Investigation on nozzle zone agglomeration during spray drying using response surface methodology

During spray drying, dry powder is circulated into the nozzle zone to force collisions, inducing agglomeration. This study systematically determined the effect of fine powder mass flowrate (varying from 7.1 ± 1.2 – 15.9 ± 0.5 kg∙h -1 ), drying air temperature (160 – 200 °C), and drying air mass flowrate (472.8 ± 6.2 – 590.8 ± 9.9 kg∙h -1 ) on agglomerate size and morphology using a central-composite trial design. Agglomeration was quantified using an agglomeration index based on laser diffraction and by quantifying particle morphology using static image analysis. Response surface models were used to quantify factor effects. Increasing the fines mass flowrate had the largest positive effect on particle size enlargement and development of grape-like agglomerates. Increasing drying air temperature had a small negative effect on particle size enlargement and no significant effect on morphology. Increasing drying air mass flowrate had a small negative effect on particle size enlargement, but a positive effect on morphology. Finally, image analysis was found to be the preferred method to quantify the onset of agglomeration.

Mobile Minor is not very realistic. Furthermore, having a rotary atomizer or a pressure nozzle, or a cocurrent or counter-current air pattern can have a major impact on the results.
At small pilot scale, researchers measured air properties in a GEA Niro Minor [3] to predict regions in the dryer where particles would be sticky (Gianfrancesco et al., 2009). They also extended this work with computational fluid dynamics (CFD) modeling of spray drying of maltodextrin solutions [5] and performed agglomeration trials with fines dosing [6]. They concluded that for successful agglomeration, the particle surface condition is more important than the collision rate. However, the experiments were conducted on a system with a rotary atomizer, and it would be interesting to test the findings with a pressure nozzle. Williams et al. (2009) already compared the GEA Niro Minor to a bigger pilot scale spray dryer (evaporation capacity of 75 kg•h -1 ) and concluded that fines addition promotes agglomeration. Most favorable for agglomeration was a high feed flow rate with a low total solids content (TS) combined with a high flow of small fines. However, larger particles were obtained in situations with a high solids content of the feed and larger dry particles. Fröhlich et al. (2021) found that an increased feed TS created bigger agglomerates by shifting the collision outcomes from coalescence to agglomeration. However, Fröhlich et al. (2022) also found that an increase in TS decreases the relative fines mass flow, and that decreases the extent of agglomeration. In a multi-stage dryer, particles circulate until they are sufficiently large to Conventional experimental approaches are often one factor at a time (OFAT). A major drawback of this is that a high number of experiments are required, which is expensive and time-intensive. Additionally, interactions between variables and their combined effects on the response are not taken into account [11]. Design of experiments (DoE) approaches can assist in explaining variation of output as a function of a set of changing conditions in an experiment. In this respect response surface methodologies (RSM) such as full factorial design and central composite design are useful as these allow investigating the effects of multiple varying process variables on an output variable [12,13].
Therefore, this study aimed to investigate how spray drying processing factors affect the onset of nozzle zone agglomeration using a response surface methodology. A face-centered, central composite trial design was used for agglomeration trials on a pilot-scale spray dryer. A single-stage spray dryer was used, during which dry, small powder particles ("fines") were dosed in the nozzle zone to simulate one pass in an industrial multistage spray dryer. Input variables were the amount of fines dosed, drying air flow rate, and drying air inlet temperature, which were expected to affect both the collision probability and the sticking probability. The extent of agglomeration in the obtained powders was analyzed by calculating an agglomeration index (AI) based on the particle size distribution and by visual observations of particle morphology distributions using scanning electron microscopy (SEM) and a morphology analyser. Table 3). Runs were divided over three blocks or days. Within the blocks the runs were ordered on increasing temperature for practical reasons. The response factor was the agglomeration index (see 2.3.2). For every condition, a sample with and without fines dosing was run to correct for natural occurring agglomeration. A 40% w/w MD21 feed was prepared by adding the MD21 to hot tap water and stirring for at least 30 minutes until the feed became transparent. The atomization was kept constant throughout the trials at 21.2 ± 0.4 kg•h -1 with a pressure of 40 ± 2 bar using a SIY78/SKY16 high pressure nozzle from Spraying Systems Co. (Wheaton, Illinois, USA), therefore the outlet temperature varied. The feed temperature at atomization varied between 29.5 and 34.7°C.

