Systematic Study of the Effects of High Shear Granulation Parameters on Process Yield, Granule Size, and Shape by Dynamic Image Analysis

The aim of the work was to analyze the influence of process parameters of high shear granulation on the process yield and on the morphology of granules on the basis of dynamic image analysis. The amount of added granulation liquid had a significant effect on all monitored granulometric parameters and caused significant changes in the yield of the process. In regard of the shape, the most spherical granules with the smoothest surface were formed at a liquid to solid ratio of ≈1. The smallest granules were formed at an impeller speed of 700 rpm, but the granules formed at 500 rpm showed both the most desirable shape and the highest process yield. Variation in the shape factors relied not only on the process parameters, but also on the area equivalent diameter of the individual granules in the batch. A linear relationship was found between the amount of granulation liquid and the compressibility of the granules. Using response surface methodology, models for predicting the size of granules and process yield related to the amount of added liquid and the impeller speed were generated, on the basis of which the size of granules and yield can be determined with great accuracy.

• Dynamic image analysis was used to analyze the morphology of high shear granules; • The granules were investigated for size, shape, and surface roughness; • A small change in liquid content caused changes in granule shape and process yield; • The amount of granulation liquid had a significant effect on all parameters; • A model for predicting the granule size was created from the experimental data.

Introduction
A wet granulation process is predominantly used in the pharmaceutical industry to manufacture tablets, which are the most widely used dosage form. The high shear wet granulation (HSWG) represents the most common method, however, fluid bed granulation (FBG) process is recently also frequent [1]. HSWG is a process of size enlargement of primary particles joined together by agitation and a liquid binder. Major advantages Several authors have investigated various morphological parameters for the characterization of particle shape [42][43][44][45][46][47]. Xiu et al. [48] pointed out that the flow properties of different types of MCC are directly influenced by fractal dimension and circularity. However, Almeida-Prieto et al. [49] have shown that the shape of the granules cannot be characterized correctly with only one shape factor. To fill the literature gap, the aim of our work is to systematically analyze the influence of selected HSWG process parameters on the granule morphology and process yield. Using the DIA method on a PartAn 3D device, the shape parameters sphericity, aspect ratio, roundness, and concavity will be monitored at different L/S ratios and impeller speeds; selected samples of granules will be analyzed using an electron microscope. Employing response surface methodology, the possibility to reach an acceptable model for prediction of the size of the granules will be investigated.

Material and Methods
All measurements and manipulations were carried out at a controlled ambient temperature of 24.0 ± 2.0 • C and relative air humidity 40.0 ± 12.0% analyzed by hygrometer E5005 (Emos, Prerov, Czech Republic).

Experimental Material
Avicel ® PH 101-microcrystalline cellulose (FMC Biopolymer, Cork, Ireland), a common pharmaceutical excipient obtained from the IMCD Czech Republic, was used in the experiments. This material is regularly used in the pharmaceutical industry as a filler or binder in oral dosage forms. Information about the true density (1580 kg/m 3 ) and the bulk density (333 kg/m 3 ) was provided by the manufacturer. An aqueous solution of polyvinylpyrrolidone powder K30 (BioChemica, Billingham, UK) was used as a granulation binder. The polyvinylpyrrolidone was prepared as a 2% w/w aqueous solution in distilled water at a temperature of 38 • C by a Stuart 162 magnetic stirrer (Stuart Equipment, Staffordshire, UK) until the powder particles were completely until the powder particles were completely dissolved.

Particle Size Analysis
The particle size and size distribution were analyzed using a Malvern Mastersizer 3000 Laser Diffraction Analyzer (Malvern Panalytical, Malvern, UK). Characteristic dimensions of the experimental material were assessed volumetrically, and samples were analyzed by the dry cell method. The experiment was repeated three times. The results were then interpreted as an average from three measurements. Measurements were performed using an Aero S dry powder dispersion unit. The polystyrene latex standard was used for calibration according to ISO 13320:2009. The particle refractive index and the particle absorption index were 1.5 and 0.1, respectively.

Water Sorption Analysis
The sorption and desorption kinetics of water vapor applied to a sample of the investigated material were measured using a gravimetric fully automated Aquadyne DVS dynamic vapor sorption device (Quantachrome Instrumens, Hartley Wintney, UK). The measurements were performed at a constant temperature of 22.5 • C. Microcrystalline cellulose CRM 302 MCC was used as a standard for the device calibration. A sample of a defined weight of 45 mg was placed on analytical balance in the instrument, then a re-testing process was started, where the sample was cyclically exposed to changes in relative humidity and the dependence of the change in weight of the examined material on ambient humidity was measured. When the humidity limit is reached, the humidity increase cycle stops and the desorption process begins. The output of the measurement is the dependence between the change in the weight of the examined sample and the relative humidity for the adsorption and desorption process. The actual moisture of the powder was analyzed using a moisture analyzer MB 60 (VWR, Lutterworth, UK).

High Shear Granulation
Avicel PH101 was granulated by high shear wet granulation (HSWG) with a vertical, three-bladed bottom driven granulator of our own production [50]. The amount of Avicel per batch was 500 mL, i.e., 167 g. As the granular material was pure Avicel, it was not necessary to mix the material before the HSWG process. At the set chopper speed (1000 rpm) and impeller speeds (300, 500, and 700 rpm), granulation liquid was sprayed into the granulator at a flow rate of 22.5 mL/min. After the addition of the granulation liquid was completed, the whole wet mass was mixed for another 5 min, at the same impeller and chopper speeds. An appropriately set combination of impeller and chopper speeds ensures a balance, allowing establishment of good toroidal flow to promote granule growth [51]. The experiment was repeated twice, for the same duration of the granulation liquid adding.

Classification of Franules
The wet granules from the HSWG process were weighed using a Nimbus NBL 254i (Adam Equipment, Kingstone, UK). After weighing, the wet granules were dried at 60 • C for 24 h. The dried granules were weighed again. The liquid to solid ratio (L/S ratio) parameter was determined as the difference between the weight of the wet and dry granulate divided by the weight of the dry granulate to accurately express the liquid content of the granules. The dried granules were classified as fine (<0.4 mm), yield (product) (0.4 up to 3 mm) and coarse (>3 mm) using sieve analysis by a Vibratory Shaker (Fritsch, Oberstein, Germany, with sieves of 0.4 mm and 3 mm. The individual fractions of the classified granules were weighed again.

Measurement of Granule Size and Shape
The size distribution and shape of the granules were analyzed using a PartAn 3D device (Microtrac MRB, Haan, Germany). The measurements were performed according to ISO 13322-2 and 9276-6 standards for dynamic image analysis [52]. Dynamic image analysis (DIA) is a technique that characterizes granules in motion by digitalizing photos of each granule from a CCD (charge-coupled device) camera (100 photos per second) and storing them in an image file. The particles tumble and rotate and are illuminated by stroboscopic light. The images are used to calculate morphological parameters based on the known size and location of the pixels in each image. The principle of measurement is shown in the figure (Figure 1).

