrial applications from painting to the spray drying of food and pharmaceutical products [1]. Spray drying processes have gained popularity in the pharmaceutical production of both excipients, such as lactose, as well as active pharmaceutical ingredients (API). This process offers the ability to combine formulation science with particle engineering to create powder of the desired chemistry as well as physical properties for innovative drug delivery systems. The advantages of this scaleable, economical, one-step process from liquid solution to final particle were implemented in the recently approved first inhaled insulin product [2]. For drying applications utilizing rotary atomization, the liquid is delivered to a high speed rotating wheel where is it ejected from the perimeter and subjected to shearing forces between the wheel tip and the surrounding gas. The resulting droplets are exposed to a gas environment with suitably low solvent partial vapor pressures to promote evaporative mass transfer and convert the liquid droplet into a solid particle. The method of atomization impacts the particle production by breaking the bulk liquid into individual droplets, which directly controls product issues such as the final particle size and powder flow characteristics. In addition, how the droplets are introduced into the gas stream can impact the effective drying rate and, hence, particle properties such as density, surface area and surface composition. Rotary designs have the inherent advantage of a wide spray angle, which acts to reduce the probability of multiple droplet interactions prior to particle formation. When operated in a direct droplet production regime, rotary atomizers yield narrower particle size distributions in comparison to hydraulic or twin-fluid types [1]. One drawback to using a rotary atomizer for fine particle spray drying applications is limited ability to produce small droplets (< micron diameter). This directly reduces production rates for the production of particles below 10 microns as the relatively large droplet size forces the processing of highly dilute solution feedstocks. In addition, process scale-up based upon a rotary design is complicated by both liquid length scale changes and relative velocity shifts as rotary wheel diameters increase and rotational rates are lowered due to mechanical stress concerns. Factors impacting rotary atomizer performance include the wheel design, liquid rheology and operating conditions (wheel speed and liquid feedrate). The liquid droplets will form directly at the wheel, or through sheet/ligament structures [3]. For large commercial applications, the motive force for wheel rotation is typically provided by electric motors; however, lab-scale units can be driven by pneumatic turbines integrated into the atomizer housing. The majority of rotary atomizer droplet size correlations reported in the literature [1, 3] were developed before the advent of currently-available noninvasive, in situ optical measurement assays. Improved predictive capability would enable more efficient scale-up of spray drying processes in which precise control of the final particle size across production scales is required. This paper reports on a study performed with a commercial rotary wheel atomizer (Niro model M02/b) commonly utilized on the Niro Mobile Minor lab spray dryer. Non-invasive, in situ droplet size measurements were conducted using a avision laser-based imaging system over a range of liquid flows and wheel rotation rates. The application of optical imaging to liquid droplet sizing is a direct measurement technique that is dependent upon numerous assumptions for appropriate data analysis. This paper briefly reviews the resulting sensitivity to the analysis software settings. Results will be further processed using a multivariate software optimization package to fit the data to previously published functional forms with updated coefficient and exponent values. Experimental Setup Figure 1 illustrates the commercially available Niro Model M02/b rotary atomizer from GEA Niro Inc. tested in this study. The atomizer consists of a rotary wheel powered by an air turbine, driven by a pressurized gas source. The rotary wheel contains equispaced vanes around the perimeter. The liquid is slung out of these openings and the atomization process is driven by the high relative velocity between the liquid and the surrounding air provided by the high speed rotation of the wheel. Figure 2 shows a closeup view of the Niro rotary atomizer wheel along with characteristic dimensions, d (wheel diameter), and h (vane height). The Niro rotary atomizer tested in the present study contains 24 vanes (n=24) and has a wheel diameter of d = 5 cm (d = 5cm), and a vane height of 5 mm (h = 5mm).