Spray drying
All spray drying experiments were performed on a DW-350 single stage pilot-scale spray dryer from Spray Dry Works (the Netherlands) ( Figure 1) having a maximum drying capacity of 25 kg•h -1 and a drying chamber of 2 m in length and 1.5 m diameter. Ambient air was dehumidified using a Condair DA 1400 desiccant dryer (Switzerland) to a relative humidity of 5 % before an electrical heater heated it to the desired drying air inlet temperature ( ). The DW-350 spray dryer has a Rotaswirl RC350 air distributor (Spray Dry Works, the Netherlands), with a perforated inserted to reduce the air vortex. Outgoing air and powder were separated using a cyclone and the powder was collected through a rotary valve.
Fines were dosed with a screw, conveyed with ambient air using a small fan to the top of the spray dryer, and inserted concentrically to the nozzle at the inner inlet position ( Figure 2).

Powder analysis 2.3.1. Moisture content
The powder moisture content ( (%)) on total basis was determined in duplicate by oven drying at 105 °C overnight and calculated using Eq. (1).
where (g) and (g) are the sample masses before and after drying overnight, respectively.

Particle size analysis and agglomeration index
Particle size distributions (PSD) of the samples were determined in triplicate using laser diffraction with a Mastersizer 3000 (Malvern Panalytical, UK). An Aero S dry dispersion unit combined with a standard venturi dispenser was used to disperse the powder with a dispersion pressure of 3.5 bar.
To quantify the agglomeration caused by the fines dosing, the Agglomeration Index (AI) was calculated, as proposed by Williams (2007). The AI can be considered as a measure for the relative size enlargement occurring during one pass through a spray dryer. For this the mass fraction of primary feed MD21 particles ( 21 in kg•kg -1 ) in the final product was calculated using Eq. (2): where (kg•kg -1 ) is the total solids content of the feed, ̇ In which ((% volume)(log 10 (µm)) -1 ) is the volume weighted percentage of particles with diameters inside of a bin i (i.e. the volume of particles with diameters larger than d i-1 (µm) and smaller than d i (µm)).

J o u r n a l P r e -p r o o f
Journal Pre-proof The agglomeration index assigns a quantitative value for samples with a larger number of formed agglomerates. Since the PSDs are volume-based distributions, the formation of large agglomerates will skew the difference distribution since larger particle diameters contribute more volume relative to smaller particles. Thus, the formation of larger agglomerates will enlarge the AI more drastically than the formation of relatively smaller agglomerates.

Particle shape analysis
The obtained powder was visually observed by using a JCM-7000 SEM (JEOL, Japan). Carbon tape was used to secure the samples on the aluminum sample holder. Loose powder was removed using pressurized air and the samples were then coated with gold using a JEOL Smart-Coater (JEOL, Japan). The images were taken at an acceleration voltage of 5 kV.
Particle morphology of the powders was analyzed with a Malvern Morphologi 4 (Malvern, UK). For each sample, 19 mm 3 of powder was dispersed onto the glass plate and individual particles were photographed and using image analysis descriptive shape factors such as HS circularity, elongation and solidity were determined. For each sample >30,000 particles were analyzed. Using the shape factors, the particles were divided into primary, partially coalesced and agglomerated particles ( J o u r n a l P r e -p r o o f Table 2). Particles that did not fall within these classes, were classified as 'other' (N o ).