High Shear Granulation
Avicel PH101 was granulated by high shear wet granulation (HSWG) with a vertical, three-bladed bottom driven granulator of our own production [50]. The amount of Avicel per batch was 500 mL, i.e., 167 g. As the granular material was pure Avicel, it was not necessary to mix the material before the HSWG process. At the set chopper speed (1000 rpm) and impeller speeds (300, 500, and 700 rpm), granulation liquid was sprayed into the granulator at a flow rate of 22.5 mL/min. After the addition of the granulation liquid was completed, the whole wet mass was mixed for another 5 min, at the same impeller and chopper speeds. An appropriately set combination of impeller and chopper speeds ensures a balance, allowing establishment of good toroidal flow to promote granule growth [51]. The experiment was repeated twice, for the same duration of the granulation liquid adding.

Classification of Franules
The wet granules from the HSWG process were weighed using a Nimbus NBL 254i (Adam Equipment, Kingstone, United Kingdom). After weighing, the wet granules were dried at 60 °C for 24 h. The dried granules were weighed again. The liquid to solid ratio (L/S ratio) parameter was determined as the difference between the weight of the wet and dry granulate divided by the weight of the dry granulate to accurately express the liquid content of the granules. The dried granules were classified as fine (<0.4 mm), yield (product) (0.4 up to 3 mm) and coarse (>3 mm) using sieve analysis by a Vibratory Shaker (Fritsch, Oberstein, Germany, with sieves of 0.4 mm and 3 mm. The individual fractions of the classified granules were weighed again.

Measurement of Granule Size and Shape
The size distribution and shape of the granules were analyzed using a PartAn 3D device (Microtrac MRB, Haan, Germany). The measurements were performed according to ISO 13322-2 and 9276-6 standards for dynamic image analysis [52]. Dynamic image analysis (DIA) is a technique that characterizes granules in motion by digitalizing photos of each granule from a CCD (charge-coupled device) camera (100 photos per second) and storing them in an image file. The particles tumble and rotate and are illuminated by stroboscopic light. The images are used to calculate morphological parameters based on the known size and location of the pixels in each image. The principle of measurement is shown in the figure (Figure 1).  Both the granules and the powdered Avicel were evaluated volumetrically. The description of the size parameters is in Table 1. The characteristic dimension for determining the size of a granule was the Area Equivalent Diameter D a Equation (1) obtained from multiple photos of individual granules.
where A is the area of the projected particle. Mean granule size was calculated according to the relationship in Equation (2).
where FL is the Feret length. The aspect ratio AR Equation (5) was defined as thickness to the particle length expressed by Feret parallel tangent dimensions. The value AR = 1 corresponds to a perfect sphere.
= , where FT is the Feret thickness. The largest Feret size measured in a series of individual particle images was assigned as the 3D length FL and the smallest Feret size was assigned as the thickness FT of that particle [52]. The last parameter describing the shape of the particles was concavity c Equation (6). The value c = 1 describes an extremely rough, spikey surface.
where CHA is convex Hull area. In the same way as dmean, mean sphericity, roundness, aspect ratio and concavity were also determined as mean values from all granules classified as product in the whole batch.
where FL is the Feret length. The aspect ratio AR Equation (5) was defined as thickness to the particle length expressed by Feret parallel tangent dimensions. The value AR = 1 corresponds to a perfect sphere.
= , where FT is the Feret thickness. The largest Feret size measured in a series of individual particle images was assigned as the 3D length FL and the smallest Feret size was assigned as the thickness FT of that particle [52]. The last parameter describing the shape of the particles was concavity c Equation (6). The value c = 1 describes an extremely rough, spikey surface.
where CHA is convex Hull area. In the same way as dmean, mean sphericity, roundness, aspect ratio and concavity were also determined as mean values from all granules classified as product in the whole batch.
where FL is the Feret length. The aspect ratio AR Equation (5) was defined as thickness to the particle length expressed by Feret parallel tangent dimensions. The value AR = 1 corresponds to a perfect sphere.
= , where FT is the Feret thickness. The largest Feret size measured in a series of individual particle images was assigned as the 3D length FL and the smallest Feret size was assigned as the thickness FT of that particle [52]. The last parameter describing the shape of the particles was concavity c Equation (6). The value c = 1 describes an extremely rough, spikey surface.
where CHA is convex Hull area. In the same way as dmean, mean sphericity, roundness, aspect ratio and concavity were also determined as mean values from all granules classified as product in the whole batch.

Feret thickness FT
Minimum distance between parallel tangents. Measured as minimum particle width in the sequence of 3D images.

Microscopic Analysis
The surface of the granules was analyzed using a JSM electron microscope (JEOL Ltd., Tokyo, Japan). The examined sample was sprayed with a layer of platinum nanoparticles before measurement in order to obtain a conductive surface. Next, the sample was placed in the working chamber of the microscope, where a vacuum was created. By the action of a beam with voltage 10.0 kV, the material was irradiated and the subsequent application of the secondary electron detector method (SED) revealed electrons reflected from the surface of the sample.

Equivalent Perimeter diameter D p
Spheres of equal perimeter to the original particle. Average area in the sequence of 3D images.

Area Equivalent diameter Da
Spheres of equal area to the original particle. Average area in the sequence of 3D images.
Equivalent Perimeter diameter

Dp
Spheres of equal perimeter to the original particle. Average area in the sequence of 3D images.

Feret length FL
Maximal distance between parallel tangents.

Feret thickness FT
Minimum distance between parallel tangents. Measured as minimum particle width in the sequence of 3D images.

Microscopic Analysis
The surface of the granules was analyzed using a JSM electron microscope (JEOL Ltd., Tokyo, Japan). The examined sample was sprayed with a layer of platinum nanoparticles before measurement in order to obtain a conductive surface. Next, the sample was placed in the working chamber of the microscope, where a vacuum was created. By the action of a beam with voltage 10.0 kV, the material was irradiated and the subsequent application of the secondary electron detector method (SED) revealed electrons reflected from the surface of the sample. Spheres of equal area to the original particle. Average area in the sequence of 3D images.
Equivalent Perimeter diameter

Dp
Spheres of equal perimeter to the original particle. Average area in the sequence of 3D images.

Feret length FL
Maximal distance between parallel tangents.

Feret thickness FT
Minimum distance between parallel tangents. Measured as minimum particle width in the sequence of 3D images.

Microscopic Analysis
The surface of the granules was analyzed using a JSM electron microscope (JEOL Ltd., Tokyo, Japan). The examined sample was sprayed with a layer of platinum nanoparticles before measurement in order to obtain a conductive surface. Next, the sample was placed in the working chamber of the microscope, where a vacuum was created. By the action of a beam with voltage 10.0 kV, the material was irradiated and the subsequent application of the secondary electron detector method (SED) revealed electrons reflected from the surface of the sample.