3 iquid Inlet Air Inlet diffuser and directed vertically upwards from below the spray with a turning mirror. Air was provided to the atomizer through the house CDA supply and water or ethanol flows were supplied using a gear pump. Air pressure was measured just upstream of the air inlet on the rotary atomizer via pressure transducer. CCD Camera Rotary Wheel Figure 1. Niro Model M02/b Rotary Atomizer. Rotary Atomizer Pulsed Nd:YAG asers QM-100 ong- Distance Microscope Measurement Volume Illuminating Beam Turning Mirror Figure 3. Schematic of Experimental Setup iquid exit ports Figure 2. Rotary wheel parameters. Droplet-size testing of the rotary atomizer was carried out in the Nektar Atomization Test Facility (ATF) spray chamber. This ambient temperature, controlled flow environment is designed to accommodate multiple spray diagnostics. Figure 3 contains a schematic of the optical setup for the present experiments. The rotary atomizer was fixed vertically on the translational stage mount of the ATF threedimensional traverse, thus generating a circular spray with its axis aligned to the horizontal direction in a lab-fixed coordinate system. A QM-100 longdistance microscope with CCD camera capture was placed above the spray with the focal plane aligned to the horizontal spray axis plane at a radial distance of 1 inch from the wheel. Image backlighting was provided by a pulsed Nd:YAG laser passed through a d h Droplet-size measurements were performed in the present study using high-magnification shadowgraphy with pulsed-laser illumination. Acquired images were processed with the droplet-sizing software, SizingMaster, from avision Inc. This software employs a sizing algorithm involving a two-step segmentation of each acquired shadowgraph image. In the first segmentation, a user-defined global intensity threshold serves to locate the droplets. The second segmentation of the sizing algorithm then focuses on each of these droplets and employs a high and low intensity level relative to the maximum intensity of the droplet to calculate both a high level and low level droplet diameter. The user may reject droplets by limiting the allowable difference between these high and low values, termed the droplet halo. This parameter can have a marked effect on the measured droplet distribution, and care should be taken in setting this value appropriately for the size range of droplets being measured, as well as for the depth of field of the imaging setup. Figure 4 contains a diagram that illustrates the determination of the halo width of a measured droplet. The software ultimately reports an average of the high and low diameter values, with these diameters calculated assuming circular areas. An absolute image scaling factor is applied to convert pixel values to microns. This absolute scaling factor is determined using a software routine that requires focusing on a calibration plate or object of known size. Figure 5 contains a SizingMaster screenshot from the present experiments, which illustrates the location of two distinct droplets as well as the software-determined sizes and major/minor axes. Additional details of operation and principles behind

4 this measurement system can be found in the avision SizingMaster Manual [4]. rates are reported in revolutions per second (RPS), ranging from approximately 0 RPS (100 RPM) at 68.1 kpa to approximately 6 RPS (39000 RPM) at kpa. A regression fit for the rotation rate based upon a single parameter power law formula yielded: RPS=31.05*(kPa) Figure 4. Illustration of Halo Width Determination Atomizer Wheel rotation (RPS) RPS = 31.05(kPa) R 2 = Rotary atomizer gas pressure (kpa) Figure 6. Rotary Wheel Rotation Rate as a function of Gas Pressure Figure 5. SizingMaster Droplet Sizing Image Sets of 1000 images each were acquired with the SizingMaster software at atomizer pressures of,,, 80, and 90 psi and liquid volume flow rates of,,, and 80 ml/min. Water was tested across the entire range of atomizer pressures and liquid flows, while ethanol was tested only at the ml/min condition over the full range of pressures. Sauter Mean (SMD) were calculated from the droplet-size distributions generated by the SizingMaster software at each condition tested. The 1000-image data sets translated to ~10,000 samples at the high wheel rotation rates and to hundreds of samples at the lowest wheel rotation rates. Results and Discussion Figure 6 contains rotation rate data for the Niro rotary atomizer