Response surface model development
A response-surface model was fit relating the AI values to the input variables of the CCD experiments using a linear least-squares regression algorithm. However, for the drying air flow rate there were large fluctuations because the flow rate at a given capacity depends on the air density, which changes with temperature. Therefore, the actual factor levels were calculated (Eq.(6), Table A. 1) and used instead of the expected levels in the response surface model.
where x i is the coded factor of the natural variable X i , where i is the subscript indicating the factor. The drying air temperature was x 1 , the fines mass flowrate was x 2 , and the drying air mass flowrate was x 3 .
,0 represents the value of the process variable at its midpoint: 1,0 = 180℃; 2,0 = 10.9 • ℎ −1 ; 3,0 = 541.3 • ℎ −1 . Δ ,1 and Δ ,−1 are the process variables at the low and high factorial levels. Δ is the average difference in the level as it is increased by 1 unit of : The coded values of all the process conditions were then used to fit the RSM models.
Firstly, a second-order model was fit to the AI using the rsm package (version 2.10.3) for R in RStudio (version 4.2.2). This model fits a second-order model with pure quadratic terms, two-way interaction terms, and linear terms (Eq. (7)).

(7)
Where 1 , 2 , 3 are the coded variables for drying air temperature, fines mass flowrate, and drying air flowrate respectively, 0 is the intercept (fitted value of AI when ( 1 , 2 , 3 ) = (0,0,0)), 1 , 2 , 3 are the J o u r n a l P r e -p r o o f Journal Pre-proof coefficients for the linear terms, 12 , 13 , 23 the coefficients for the two-way interaction terms between coded variables and 11 , 22 , 33 the coefficients for the pure quadratic terms.
Next, the model was refined by following stepwise regression. Here, terms are removed (i.e. interaction terms or quadratic terms) and analysis of variance (ANOVA) is computed to obtain the corrected Akaike Information Criteria (AIC). Via this procedure the model with the lowest AIC was found, describing the data, containing only significant terms, and lowest residuals. The model adequacy was checked by calculating the R 2 and adjusted-R 2 values. The same procedure was repeated to fit the model to the N agg .

Response Surface Model Development
The AI values were calculated ( J o u r n a l P r e -p r o o f Table 3) and fit to the second-order model presented in Eq. (7). From the model comparison, the model that described the data with the lowest AIC was derived (Eq. (8)), which also contained a blocking factor (D) distinguishing the days on which the experiments were performed. The model parameter estimates, standard error and significance of each term were determined (Table A.

Effect of processing conditions on AI
All three variables affected the AI. The AI was found to decrease as the drying air temperature ( 1 ) increased. This may be related to the effect of temperature on the drying rate, which changes the sticking probability. At higher temperatures, droplets dry more quickly and become more surface-dry and less sticky before colliding with incoming fines, resulting in reduced adhesion. These findings match earlier findings from Both et al. (2020), who observed a faster surface-dry skin forming around the droplets at higher air inlet temperatures, causing a lack of adhesion upon collision. Additionally, the authors found a higher fraction of particles that were likely to be agglomerates at lower temperatures.
The fines flowrate ( 2 )  are the specific heat capacities of air and water vapor, (kg·kg -1 ) is the absolute humidity of the inlet air, (K) is the inlet temperature of the drying air, Δℎ (kJ·kg -1 ·K -1 ) is the enthalpy of vaporization for water, and ̇ is the mass flowrate of the feed on total basis (kg·h -1 )). The AI decreases for increasing supplied ̇ (Figure 4), confirming that a higher drying rate reduces agglomeration. The AI drops less upon increasing ̇ for lower fines flowrates, which can be linked to the interaction term 12 1 2 from Eq. (5). For higher fines flowrates, this meant that the AI dropped more drastically because more fines were bouncing off and not contributing to particle enlargement of the sprayed droplets. The spread in AI values within each class of ̇ is partly caused by the different days on which the experiments were carried out, which has been taken into account in the model by the introduction of blocking factor D (eq. (8)).
When a dry fine particle collides with a completely wet droplet they coalesce and the dry particle is fully absorbed. The volume of such droplets after collision hardly increases and this type of collision is therefore undesired. During drying, the droplet changes from wet to sticky to dry, and the collision outcomes shift from coalescence, to sticking to bouncing. It was therefore expected that there would be an optimum drying rate. However, no drying rate optimum was found to be significant within the experimental range. It seems that more drying leads to less agglomeration, indicating a shift from sticking to bouncing. A drying rate optimum could be suspected from the curvature when plotting drying air flowrate versus temperature (Figure 3), where there was some curvature leading to an increase in AI as 1 was varied from -α to 0, and then a decrease as 1 was varied from 0 to α. However, since the quadratic term for 1 was not significant, it is not possible to conclude if this optimum of the drying rate