Feret thickness F T
Minimum distance between parallel tangents. Measured as minimum particle width in the sequence of 3D images.

Area Equivalent diameter Da
Spheres of equal area to the original particle. Average area in the sequence of 3D images.
Equivalent Perimeter diameter

Dp
Spheres of equal perimeter to the original particle. Average area in the sequence of 3D images.

Feret length FL
Maximal distance between parallel tangents.

Feret thickness FT
Minimum distance between parallel tangents. Measured as minimum particle width in the sequence of 3D images.

Microscopic Analysis
The surface of the granules was analyzed using a JSM electron microscope (JEOL Ltd., Tokyo, Japan). The examined sample was sprayed with a layer of platinum nanoparticles before measurement in order to obtain a conductive surface. Next, the sample was placed in the working chamber of the microscope, where a vacuum was created. By the action of a beam with voltage 10.0 kV, the material was irradiated and the subsequent application of the secondary electron detector method (SED) revealed electrons reflected from the surface of the sample.

Compressibility of Granules
Various shape parameters such as sphericity ϕ were used to determine the shape of the granulate Equation (3). Sphericity value ϕ = 1 corresponds to a perfect sphere.
where D p is the equivalent perimeter diameter calculated as (p/π). Another parameter is roundness r Equation (4). Value r = 1, corresponds to a perfectly circular particle.
where F L is the Feret length. The aspect ratio AR Equation (5) was defined as thickness to the particle length expressed by Feret parallel tangent dimensions. The value AR = 1 corresponds to a perfect sphere.
where F T is the Feret thickness. The largest Feret size measured in a series of individual particle images was assigned as the 3D length F L and the smallest Feret size was assigned as the thickness F T of that particle [52]. The last parameter describing the shape of the particles was concavity c Equation (6). The value c = 1 describes an extremely rough, spikey surface.
where CH A is convex Hull area. In the same way as d mean , mean sphericity, roundness, aspect ratio and concavity were also determined as mean values from all granules classified as product in the whole batch.

Microscopic Analysis
The surface of the granules was analyzed using a JSM electron microscope (JEOL Ltd., Tokyo, Japan). The examined sample was sprayed with a layer of platinum nanoparticles before measurement in order to obtain a conductive surface. Next, the sample was placed in the working chamber of the microscope, where a vacuum was created. By the action of a beam with voltage 10.0 kV, the material was irradiated and the subsequent application of the secondary electron detector method (SED) revealed electrons reflected from the surface of the sample.

Compressibility of Granules
A FT4 powder rheometer (Freeman Technology, Tewkesbury, UK) was used to determine the compressibility of Avicel powder and granules in accordance with ASTM-D7891-15 (2015) standard [53]. The batch of experimental material for one measurement was 60 g for Avicel powder and 80 g for granules. Compressibility measurements were performed in two vessels with a diameter of 50 mm and a volume of 85 mL. Prior to the compressibility measurement itself, the sample was subjected to three conditioning cycles, using a blade, in order to obtain a homogeneously packed powder bed. Excess material was split to achieve a defined bottom vessel volume 85 mL. Normal stress in a range of 0.5-15 kPa was applied to the pretreated sample. Based on the position of the piston, the height occupied by the test material before and after the compression test was determined and the volume of material was calculated using the vessel geometry and the measured heights. Compressibility (CPS) is calculated from the change in volume of the tested sample after compression, expressed as a percentage [54]. Despite the fact that compressibility is not a direct indicator of flow properties, it is related to several operations such as storage in hoppers or the behavior of bulk materials during roller compaction [55].

Data Processing
OriginPro 9.0 software (OriginLab Corporation, Northampton, MA, USA) was used to process the experimental data and box charts. Design Expert 13 (StatEase Inc., Minneapolis, MN, USA) was used for the creation of models. Response surface methodology with a quadratic model was used to correlate responses and factors (independent variables). The process parameters impeller speed and L/S ratio were chosen as independent variables.
Parameters d mean and process yield were chosen as the response (dependent variable). A general polynomial equation was used to create the model Equation (7) where Y is a dependent variable (response), β 0 is a constant, β 1 and β 2 are linear coefficients, β 11 and β 22 are quadratic coefficients, β 12 is the interaction coefficient. X 1 and X 2 are the coded values of independent variables. Multilinear analysis and analysis of variance (ANOVA) were performed to established statistical significance [56].

Evaluation of Particle Size of Avicel PH 101
Values d 10 (23.9 µm), d 50 (64.2 µm) and d 90 (140 µm) were determined by the dry cell method. These are the particle sizes corresponding to 10%, 50%, and 90% of the cumulative volume distribution curve. The distribution curve of the particles was unimodal ( Figure 2). The span value of 1.807 calculated as (d 90 − d 10 )/d 50 and Sauter volume/surface diameter d 32 of 46 µm were detected. The particle shape will be discussed later.
Pharmaceutics 2021, 13, x FOR PEER REVIEW variables. Parameters dmean and process yield were chosen as the response (dependen iable). A general polynomial equation was used to create the model Equation (7) = 0 + 1 1 + 2 2 + 11 1 2 where Y is a dependent variable (response), 0 is a constant, 1 and 2 are linear ficients, 11 and 22 are quadratic coefficients, 12 is the interaction coefficient. X X2 are the coded values of independent variables. Multilinear analysis and analysis o iance (ANOVA) were performed to established statistical significance [56].

Evaluation of Particle Size of Avicel PH 101
Values d10 (23.9 μm), d50 (64.2 μm) and d90 (140 μm) were determined by the dr method. These are the particle sizes corresponding to 10%, 50%, and 90% of the cumu volume distribution curve. The distribution curve of the particles was unimodal (Figu The span value of 1.807 calculated as (d90 − d10)/d50 and Sauter volume/surface diamet of 46 μm were detected. The particle shape will be discussed later.

Evaluation of Water Sorption of Avicel PH 101
Avicel PH101 particles tend to increase in volume with increasing liquid con which is desirable when tablets disintegrate. However, due to changes in relative hu ity, Avicel PH101 particles can absorb significant amounts of moisture. Figure 3 show adsorption and desorption isotherms analyzed by Aquadyne DVS. At 95% relativ midity, the increase in the weight of the analyzed sample was up by 15.8%, meaning the weight of the sample increased to 71.1 mg compared to the original 45 mg. Th sorption cycle was able to dry the sample at a relative humidity of 12.6% to a value w the weight gain was 0.17%. Sun [57] pointed out that moisture trapped in Avicel par can significantly influence its behavior during the compaction process. The moistu our Avicel PH101 powder, analyzed by a MB 60 moisture analyzer, was 4.0 ± 0.2%.