obtained via laser tachometer. Rotation rates were measured in 68.9 kpa atomizer pressure increments from 68.9 to kpa. Rotation Figure 7 contains Sauter Mean Diameter (SMD) data as a function of rotation rate for the range of flows from ml/min to 80 ml/min for water and ml/min for ethanol. It can be observed in this figure that increasing wheel rotation rate leads to smaller mean droplet sizes at all liquid flow rates, as expected. However the impact of liquid flow rate on droplet sizes diminishes with increasing rotation rates, such that all the water curves in this figure approach a performance plateau (i.e. equivalent SMD) at the highest rotation rate of 619 RPS (371 RPM). It can also be observed in this figure that the ethanol results at ml/min exhibit SMD values which are 25 - % smaller than the comparable water values at all rotation rates. This result is expected due to the significantly lower surface tension of ethanol. Sauter Mean Diameter (microns) Atomizer Wheel Rotation Rate (rps) ml/min Water ml/min Water ml/min Water 80 ml/min Water ml/min Ethanol Figure 7. Sauter Mean Droplet Diameter as a Function of Atomizer Wheel Rotation Rate

5 Figure 8 contains SMD values for water as a function of liquid flow rate at fixed rotation rates ranging from 312 RPS (187 RPM) to 619 RPS (371 RPM). It can be observed in this figure that at low wheel rotation rates (<0 RPS or 000 RPM), increasing liquid flow rate translates to increased droplet sizes, while at higher rotation rates (>0 RPS), the SMD response to increasing liquid flow rate is very flat. Sauter Mean Diameter (microns) psi (312 rps) Water psi (429 rps) Water psi (0 rps) Water 80 psi (595 rps) Water 90 psi (619 rps) Water iquid Flow Rate (ml/min) Figure 8. Droplet Sauter Mean Diameter as a function of iquid Flow Rate The sensitivity of the results presented in Figures 7 and 8 to the software filters provided in the avision SizingMaster software was examined in the present study. In particular, the impact of both maximum halo width settings and depth of field corrections (DOF) were analyzed. The depth of field for the optics utilized in the present study was approximately 80 microns. This narrow depth of field associated with highmagnification imaging can present some challenges for ensuring detection of the largest target droplets, which can be roughly the same width. In particular, this has implications on setting of the maximum halo width, which can altogether eliminate the counting of large droplets if it is set too low (below a threshold of a few pixels). On the other hand, setting this parameter to a higher pixel count or disabling the halo filter completely will include more out of focus droplets and increase the expected errors in derived diameters. Figure 9 contains normalized number frequency distributions for a single condition in the present experiments using three different maximum halo settings (2- pixel, 5-pixel, and 10-pixel). Normalized Number Frequency Droplet Size (microns) 10-pixel max. halo 10-pixel log-normal fit 5-pixel max. halo 5-pixel log-normal fit 2-pixel max. halo 2-pixel log-normal fit Figure 9. Effect of Maximum Halo Width Filter on Measured Size Distribution It can be observed clearly in this figure that reducing the maximum halo width leads to a reduction in counting of the larger droplets in the field. In the extreme, we see that the 2-pixel maximum halo setting will eliminate all droplets in the distribution above microns, and will under size the spray in general. The differences in the 5-pixel and 10-pixel maximum halo distributions are also significant; the 5-pixel maximum halo removes all droplets above microns from counting. The impact of this filter on the large end of the size distribution has a strong impact on derived diameters such as the Sauter Mean Diameter. Figure 10 contains plots of SMD as a function of wheel rotation rate for the ml/min condition for each of the three halo settings in Figure 9. Sauter Mean Diameter (microns) Atomizer Wheel Rotation Rate (rps) 10-pixel halo 5-pixel halo 2-pixel halo Figure 10. Effect of Maximum Halo Width Filter on SMD as a function of Atomizer Wheel Rotation Rate, ml/min It can be observed in Figure 11 that reducing the maximum halo from 10 pixels to 5 pixels reduces the calculated SMD values by ~5 microns at most wheel speeds, while maintaining the trend with rotation rate. The 2- pixel maximum halo setting eliminates a much larger