Predicting power of RSM
To test the predicting power of the RSM, during block 4 a test point (α,α,-α) was included to see how well the model could predict the AI for conditions outside of the experimental range. The model was capable of predicting the AI up to a 95% confidence interval ( Figure 5). In terms of agglomeration performance, this condition had a larger AI (0.189) than the CCD points, indicating that operating at a high fines flowrate and low massflow of drying air leads to more agglomeration. This verifies the trends uncovered by the model that a high fines flowrate and low massflow of drying air are beneficial for agglomeration. Possibly, with a low temperature (-α,α,-α) an even higher AI would have been measured, (AI predicted = 0.250), but the effect is assumed to be smaller than that of the other factors. There is less deviation from the trend in Figure 5 than in Figure 4 because the model incorporates the blocking factor D to correct for the spread caused by the different days on which the experiments were carried out.

Effect of processing conditions on particle morphology
Although the AI shows that particle size enlargement occurred, it does not indicate whether larger particles were formed by (partial) coalescence or agglomeration. However, different structures have distinct functional properties, leading to a preference for, for example, a grape structure. In literature, The number of agglomerated particles (N agg ) increases for increasing fines flowrates ( 2 ). This is in line with the AI, which is higher for those conditions. When no agglomeration would have occurred during fines dosing, the number of primary particles (N pp ) should have increased drastically for higher fines dosing rates due to the presence of more unagglomerated fines. The effect of the fines flowrate on the morphology of the samples is not as strong as on the AI. This can be explained since the morphology analyses were not compared to the "no-fines" condition, it is difficult to distinguish if the changes occur by increasing natural or forced agglomeration. The main conclusion that can be drawn is that adding fines results in the formation of more agglomerate clusters, which could be expected.
However, within each fines dosing rate, there is also variation, meaning that the drying air temperature and mass flow also affected the obtained morphologies. The drying air flowrate ( 3 ) increased N agg , which is opposite to its effect on AI. It is hypothesized that the increase in the air flowrate affected the smaller particles in the nozzle zone more than the larger ones. This makes physical sense as smaller particles have less inertia (Williams, 2007). This could have led to more collisions between fines and smaller spray particles, forming rather small agglomerates. This outcome could explain a decrease in AI J o u r n a l P r e -p r o o f Journal Pre-proof and an increase in N agg because the agglomerate subclass did not filter for size and can include small grape agglomerates. This result must be taken with caution as the smaller the particles are, the lower the relative resolution of the particles. Hence, the shape parameters calculated from the projection of a small agglomerate do not have the same accuracy as a large agglomerate. However, this trend describes a trade-off between particle enlargement (AI) and particle morphology (N agg ) since increasing the air flowrate decreases AI while increasing N agg .
Again, the effects of drying air temperature and drying air flowrate on morphology can be combined by calculating the ̇. Increasing the thermal energy supplied by drying air led to a decreasing trend in N pc at the expense of an increase in N pp . For a given fines flowrate, increasing the thermal energy seems to have reduced the partially coalesced fraction while keeping the agglomerate fraction unaffected ( Figure   7). Partial coalescence is only possible if the particles are still liquid enough so that they can penetrate before the larger particle dries. The results indicate that when ̇ was increased in the nozzle zone, the droplets dried more rapidly, reducing the incidence of coalescence while relatively increasing the incidence of agglomeration (particles penetrated less than 50% of their diameter). Moreover, increasing ̇ led to the formation of more primary particles, which is most probably related to the over-drying of droplets resulting in surface-dry skins [15,20]. These results point to lower drying rates yielding a size enlargement that might not be as desired as higher drying rates as partial coalesced particles do not bring the functional benefits that grape-type agglomerates do. However, these results only apply for a single pass through the nozzle zone and may change when the size of the dry particles changes upon circulating through the nozzle zone more times.
J o u r n a l P r e -p r o o f