Evaluation of Water Sorption of Avicel PH 101
Avicel PH101 particles tend to increase in volume with increasing liquid content, which is desirable when tablets disintegrate. However, due to changes in relative humidity, Avicel PH101 particles can absorb significant amounts of moisture. Figure 3 shows the adsorption and desorption isotherms analyzed by Aquadyne DVS. At 95% relative humidity, the increase in the weight of the analyzed sample was up by 15.8%, meaning that the weight of the sample increased to 71.1 mg compared to the original 45 mg. The desorption cycle was able to dry the sample at a relative humidity of 12.6% to a value where the weight gain was 0.17%. Sun [57] pointed out that moisture trapped in Avicel particles can significantly influence its behavior during the compaction process. The moisture of our Avicel PH101 powder, analyzed by a MB 60 moisture analyzer, was 4.0 ± 0.2%.

Evaluation of Granule Size
Graphical dependence of dmean on the L/S ratio parameter is illustrated in Figure 4a for all impeller speeds. With increasing L/S, the dmean of the formed granules increased almost linearly, up to the value L/S = 1.0. Within the L/S ratio 0.9-1.0, a "stagnant zone" (or the mild decrease in size) in the value of dmean was noted. After exceeding the L/S ratio 1.0, the size of the granules increased exponentially, which indicates that the saturation point has been achieved and surpassed. The excess binder solution is then available for extensive granule growth by coalescence. A similar trend can be found in the work by Shi et al. [58] during granulation of Avicel PH101 using distilled water and also in the work by Chitu et al. [59] during granulation of MCC with lactose. Osei-Yeboah et al. [60] found that at more than 45% water content, MCC-PVP granules flow well but cannot be compressed into intact tablets. Such changes in powder performance correspond to the rapid growth into large and dense spheres with a smooth surface.
A significant effect of impeller speed on the mean granule sizes for all experiments is summarized in Figure 4b in a visual form of box charts. For better clarity, the points in the graph relating to individual experiments are included. In order to maintain a uniform

Evaluation of Granule Size
Graphical dependence of d mean on the L/S ratio parameter is illustrated in Figure 4a for all impeller speeds. With increasing L/S, the d mean of the formed granules increased almost linearly, up to the value L/S = 1.0. Within the L/S ratio 0.9-1.0, a "stagnant zone" (or the mild decrease in size) in the value of d mean was noted. After exceeding the L/S ratio 1.0, the size of the granules increased exponentially, which indicates that the saturation point has been achieved and surpassed. The excess binder solution is then available for extensive granule growth by coalescence.

Evaluation of Granule Size
Graphical dependence of dmean on the L/S ratio parameter is illustrated in Figure 4a for all impeller speeds. With increasing L/S, the dmean of the formed granules increased almost linearly, up to the value L/S = 1.0. Within the L/S ratio 0.9-1.0, a "stagnant zone" (or the mild decrease in size) in the value of dmean was noted. After exceeding the L/S ratio 1.0, the size of the granules increased exponentially, which indicates that the saturation point has been achieved and surpassed. The excess binder solution is then available for extensive granule growth by coalescence. A similar trend can be found in the work by Shi et al. [58] during granulation of Avicel PH101 using distilled water and also in the work by Chitu et al. [59] during granulation of MCC with lactose. Osei-Yeboah et al. [60] found that at more than 45% water content, MCC-PVP granules flow well but cannot be compressed into intact tablets. Such changes in powder performance correspond to the rapid growth into large and dense spheres with a smooth surface.
A significant effect of impeller speed on the mean granule sizes for all experiments is summarized in Figure 4b in a visual form of box charts. For better clarity, the points in the graph relating to individual experiments are included. In order to maintain a uniform A similar trend can be found in the work by Shi et al. [58] during granulation of Avicel PH101 using distilled water and also in the work by Chitu et al. [59] during granulation of MCC with lactose. Osei-Yeboah et al. [60] found that at more than 45% water content, MCC-PVP granules flow well but cannot be compressed into intact tablets. Such changes in powder performance correspond to the rapid growth into large and dense spheres with a smooth surface.
A significant effect of impeller speed on the mean granule sizes for all experiments is summarized in Figure 4b in a visual form of box charts. For better clarity, the points in the graph relating to individual experiments are included. In order to maintain a uniform Pharmaceutics 2021, 13, 1894 9 of 21 range of L/S ratio, three experiments were excluded from the analyzes at L/S < 0.70 and impeller speed 500 rpm. In a numerical form, data of d mean for different impeller speeds are presented in Table 2. The granules obtained at higher impeller speeds have a smaller particle size due to more pronounced dynamic stresses followed by subsequent breakage and attrition. The granules at an impeller speed of 300 rpm ranged in size from 0.89 to 2.02 mm. At an impeller speed of 500 rpm, the granules had a size of 0.72-1.6 mm, while the size of 0.66-1.23 mm was detected at 700 rpm. Rahmanian et al. [61] proved that granules formed at higher impeller speeds also have higher crushing strength due to lower porosity.
The width of the granule size distribution decreased as the impeller speed increased; it was the narrowest at a speed of 700 rpm (Figure 4b). Mean granule size from all experiments, analyzed by box charts at an impeller speed 700 rpm was d mean = 0.826 mm. At an impeller speed of 300 rpm it was up to 1.244 mm. Since some of the experiments are located in the area of L/S < 1, where the growth of the granules was not so significantly affected by the amount of added liquid, the median of the distribution curve was 1.06 mm (300 rpm), 0.88 mm (500 rpm), and 0.78 mm (700 rpm).

Evaluation of Process Yield
A graphical dependence between the process yield and the L/S ratio is shown in Figure 5a. With an increasing L/S ratio, the proportion of required granule fraction marked as yield (0.4 up to 3 mm) increased, however, the results were influenced by impeller speed as well. At an impeller speed of 300 rpm, the increase in yield is slower than at higher speeds and reached the value of 80% at the high L/S ratio of 1.12. With a further increase in L/S, the yield fraction increased to above 90%. The decrease in the curve at L/S > 1.25 shown in Figure 4a at 300 rpm was due to the removal of larger granules, which were classified as coarse. At impeller speeds of 500 and 700 rpm, contrarily, the content of yield granule portion raised sharply when exceeding the L/S ratio of 0.87 and significant changes in the process yield were observed particularly in the L/S range from 0.9 to 1.0. The 80% limit was achieved at a L/S ratio of approximately 0.97 at both speeds. After exceeding this value, the yield continued to grow slowly, achieving process yield >90% at a L/S ratio of 1.2 The graph shows that the highest process yields were achieved at higher L/S ratio values and higher impeller speeds. value, the yield continued to grow slowly, achieving process yield >90% at a L/S ratio of 1.2 The graph shows that the highest process yields were achieved at higher L/S ratio values and higher impeller speeds. In Figure 5b, a summary box chart of the yield fraction distribution for each impeller speed are shown; in a numerical form, data are summarized in Table 2. The highest yield with the narrowest distribution was detected at an impeller speed of 500 rpm with the mean yield 67.6%. However, most experiments had a process yield above 80%, which was reflected in a median size yield of 84.6% at 500 rpm. At 700 rpm, the median was 79.8%. On the other hand, the lowest median yield value of 61.6% was noted in experiments at 300 rpm impeller speed.