6 portion of the size distribution, removing the impact of rotation rate all together. These data clearly illustrate the importance of appropriately setting this filter. In practice, one can target a reasonable setting for this filter parameter by examining well-focused individual droplet images of various sizes to determine the minimum halo width necessary for detection. Figure 11 illustrates such a focused droplet of approximately microns in size. For the configuration tested, this droplet was not detectable for maximum halo widths of less than 7 pixels. A maximum halo width of 10 pixels was employed in the present study to ensure detection of droplets up to 90 microns in size. Figure 11. SizingMaster Image of a -micron Droplet The avision SizingMaster software also includes a Depth of Field (DOF) correction option that was not enforced on the data presented in this paper. A sensitivity analysis of this DOF correction on select data in this study was performed, however, and is presented in Table 1. The DOF function adjusts the size distribution to account for the probability bias in the detection of large droplets due to the depth-of-field size dependence [4]. The DOF correction shifts the SMD values down by ~4 microns for the m/min water data shown in Table 1. Table 1. Impact of DOF Correction on ml/min SMD Results Rotation Rate (rps) Uncorrected D 32 (m) DOF Corrected D 32 (m) The observed difference in droplet-size behavior between the low and high wheel rotation rates in the present study suggests a fundamental difference in the droplet production mechanism at these two conditions. Three unique droplet production mechanisms have been previously identified for rotary atomizer disks, including direct drop formation, atomization by ligament formation, and atomization by film formation [3]. The critical liquid flow rate, q, which defines the boundary between the direct drop formation and ligament formation regimes has been shown to follow the empirical formula q d N d (1) 0.5 For the minimum wheel rotation rate encountered in the present study (312 rps or 187 RPM), equation 1 predicts direct droplet formation for liquid flow rates less than ml/min, while at the highest wheel rotation rate (619 RPS or 371 RPM) direct droplet formation should occur at liquid flows less than ml/min. Similar calculations from Equation 1 at the other wheel rotation rates encountered in the present experiments suggests that for rotation rates greater than 0 rps, only the lowest liquid flow value of ml/min would fall within the direct drop formation regime. This suggests that ligament formation was the dominant mechanism of droplet formation over the bulk of the high-rotation-rate conditions in the present experiments, and supports the observed shift in performance with increasing rotation rate. A comparison of the measured droplet-size distributions underlying the derived diameters plotted in Figures 7 and 8 was also performed in the present study. Figure 12 contains a comparison of the droplet volume frequency distributions for two different wheel rotation rates with water at a flowrate of ml/min. It can be observed in this figure that the distribution corresponding to the lower rotation rate of 429 rps is both shifted to the right and broadened slightly in comparison to the high-rotation-rate distribution. A bimodality is also noted in the low rotation rate distribution, with peaks occurring at approximately and microns.

7 Figure 13 contains a similar comparison of volume frequency distributions for a fixed rotation rate of 619 RPS (371 RPM) with water at flowrates of ml/min and 80 ml/min. In spite of the 4X difference in liquid mass loading to the wheel between these two conditions, the high rotation rate produces very similar volume distributions for both cases, with the exception of a number of droplets greater than 55 microns in size in the 80 ml/min case, which do not appear at ml/min. Figure 14 contains a comparison of the volume frequency distributions for water and ethanol at conditions of ml/min and 619 RPS. These two distributions are distinctly different, with ethanol producing a volume distribution that is both narrower than the water distribution and shifted significantly toward smaller diameters. This significant difference in droplet-size performance is driven primarily by the factor of three reduction in surface tension for ethanol compared with water. Normalized Volume Frequency Normalized Volume Frequency ml/min, 429 rps 529 Drops, SMD=42.7 