Comparing nozzle zone agglomeration by microscopy
Because the AI and the N agg give contradicting results, the samples were also visually compared using SEM ( Figure 8). The samples that seem to be more agglomerated, have a higher N agg value, but lower AI values. It is important to bear in mind that the AI is a measure for particle volume increase upon fines addition. Samples with higher AI values do not necessarily have a larger particle size, but they have grown the most compared to the condition without fines addition. The AI is based on measurements with laser diffraction, a method based on the principle that particles of different sizes scatter the light differently. The particle size is then presented as the diameter of a sphere with the same volume [21].
However, the particles are often largely spherical, but with a fine particle attached to them ( Figure 8).
The AI is based on volume increase, but the adherence of a fine to a coarse particle hardly affects its total volume. With the morphological classes from the image analysis, these small attached particles are better taken into account. This means that the morphological analyses represent the samples better, making image analysis the preferred method to compare the agglomeration between the samples. This applies to the investigation of the onset of nozzle zone agglomeration. To investigate the agglomeration of powders produced in a multi-stage spray dryer, the AI may be much more indicative than here. This is because the powder in that case would consist only of agglomerates and not a mixture of primary particles, partially coalesced particles, and agglomerates.

Conclusion
The effect of drying air temperature (160 -200 °C), fines flowrate (7.1 ± 1.2 -15.9 ± 0.5 kg•h -1 ), and drying air flowrate (472.8 ± 6.2 -590.8 ± 9.9 kg•h -1 ) on nozzle zone agglomeration was systematically investigated using response surface modeling. The results showed that fitting a response-surface model to a designed experiment provided a way to statistically study a complex system with relative ease and efficiency. The experimental design consisted of a central-composite design with eight factorial J o u r n a l P r e -p r o o f Journal Pre-proof treatments, twelve axial treatments (six treatments in duplicate), and six center treatments. To compare size enlargement due to fines supplementation across treatments, an agglomeration index (AI) was applied. To compare morphological differences across treatments, static image analysis methods were applied to separate the particles into subclasses.
It was found that conditions that favored size enlargement of powder particles were those with low drying air temperatures and flowrates, and high fines supplementation flowrates. Regarding morphology, an increase in fines flowrate and air flowrate led to more grape-like cluster agglomerates being formed. From these three factors, the fines supplementation flowrate was the most significant in impacting both the size enlargement as well as the development of grape-like clusters.
It was hypothesized that conditions that increased the collision frequency in the nozzle zone (i.e. increase in fines mass flowrate, decrease in air flowrate) would lead to higher extent of agglomeration.
This was confirmed when analyzing the size enlargement due to fines supplementation. Interestingly, increasing the air flowrate negatively affected size enlargement while improving agglomerate quality. An improvement in agglomerate quality consisted of an increase in the fraction of non-circular, irregularlyshaped grape clusters. Considering the limitations of laser diffraction, which form the basis of the AI, division into morphological classes based on image analysis is the preferred method to compare the onset of agglomeration between samples.
It was hypothesized that increasing the drying air temperature would show a maximum positive effect on agglomeration outcomes up to a critical value as the droplet surface changed from wet to sticky. After this critical value, the droplet surface would become too dry, causing more fines to bounce off it instead of agglomerating. The obtained results were not conclusive when analyzing the drying air temperature, as the effect of temperature on AI was small and not significant on N agg . However, considering the thermal energy provided by the drying air (Q da ), samples that were exposed to higher drying rates J o u r n a l P r e -p r o o f   J o u r n a l P r e -p r o o f Journal Pre-proof