Evaluation of Sphericity
Spherical shape is a significant advantage for good flowability [62]. The relationship of mean sphericity φ of the formed granules on the L/S ratio and impeller speeds used was investigated and is illustrated in Figure 6a. At an impeller speed of 300 rpm, sphericity firstly decreased with increasing L/S ratio up to 1.0; then started to rise again. At L/S ratios >1.0, almost all granule samples showed a sphericity value φ > 0.9. At the highest impeller speed of 700 rpm, a similar sphericity decrease at low L/S ratios was detected with the values of φ < 0.9 in the L/S ratio range of 0.7-0.85. Then, the granules showed higher values of φ (around 0.94) within the L/S ratio range of 0.92-1.04, decreasing mildly at higher L/S ratios, but still remaining above 0.9. The best results were noted at 500 rpm. The granules sphericity near to 0.9 (φ = 0.87) occurred even at low L/S. The value raised above 0.9 at L/S ratios higher than 0.87. Bouwman et al. [63] found that spherical and regularly shaped granules in the formulation containing MCC were formed particularly with a L/S ratio = 1.0. Our results at higher impeller speeds are in a good agreement with this observation. In Figure 5b, a summary box chart of the yield fraction distribution for each impeller speed are shown; in a numerical form, data are summarized in Table 2. The highest yield with the narrowest distribution was detected at an impeller speed of 500 rpm with the mean yield 67.6%. However, most experiments had a process yield above 80%, which was reflected in a median size yield of 84.6% at 500 rpm. At 700 rpm, the median was 79.8%. On the other hand, the lowest median yield value of 61.6% was noted in experiments at 300 rpm impeller speed.

Evaluation of Sphericity
Spherical shape is a significant advantage for good flowability [62]. The relationship of mean sphericity ϕ of the formed granules on the L/S ratio and impeller speeds used was investigated and is illustrated in Figure 6a. At an impeller speed of 300 rpm, sphericity firstly decreased with increasing L/S ratio up to 1.0; then started to rise again. At L/S ratios > 1.0, almost all granule samples showed a sphericity value ϕ > 0.9. At the highest impeller speed of 700 rpm, a similar sphericity decrease at low L/S ratios was detected with the values of ϕ < 0.9 in the L/S ratio range of 0.7-0.85. Then, the granules showed higher values of ϕ (around 0.94) within the L/S ratio range of 0.92-1.04, decreasing mildly at higher L/S ratios, but still remaining above 0.9. The best results were noted at 500 rpm. The granules sphericity near to 0.9 (ϕ = 0.87) occurred even at low L/S. The value raised above 0.9 at L/S ratios higher than 0.87. Bouwman et al. [63] found that spherical and regularly shaped granules in the formulation containing MCC were formed particularly with a L/S ratio = 1.0. Our results at higher impeller speeds are in a good agreement with this observation.
In Figure 6b, a box chart of mean sphericity distributions for each impeller speed is presented; a numerical form is summarized in Table 2. The graph confirms the narrowest distribution of sphericity at 500 rpm. Mean sphericity values of 0.875 ± 0.049, 0.923 ± 0.019, and 0.908 ± 0.043 were noted at 300, 500, and 700 rpm, respectively. The median value of ϕ = 0.91 was observed at 300 rpm; at 500 and 700 rpm, the same value 0.93 was found. These median values from box chart analysis values were not so significantly affected by sphericity values at lower L/S ratios. Thus, it can be concluded that more spherical granules were formed at higher impeller speeds.
On the basis of sphericity, Maroof et al. [64] classified the particles as follows ϕ = 0.45 to 0.6-medium sphericity, ϕ = 0.6 to 0.8-spherical and ϕ = 0.8 to 1.0-high sphericity. According to this classification, most granules prepared in this work were categorized as highly spherical.  In Figure 6b, a box chart of mean sphericity distributions for each impeller speed is presented; a numerical form is summarized in Table 2. The graph confirms the narrowest distribution of sphericity at 500 rpm. Mean sphericity values of 0.875 ± 0.049, 0.923 ± 0.019, and 0.908 ± 0.043 were noted at 300, 500, and 700 rpm, respectively. The median value of φ = 0.91 was observed at 300 rpm; at 500 and 700 rpm, the same value 0.93 was found. These median values from box chart analysis values were not so significantly affected by sphericity values at lower L/S ratios. Thus, it can be concluded that more spherical granules were formed at higher impeller speeds.
On the basis of sphericity, Maroof et al. [64] classified the particles as follows φ = 0.45 to 0.6-medium sphericity, φ = 0.6 to 0.8-spherical and φ = 0.8 to 1.0-high sphericity. According to this classification, most granules prepared in this work were categorized as highly spherical.

Evaluation of Aspect Ratio
Similarly to sphericity, the Aspect ratio AR parameter showed higher values with an increasing L/S ratio. The dependence between AR and L/S ratios is shown in Figure 7a. Out of all experimental runs, the granules formed at an impeller speed 300 rpm reached the highest values of AR as well as the highest data variability. The AR firstly decreased (minimum AR = 0.4 at L/S = 0.92), then it rose sharply with an increasing content of the liquid used to form the granules. The highest values were noted at L/S > 1.1 (maximum AR = 0.58). At the highest impeller speed 700 rpm, a decrease in AR values was observed at low L/S ratios (minimum 0.42 at L/S = 0.84) following with the AR increase to approximately 0.50-0.51 if the L/S ratio increased up to 1.05. Then, a further slight increasing trend of AR values at higher L/S was noted. At an impeller speed of 500 rpm, the most consistent data were detected; all granules had the AR values of at least 0.51 regardless of the L/S ratio used. The higher AR values indicate better flow properties [65].