m ml/min, 619 rps 49 Drops, SMD=31.4 m Droplet Size (microns) Figure 12. Effect of rotation rate, Volume Frequency Distributions - ml/min, Water, 429 vs. 619 RPS Rotation Rates Droplet Size (microns) 80 ml/min, 619 rps 7947 Drops, SMD=32.1 m ml/min, 619 rps 1274 Drops, SMD=31.9 m Figure 13. Effect of flow, Volume Frequency Distributions 619 RPS, ml/min vs. 80 ml/min, Water Normalized Volume Frequency Droplet Size (microns) ml/min, 619 rps, Water 49 Drops, SMD=31.4 m ml/min, 619 rps, EtOH 33 Drops, SMD=23.6 m Figure 14. Effect of liquid type, Volume Frequency Distributions, ml/min, 619 RPS, Water vs. Ethanol In order to provide practical predictive capability for droplet-size performance the data collected in the present study was compared against historical literature correlations for rotary atomizer wheels [1][3]. These correlations were developed primarily in the 19 s to support the broad use of rotary atomizers in the chemical processing industry and can differ substantially in predicted results for a given operating condition and liquid rheology. It should be noted that the droplet-sizing techniques utilized at the time of these studies (e.g., sieving, wax particle collection and counting, oil-in-water droplet imaging, etc.) were inherently limited in comparison to sophisticated sizing techniques available today. The non-intrusive optical approach used in the present study represents a significant advance in modern characterization of this commonly-used atomizer type. Predictions from three frequently quoted correlations for rotary atomizer wheels were selected and compared to the current droplet size data. Quality of fit was measured by 1) the coefficient of determination, R 2, of the measured and predicted SMD data, and 2) the sum of the SMD squared differences. A summary of these values for each correlation is provided in Table 2. Table 2. Summary of Correlation Comparisons with SMD Data Equation Herring and Marshall SMD east Squares Sum (m 2 ) Coefficient of Determination, R 2 Dimensionally Correct No Friedman et al No Fraser et al Yes Fraser-Evolver Optimized 1 Fraser-Evolver Optimized No Yes

8 Figure 15 contains a comparison of the present SMD data set in its entirety (i.e. all rotation rates, flow rates, and liquid rheology variation) to the correlation provided by Herring and Marshall [1], which takes the form, x10 Km SMD (2) Nd 0.83 nh where K = 8.5 x 105 to 9.5 x 105 is an empirically determined factor that depends upon both wheel speed and liquid mass loading [3]. Based upon the operating range for the current data set, a value of K = 8.5 x 105 was selected. SMD predictions from this correlation compared with measured values from the present study exhibited an R 2 value of 0.77 and a large residual (least squares sum). Additional weaknesses of this correlation include a lack of explicit terms containing liquid surface tension and viscosity, and incorrect dimensions. Measured SMD, m (SizingMaster) Predicted vs Measured x column vs y column Predicted SMD, m (Herring and Marshall Equation) Figure 15. SMD Data Correlation with Herring and Marshall Equation (1) Figure 16 contains a comparison of the present SMD data set to the correlation provided by Friedman et al. [1] which takes the form, m 0.44 nh SMD d 2 2 Nd m m (3) This correlation is an improvement over the correlation of Herring and Marshall in that it includes terms which represent liquid surface tension and viscosity; however, its representation of the current rotary droplet-size data is also inadequate, exhibiting an R 2 value of just 0.66 and the largest residual of the three correlations evaluated. Measured SMD, m (SizingMaster) Predicted vs Measured inear Regression, R 2 =0.66 Predicted SMD, m (Friedman et al. Equation) Figure 16. SMD Data Correlation with Friedman et al., Equation (2) Figure 17 contains a comparison of the present SMD data set to the correlation of Fraser et al. [1] which takes the form, SMD 0.483N m d nh (4) This correlation yielded the most satisfactory representation of the current droplet-size data set and, similar to the Friedman et al. correlation, contains terms that explicitly account for surface tension and viscosity effects. Furthermore, this equation is dimensionally correct, in contrast to the previous two correlations. Measured SMD, m (SizingMaster) Predicted vs Measured inear Regression, R 2 = 0.66 Predicted SMD, m (Fraser Equation) Figure 17. SMD Data Correlation with Fraser Equation (3) Based upon the satisfactory results obtained with the Fraser correlation as presented in the literature compared to the other two correlations, an optimized correlation with the same functional form was sought for the present droplet-size data using a tailored regression analysis. This analysis was performed using

