Improved process understanding through continuously gathered GSD information.
The average D50 values, obtained from the in-line SFV measurements during the last minute of granulation, were used as DOE
responses. The experimental results revealed three significant DOE variables. HPMC concentration, inlet air temperature during
drying, and the interaction between these factors had a positive effect on the D50 values. Because GSD information was obtained
every 10 s through in-line SFV measurements, the authors took a closer look at the individual granule-size profiles of several
DOE batches to better understand the significance of the studied DOE factors.
Influence of the HPMC concentration.
The distinctive particle-size trajectory during fluid-bed granulation consisting of three phases can be distinguished clearly
in Figure 3. Between the first captured granule-size data and time point 1, the particle size remains constant, which corresponds
to the mixing phase. Between time points 1 and 2, the granule size increases because of the agglomeration of powder particles.
The drying period is depicted after time point 2. The graph reveals that the positive effect of the HPMC concentration on
granule size was caused by the following two factors:
Figure 3: D50 profile of batches 13 and 14 (1%/3% hydroxypropyl methylcellulose, 0,2% Tween 20, 50 °C spraying T, 70 °C drying
T) at (1) the beginning of the spraying phase, (2) the end of the spraying phase, and (3) the end of the drying period.
- Large amounts of HPMC resulted in large particles throughout the spraying period. At time point 2 (see Figure 3), Batch 13
(1% HPMC) had an average granule size of 200 μm, while Batch 14 (3% HPMC) displayed an average granule size of 415 μm.
- Large amounts of HPMC created less friable granules, which resulted in less fines during the drying period. Between time points
2 and 3 (see Figure 3), Batch 13 (1% HPMC) showed approximately a 60-μm average decrease in granule size, while the granule
size of Batch 14 (3% HPMC) decreased by only 25 μm.
Influence of the inlet air temperature during the drying period.
DOE analysis showed that the drying inlet air temperature was of less importance (statistical significance p < 0.05) to the GSD than the HPMC concentration (statistical significance p < 0.001). An explanation can be found in Figure 4. Figure 4a displays the in-line measured D50 profiles of Batches 6 (50
°C drying temperature) and 14 (70 °C drying temperature). The other settings were the same for both batches. These batches
showed a similar granule-size trajectory throughout the spraying phase. During the subsequent drying period, Batch 6 yielded
smaller end-sized granules because of higher levels of attrition during a longer drying phase.
Figure 4: D50 profile of (a) batches 6 and 14 (3% hydroxypropyl methylcellulose [HPMC], 0.2% Tween 20, 50 °C spraying T, 50
°C, and 70 °C drying T) and (b) batches 1 and 9 (1% HPMC, 0.2% Tween 20, 40 °C spraying T, 50 °C, and 70 °C drying T).
Figure 4b displays the granule-size data of two other batches with different drying conditions. Batch 1 was dried at 50 °C,
and Batch 9 at 70 °C. Both batches had identical initial process conditions, but a different granule growth profile occurred
during the spraying period. Although the difference in drying temperature between the two batches caused a clear difference
in particle-size evolution during the drying period (i.e., a larger decrease in particle size at the lower temperature of
Batch 1), the end granule sizes used for the DOE analysis were similar. Hence, for these batches, a similar response value
was used in the DOE analysis, despite the fact that the batches showed different granulation trajectories. This result might
explain why the drying temperature was of limited significance according to the DOE. Only through the information obtained
continuously in-line from the SFV probe was it possible to develop this in-depth understanding.
Influence of the inlet air temperature during the spraying period.
Figure 5 shows the D50 data obtained in-line during the spraying period of two batches manufactured at different spraying
temperatures. During the agglomeration phase of Batch 5 (spraying temperature of 50 °C), a higher fluctuation in granule size
was observed, compared with Batch 1 (spraying temperature of 40 °C). This result was caused by the continuous entrapment of
small particles in and subsequent discharge from the filter bags. At the higher temperature, small particles were present
for a longer period of time than they were in Batch 1 because the faster evaporation of binder liquid led to slower agglomeration.
However, at the end of the spraying periods, a similar particle size was observed at both temperatures. Although the detailed
profile of granule growth through in-line granule-size monitoring showed a difference in agglomeration kinetics, the different
spraying temperature did not create a difference in granule size at the end of the spraying period.
Figure 5: D50 profile during the spraying period of batches 1 and 5 (1% hydroxypropyl methylcellulose, 0.2% Tween 20, 40 °C/50
°C spraying T, 50 °C drying T).
Estimation of granule tapped density from in-line SFV measurements.
The tapped density of the end product is, like GSD, also important to further processing. Univariate, multivariate, and multiway
approaches were considered to relate the in-line determined GSD to the tapped density of the 19 DOE batches (tapped density
as dependent variable: 19 × 1 Y-vector). A univariate linear model was built using the D50 SFV values of the end granules of the 19 DOE batches as independent
variables (19 × 1 X-vector). A multivariate partial least squares (PLS) model was built using the D01, D10, D25, D50, D63, D75, D90, and D99
SFV values of the end granules of the 19 DOE experiments as independent variables (19 × 8 X-matrix). A multiway N-way PLS (N-PLS) model was built using the D01, D10, D25, D50, D63, D75, D90, and D99 SFV values of the 19 DOE experiments in a function
of complete batch process time as independent variables (three-way X-matrix).
Table II compares the goodness of fit (R
2) and root mean square error of estimation (RMSEE) of the different models. The multivariate PLS tapped-density model had
the highest R
2 value and the lowest RMSEE value. The low RMSEE of the model suggests the ability of the in-line SFV data to predict the
off-line measured tapped density. Nevertheless, the model performance should be tested independently because the RMSEE value
relates to the error within the calibration set and may overestimate the actual model performance.
Table II: R2 and RMSEE of univariate, multivariate, and multiway tapped-density models.
The results of this study show the feasibility of SFV for the real-time monitoring of GSD during fluid-bed granulation. The
technique was sensitive to PS changes during the performed granulations and did not underestimate the granule size because
of size segregation. Probe fouling did not occur. The continuously obtained GS information enabled a better understanding
of the significance of the studied DOE factors on granulation. This understanding was not possible based on the off-line LD
data of the end product. Finally, a multivariate PLS model was built to estimate end-product tapped density using continuously
obtained GSD during granulation, which may improve batch-release time.
Next to real-time process monitoring of critical quality attributes, which helps to improve process understanding, another
goal of PAT is to use the real-time critical information to steer or adjust the process toward its desired state based on
feedback and feed-forward loops. Granule particle-size distribution and density are important granule quality attributes.
In a future study, the authors aim to use in-line collected granule information gathered during the spraying phase to adjust
process variables from the consecutive drying phase, hence guiding the drying process to the desired granule properties.
The Institute for the Promotion of Innovation through Science and Technology in Flanders (IWT-Vlaanderen) provided financial
support for this study.
A. Burggraeve* is a doctoral student, J.P. Remon and C. Vervaet are professors of pharmaceutical technology, and T. De Beer is a professor of process analytical technology, all at Ghent University, Harelbekestraat 72, B-9000 Ghent, Belgium, tel.
+32 9 264 83 55, fax +32 9 222 82 36, email@example.com
. T. Van Den Kerkhof and M. Hellings are senior scientists, both at Johnson & Johnson.
*To whom all correspondence should be addressed.
Submitted: Feb. 22, 2011. Accepted: May 13, 2011.