Evaluation of Aspect Ratio
Similarly to sphericity, the Aspect ratio AR parameter showed higher values with an increasing L/S ratio. The dependence between AR and L/S ratios is shown in Figure 7a. Out of all experimental runs, the granules formed at an impeller speed 300 rpm reached the highest values of AR as well as the highest data variability. The AR firstly decreased (minimum AR = 0.4 at L/S = 0.92), then it rose sharply with an increasing content of the liquid used to form the granules. The highest values were noted at L/S > 1.1 (maximum AR = 0.58). At the highest impeller speed 700 rpm, a decrease in AR values was observed at low L/S ratios (minimum 0.42 at L/S = 0.84) following with the AR increase to approximately 0.50-0.51 if the L/S ratio increased up to 1.05. Then, a further slight increasing trend of AR values at higher L/S was noted. At an impeller speed of 500 rpm, the most consistent data were detected; all granules had the AR values of at least 0.51 regardless of the L/S ratio used. The higher AR values indicate better flow properties [65]. Again, an AR distribution box chart for each impeller speed can be seen in Figure 7b and in numerical form in Table 2. The graph clearly documents that the narrowest AR distribution was noted at 500 rpm impeller speed. The mean value of AR was 0.507 ± 0.054 (300 rpm), 0.531 ± 0.009 (500 rpm) and 0.505 ± 0.026 at 700 rpm. At 500 rpm, the median value (0.53) of the AR distribution was almost identical to the mean value. The median value of 0.515 at 300 rpm was almost identical to that of 0.51 at 700 rpm.
Based on the sphericity and AR values, it can be concluded that most of the granules prepared in all experimental runs had an elongated rounded shape. Again, an AR distribution box chart for each impeller speed can be seen in Figure 7b and in numerical form in Table 2. The graph clearly documents that the narrowest AR distribution was noted at 500 rpm impeller speed. The mean value of AR was 0.507 ± 0.054 (300 rpm), 0.531 ± 0.009 (500 rpm) and 0.505 ± 0.026 at 700 rpm. At 500 rpm, the median value (0.53) of the AR distribution was almost identical to the mean value. The median value of 0.515 at 300 rpm was almost identical to that of 0.51 at 700 rpm.
Based on the sphericity and AR values, it can be concluded that most of the granules prepared in all experimental runs had an elongated rounded shape.

Evaluation of Roundness
Roundness r represents another granulometric parameter. When analyzing the experimental data, a similar trend to that described above for sphericity was found in relationship to the L/S ratio and impeller speed. The linear relationship between mean r and mean ϕ illustrated in Figure 8a for all experimental data can be described by a linear regression r = 1.187 × ϕ − 0.576 with the coefficient of determination R 2 = 0.909 (p < 0.005).
Again, an AR distribution box chart for each impeller speed can be seen in Figure 7b and in numerical form in Table 2. The graph clearly documents that the narrowest AR distribution was noted at 500 rpm impeller speed. The mean value of AR was 0.507 ± 0.054 (300 rpm), 0.531 ± 0.009 (500 rpm) and 0.505 ± 0.026 at 700 rpm. At 500 rpm, the median value (0.53) of the AR distribution was almost identical to the mean value. The median value of 0.515 at 300 rpm was almost identical to that of 0.51 at 700 rpm.
Based on the sphericity and AR values, it can be concluded that most of the granules prepared in all experimental runs had an elongated rounded shape.

Evaluation of Roundness
Roundness r represents another granulometric parameter. When analyzing the experimental data, a similar trend to that described above for sphericity was found in relationship to the L/S ratio and impeller speed. The linear relationship between mean r and mean φ illustrated in Figure 8a for all experimental data can be described by a linear regression r = 1.187× φ − 0.576 with the coefficient of determination R 2 = 0.909 (p < 0.005). In a box chart of roundness distribution at each impeller speed (Figure 8b), the narrowest r distribution with the rounded granules at 500 rpm is demonstrated. Similarly to In a box chart of roundness distribution at each impeller speed (Figure 8b), the narrowest r distribution with the rounded granules at 500 rpm is demonstrated. Similarly to the evaluation of ϕ and AR, the highest values of r = 0.57 to 0.58 (L/S = 1.12 to 1.17) as well as the widest distribution were obtained at 300 rpm impeller speed; low values of r were found up to L/S = 1. Finally, the lowest r values were noted for four granules batches formed at lower L/S values at 700 rpm.
According to the classification by Maroof et al. [64], values r = 0.35 to 0.49 represent sub-rounded particles, r = 0.49 to 0.7 belong to rounded particles, and r = 0.7 to 1.0 belong to well-rounded particles. The median values of r = 0.495 (300 rpm), 0.53 (500 rpm), and 0.51 (700 rpm) were registered in this work. Mean values from box chart for each impeller speed were 0.471 ± 0.072 (300 rpm), 0.522 ± 0.023 (500 rpm), and 0.492 ± 0.046 (700 rpm). Based on the abovementioned classification, most prepared granules can be characterized as rounded particles. In a numerical form, data are also summarized in Table 2.

Evaluation of Roughness
The roughness of the granules was described by concavity c. A graphical dependence of mean c on the L/S ratio in Figure 9a shows that the roughest granules were formed at the lowest impeller speed of 300 rpm. In a range of L/S ratio 0.9-1.0, the values of c were even larger than 0.07. In opposite, the most homogeneous data were noted at 500 rpm again. At higher impeller speeds, there was first a rising roughness trend at L/S = 0.7 to 0.8, which then descended at L/S = 0.8-0.95. Starting at an L/S ratio of approximately 0.95, a linear increase in c value in regard to the increase in L/S value was visible at both 500 rpm and 700 rpm. Lower values of roughness have a positive effect on the flow properties of powder materials and granules [66]. Based on the results, the smoothest granules were formed at higher impeller speeds in the L/S range of 0.9 to 1.05.
Pharmaceutics 2021, 13, x. https://doi.org/10.3390/xxxxx www.mdpi.com/journal/pharmaceutics The roughness of the granules was described by concavity c. A graphical dependence of mean c on the L/S ratio in Figure 9a shows that the roughest granules were formed at the lowest impeller speed of 300 rpm. In a range of L/S ratio 0.9-1.0, the values of c were even larger than 0.07. In opposite, the most homogeneous data were noted at 500 rpm again. At higher impeller speeds, there was first a rising roughness trend at L/S = 0.7 to 0.8, which then descended at L/S = 0.8-0.95. Starting at an L/S ratio of approximately 0.95, a linear increase in c value in regard to the increase in L/S value was visible at both 500 rpm and 700 rpm. Lower values of roughness have a positive effect on the flow properties of powder materials and granules [66]. Based on the results, the smoothest granules were formed at higher impeller speeds in the L/S range of 0.9 to 1.05. Figure 9. Evaluation of mean concavity for impeller speeds 300, 500, and 700 rpm: (a) effect of the L/S ratio on mean concavity, (b) box charts of mean concavity for different impeller speeds.
A summary of the mean concavity for all experiments in a visual form of box charts of c for all impeller speeds is in Figure 9b and in numerical form in Table 2. At an impeller speed of 300 rpm, the highest mean concavity was observed (0.042 ± 0.013; median = 0.039). The mean concavity values as well as the median values were comparable (0.029 and 0.028, respectively) at an impeller speed of 500 rpm and 700 rpm. However, the graph confirms the narrowest distribution of c in the investigated range of the L/S ratio at an impeller speed of 500 rpm. A summary of the mean concavity for all experiments in a visual form of box charts of c for all impeller speeds is in Figure 9b and in numerical form in Table 2. At an impeller speed of 300 rpm, the highest mean concavity was observed (0.042 ± 0.013; median = 0.039). The mean concavity values as well as the median values were comparable (0.029 and 0.028, respectively) at an impeller speed of 500 rpm and 700 rpm. However, the graph confirms the narrowest distribution of c in the investigated range of the L/S ratio at an impeller speed of 500 rpm.