9 a genetic solver package for Excel called Evolver, by Palisade. The Evolver software was presented with an equation of the following form, SMD C c a b N m * d (5) d nh where C is a constant between 0 and 1, and a, b, c and d are exponents, which could vary between a user-defined range from -1 to 1. The software minimized the sum of SMD difference squared through variation of these parameters, leading to the following optimized correlation, SMD 0.69N m (6) A comparison of the present data with the optimized correlation of Equation 6 is shown in Figure 18. The coefficient of determination R 2 between the predicted and measured data sets for this comparison was 0.95 and the least squares sum was reduced to ~105 m2 in comparison to the sum from the original Fraser equation at ~916 m2 (see Table 2). Measured SMD, m (SizingMaster) Predicted vs Measured inear Regression, R 2 =0.95 d nh Predicted SMD, m (Evolver-Fraser Optimized 1) Figure 18. SMD Data Correlation with Evolver Optimized Fraser Equation 1 The optimized correlation presented in Equation 6 was not, however, constrained to be dimensionally correct. Applying an additional constraint that the equation be dimensionally correct, the following correlation for SMD is proposed, SMD 0.59N m d nh (7) This dimensionally correct correlation maintains an R 2 value of 0.95 and slightly increases the least squares sum to 116 m2. A comparison of the current SMD data to predictions from this equation is presented in Figure 19. Table 3 provides a list of the individual parameters used in this correlation along with required units and example values for reference. Measured SMD, m (SizingMaster) Predicted vs Measured inear Regression, R 2 = 0.95 Predicted SMD, m (Evolver-Fraser Optimized 2) Figure 19. SMD Data Correlation with Evolver Optimized Fraser Equation 2 (Dimensionally Correct) Table 3. Equation 7 Correlation Parameter Units and Examples Parameter/Variable Symbol Units Example Sauter mean diameter SMD m 32.9 x 10-6 wheel rotation rate N rps 595 liquid density kg/m (water) liquid dynamic viscosity Pa-s (water) liquid mass flow rate m kg/s 9.98 x 10-4 ml/min) rotary wheel diameter d m 0.05 surface tension N/m (water) # of vanes on atomizer wheel n n/a 24 vane height h m Conclusions Droplet-size performance characterization was conducted on a rotary atomizer using a highmagnification laser-illuminated imaging system. Trends of mean droplet size with rotary wheel rotation rate, liquid feed rate, and surface tension were observed to be consistent with traditional observations for rotary atomizers. SMD data were compared with three available literature correlations for rotary wheel atomizers and quality of fit was evaluated for each. A dimensionally correct correlation from Fraser et al. [1] was found to provide the most satisfactory representation of the current data set and was subsequently chosen for optimization by a genetic solver. An optimized correlation with the same functional form as the Fraser correlation was produced and was shown to provide an excellent representation of the present droplet-size data.

10 A sensitivity analysis of several software settings which impact measured droplet-size distributions in shadowgraph sizing was performed. The maximum halo width filter was shown to have a significant impact on detection of large droplets when using highmagnification optics such as the QM-100 used in the present study, where the depth of focus may approach the maximum expected droplet sizes. Nomenclature AR CDA d h kpa m air-to-liquid mass ratio Clean dry air Wheel Diameter (m) Vane height (m) kilopascals Mass flow (kg/s) N Rotation Rate of Atomizer (RPS or RPM) n Number of vanes per wheel. RPS Revolutions per second RPM Revolutions per minute SMD Sauter Mean Diameter (m) [3] iquid Density (kg/m^3) iquid surface tension (N-m) Viscosity (Pa-s) Subscripts liquid References 1. Masters, K., Spray Drying Handbook, Wiley and Sons, FDA Approves First Ever Inhaled Insulin Combination Product for Treatment of Diabetes, FDA press release, P06-13, Jan 27, efebvre, A.H., Atomization and Sprays, Hemisphere Publishing Corporation, avision SizingMaster Manual.