Effect of Granule Size on Shape Factors
As discussed above, granulometric parameters of the produced granules were influenced by the L/S ratio and the impeller speed used. Apart from these process parameters, the actual values of shape factors depended also on the actual size of the individual granules in the batch. The graphical dependence of investigated shape factors on the area equivalent diameter D a at the L/S ratio = 1.0 is summarized in Figure 10.

Effect of Granule Size on Shape Factors
As discussed above, granulometric parameters of the produced granules were influenced by the L/S ratio and the impeller speed used. Apart from these process parameters, the actual values of shape factors depended also on the actual size of the individual granules in the batch. The graphical dependence of investigated shape factors on the area equivalent diameter Da at the L/S ratio = 1.0 is summarized in Figure 10.
The graph shows that sphericity φ reaches the smallest value for granules up to 0.1 mm. The highest values of φ = 0.94 were detected for granules in the range of 0.3-0.5 mm at 500 and 700 rpm. When the area equivalent diameter Da increased further, a slight decrease in the values of φ regardless of the impeller speed was noted. The lowest values of aspect ratio AR were also detected for the smallest granules, but they increased quickly with the increase in the Da up to 0.3 mm at all impeller speeds used. Then at the impeller speed of 300 rpm, AR remained near to 0.5 up to approximately 2.5 mm granules and decreased with larger granules. At impeller speeds of 500 and 700 rpm, slightly higher invariable values of AR (around 0.55) were noted, up to approximately 1.0 mm granules, which then increased with the area equivalent diameter Da. The highest AR values were reached at the 500-rpm impeller speed. As clearly illustrated in Figure 10, roundness r is narrowly associated with the AR showing the same trend. The exclusion is observable only with the granules larger than 2 mm at higher impeller speeds.
Even though the higher values of concavity c were particularly registered for the smallest granules at 300 rpm, the low oscillating values were noted later up to the granule size 1.5 mm regardless of impeller speed. Then, they increased with the further increase in Da, achieving the highest value c = 0.1 at granule size 2.0 mm at 500 rpm. The raw particles of Avicel PH101 were analyzed by electron microscopy using the SED method at 500× magnification; the rod-shaped particles are shown in Figure 11a. In order to illustrate granulometric parameters of the prepared granules, those formed at L/S ratio = 1.0 and an impeller speed of 700 rpm are chosen. Figure 11b shows the smooth  The graph shows that sphericity ϕ reaches the smallest value for granules up to 0.1 mm.
The highest values of ϕ = 0.94 were detected for granules in the range of 0.3-0.5 mm at 500 and 700 rpm. When the area equivalent diameter D a increased further, a slight decrease in the values of ϕ regardless of the impeller speed was noted. The lowest values of aspect ratio AR were also detected for the smallest granules, but they increased quickly with the increase in the D a up to 0.3 mm at all impeller speeds used. Then at the impeller speed of 300 rpm, AR remained near to 0.5 up to approximately 2.5 mm granules and decreased with larger granules. At impeller speeds of 500 and 700 rpm, slightly higher invariable values of AR (around 0.55) were noted, up to approximately 1.0 mm granules, which then increased with the area equivalent diameter D a . The highest AR values were reached at the 500-rpm impeller speed. As clearly illustrated in Figure 10, roundness r is narrowly associated with the AR showing the same trend. The exclusion is observable only with the granules larger than 2 mm at higher impeller speeds.
Even though the higher values of concavity c were particularly registered for the smallest granules at 300 rpm, the low oscillating values were noted later up to the granule size 1.5 mm regardless of impeller speed. Then, they increased with the further increase in D a , achieving the highest value c = 0.1 at granule size 2.0 mm at 500 rpm.
The raw particles of Avicel PH101 were analyzed by electron microscopy using the SED method at 500× magnification; the rod-shaped particles are shown in Figure 11a. In order to illustrate granulometric parameters of the prepared granules, those formed at L/S ratio = 1.0 and an impeller speed of 700 rpm are chosen. Figure 11b shows the smooth surface, rounded spherically shaped (ϕ = 0.94) granules. It can be seen that the characteristic dimensions of the granules are not the same in all directions, which confirms the lower AR values. A more detailed view of the granule from this batch can be found in Figure 11c, where asperities (c = 0.022) on the surface at 250× magnification are visible. In detail, the raw rod-shaped particles of Avicel PH101 can be recognized on the surface of the granules in Figure 11d. A similar morphology of granules can be found in the work of Mahdi et al. [67]. Shi et al. [68] proved that the final shape of the granules can also be influenced by the process time of wet massing.
Pharmaceutics 2021, 13, x FOR PEER REVIEW 15 of 23 the granules in Figure 11d. A similar morphology of granules can be found in the work of Mahdi et al. [67]. Shi et al. [68] proved that the final shape of the granules can also be influenced by the process time of wet massing.

Evaluation of Granules Compressibility
Particles with lower compressibility are generally advantageous for the tableting process due to more uniform filling of the die and less frequent troubles during transport [69].

Evaluation of Granules Compressibility
Particles with lower compressibility are generally advantageous for the tableting process due to more uniform filling of the die and less frequent troubles during transport [69]. The compressibility value (CPS) obtained from the FT4 rheometer for Avicel powder was high (15.69 ± 0.48%), showing its poor flow properties. Contrary, all granules produced within the wide range of L/S ratio at three different impeller speeds showed lower values of CPS.
The effect of L/S ratio on the compressibility of granules manufactured at 300 rpm impeller speed is present in Figure 12. As discussed above, these granules were the biggest, having the lowest sphericity and high concavity. A decrease in the compressibility value with an increase in L/S ratio was observed; with a higher liquid content, the formed granules had the higher density, which was reflected in the CPS value. CPS could be analyzed up to the L/S ratio 1.07 when the value of CPS = 3.33%. For granules formed at the higher L/S ratio, the device showed an overload at 15 kPa and automatically terminated the measurement.

Application of Surface Response Methodology to Predict Granule Size and Process Yield
A general polynomial equation Equation (7) was used. Based on experimental data, models for determining dmean and process yield (dependent variables), were created using Design Expert 13 software depending on the impeller speed (IS) and the L/S ratio (independent variables). The relationship was described by equations Equations (8) and (9).
A contour plot model for prediction of dmean in relationship to the L/S ratio and the impeller speed is shown in Figure 13. A significant area with dmean = 0.8 mm (blue), which occupies more than half of the graph was detected. The increase in L/S ratio and the associated increase in the size of the granules formed is shown by the change in color. At lower impeller speeds, an orange to red area shows the largest formed granules depending on The relationship between CPS (%) and L/S ratio for 300 rpm impeller speed was described by linear regression of the experimental data: CPS = 25.319 − 20.954 × L/S, with the coefficient of determination R 2 = 0.948. This clearly demonstrates that CPS value can be efficiently predicted by L/S ratio.

Application of Surface Response Methodology to Predict Granule Size and Process Yield
A general polynomial equation Equation (7) was used. Based on experimental data, models for determining d mean and process yield (dependent variables), were created using Design Expert 13 software depending on the impeller speed (IS) and the L/S ratio (independent variables). The relationship was described by equations Equations (8) and (9). A contour plot model for prediction of d mean in relationship to the L/S ratio and the impeller speed is shown in Figure 13. A significant area with d mean = 0.8 mm (blue), which occupies more than half of the graph was detected. The increase in L/S ratio and the associated increase in the size of the granules formed is shown by the change in color. At lower impeller speeds, an orange to red area shows the largest formed granules depending on the L/S ratio. Based on the graph, the process parameters can be optimally set in order to obtain granules of a defined size. Using this, the prediction of d mean value with a standard deviation of 0.077 mm is possible. Similarly, the experimental data was processed in order to predict the process yield; the model is shown in Figure 14. In this graph, the transition between the individual areas of the process yield is significantly smoother. A significant part of the graph is the area with a yield of more than 80% shown in orange to red color. As a result, the process parameters of the impeller speed and L/S ratio can be optimally set in order to obtain a defined process yield. Similarly, the experimental data was processed in order to predict the process yield; the model is shown in Figure 14. In this graph, the transition between the individual areas of the process yield is significantly smoother. A significant part of the graph is the area with a yield of more than 80% shown in orange to red color. As a result, the process parameters of the impeller speed and L/S ratio can be optimally set in order to obtain a defined process yield.
The parameters of the analysis of variance (ANOVA) for both models can be found in Tables 3 and 4. Values of p < 0.05 indicate that the parameters evaluated are statistically significant (>95% probability that the hypothesis is true). The highest p value of 0.0169 in d mean model was detected with the independent variable IS 2 , which indicates its lowest significance. The model determination coefficient was R 2 = 0.9436 with p < 0.0001.  The parameters of the analysis of variance (ANOVA) for both models can be found in Tables 3 and 4. Values of p < 0.05 indicate that the parameters evaluated are statistically significant (>95% probability that the hypothesis is true). The highest p value of 0.0169 in dmean model was detected with the independent variable IS 2 , which indicates its lowest significance. The model determination coefficient was R 2 = 0.9436 with p < 0.0001. In the yield model, the highest p value of 0.1596 was detected with the independent variable L/S × IS. Similarly to dmean model, the value of 0.0169 was noted for IS 2 . The determination coefficient for yield model was slightly lower R 2 = 0.8905 with p < 0.0001.  In the yield model, the highest p value of 0.1596 was detected with the independent variable L/S × IS. Similarly to d mean model, the value of 0.0169 was noted for IS 2 . The determination coefficient for yield model was slightly lower R 2 = 0.8905 with p < 0.0001.
The comparison between the experimental and predicted data for both models can be found in Figure 15a,b. The better prediction efficiency was obtained for mean granule size. The comparison between the experimental and predicted data for both models can be found in Figure 15a,b. The better prediction efficiency was obtained for mean granule size. Figure 15. Relationships between experimental and predicted data: (a) model for prediction of mean granule size, (b) model for prediction of process yield.

Conclusions
The morphology of the granules based on microcrystalline cellulose and the yield of the process were analyzed in this systematic study with respect to the process parameters of high shear wet granulation. The granulometric parameters of the formed granules were analyzed by the dynamic image analysis by PartAn 3D as well as by the electron microscopy. The amount of added granulation liquid expressed by the liquid to solid ratio (L/S) in a range of 0.7 to 1.27 had a significant effect on all the monitored parameters: sphericity, roundness, aspect ratio, and concavity. Three different impeller speeds, 300 rpm, 500 rpm, and 700 rpm, were used. For the comparison reason, the same time of wet massing was employed.
Even a small change in the liquid content caused a significant change in the shape of the granules. At higher impeller speeds and with a L/S ratio ≈1, the most spherical granules with the smoothest surface were formed. In opposite, the higher dynamic stress at the highest impeller speed caused mechanical breakage and attrition, which led to the formation of granules of the smallest particle size. In terms of process yield and particle shape, the best granules were formed at an impeller speed of 500 rpm. The highest process yields were achieved at higher L/S ratio values and higher impeller speeds. Moreover, the values of the granulometric factors observed experimentally were influenced also by the size (area equivalent diameter) of the individual granules in the batch. Additionally, the linear relationship with the coefficient of determination R 2 = 0.948 was detected between the amount of added granulation liquid and the compressibility of the granules (FT4 rheometer), allowing the efficient prediction of CPS by L/S ratio. By using response surface

Conclusions
The morphology of the granules based on microcrystalline cellulose and the yield of the process were analyzed in this systematic study with respect to the process parameters of high shear wet granulation. The granulometric parameters of the formed granules were analyzed by the dynamic image analysis by PartAn 3D as well as by the electron microscopy. The amount of added granulation liquid expressed by the liquid to solid ratio (L/S) in a range of 0.7 to 1.27 had a significant effect on all the monitored parameters: sphericity, roundness, aspect ratio, and concavity. Three different impeller speeds, 300 rpm, 500 rpm, and 700 rpm, were used. For the comparison reason, the same time of wet massing was employed.
Even a small change in the liquid content caused a significant change in the shape of the granules. At higher impeller speeds and with a L/S ratio ≈1, the most spherical granules with the smoothest surface were formed. In opposite, the higher dynamic stress at the highest impeller speed caused mechanical breakage and attrition, which led to the formation of granules of the smallest particle size. In terms of process yield and particle shape, the best granules were formed at an impeller speed of 500 rpm. The highest process yields were achieved at higher L/S ratio values and higher impeller speeds. Moreover, the values of the granulometric factors observed experimentally were influenced also by the size (area equivalent diameter) of the individual granules in the batch. Additionally, the linear relationship with the coefficient of determination R 2 = 0.948 was detected between the amount of added granulation liquid and the compressibility of the granules (FT4 rheometer), allowing the efficient prediction of CPS by L/S ratio. By using response surface methodology, two separate models predicting either the size of granules or the process yield were obtained based on experimental data from more than 90 experiments. According to the amount of liquid added and the impeller speed, both proposed models allow the determination of the granule size or process yield with a high level of accuracy expressed by the coefficient of determination values R 2 = 0.9436 and R 2 = 0.8905, respectively.

Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.

Data Availability Statement:
The data presented in this study are available on request from the corresponding author. The data are not publicly available due to extensive quantity of values.