Real-time monitoring of the drying of extruded granules in a fluid-bed dryer using audible acoustic emission chemometrics

Abstract

The purpose of this study was an attempt to adapt the audible acoustic emission (AAE) sound measurement method for the on-line monitoring of the fluid-bed drying progress of pharmaceutical granules. The granules were prepared by extrusion-granulation based on a formulation of 6.7 : 2 : 1 lactose–starch–crystalline cellulose. After the granulation process, the drying process was performed in a fluid-bed dryer at 27 or 42 °C, and AAE sound was measured using a digital voice recorder. The recorded signals were transformed into frequency spectra by using the fast Fourier transformation function. Samples were collected every 60 seconds to determine the moisture content of the granules. The calibration models to predict the moisture content of the granules were constructed based on AAE frequency spectra by using the partial least squares regression method after area normalized function. In order to test the robustness of the calibration model obtained under different dry operating conditions (air temperature) with various acoustic environments (noise), the moisture content of the granules was predicted based on AAE frequency spectra containing noise. The external validation results suggested that the calibration model could be applied to any data set. The regression vector indicated that the sound in the low frequency range might have been caused by the contact of the granules upon over-hydration at the initial stage of the drying process. In contrast, the sound at high frequencies might have been caused by friction of the dried granules later in the drying process.

---

Introduction

Since granules have several advantages as a pharmaceutical dosage form compared with powder, such as, better flowability, wettability, mixing uniformity, easy control of their dust, and good compressibility, granules are prepared to agglomerate powdered materials into larger sizes by using various kinds of granulator. The properties of the final granular products are affected by operating conditions during the drying process and the kind of granulator. The drying process of agglomerates of powdered materials is, therefore, a crucial operation to make high-quality granular dosage forms in the pharmaceutical industry. These are important properties in order to achieve fast, gentle and uniform particle drying. Owing to the high drying rate, associated high quality, and economic benefits, fluid-bed drying has been proposed as the method of choice over to other drying techniques.

In recent years, in order to improve the quality of pharmaceutical products, regulatory authorities such as the US Food and Drug Administration and the International Conference on Harmonization have proposed Process Analytical Technology (PAT) initiative forms based on the pharmaceutical Good Manufacturing Practice rules for the 21st century.^1,2 They requested real-time control of drug product quality and the application of Quality by Design principles to monitor and control manufacturing processes using PAT tools.^3–5

To control the drying process, conventional chemical (Karl Fischer titration) and physical (heat balance) methods are used routinely for the determination of water content in pharmaceutical products. However, they are hard to perform as real-time monitoring of the drying because both methods are time-consuming and costly. Recently, near-infrared (NIR) spectroscopy has been introduced in the pharmaceutical industry because it is nondestructive and requires no or minimal sample preparation and provides immediate delivery of results. In particular, in combination with chemometrics, NIR spectroscopy provides an ideal method of extracting quantitative information from multicomponent chemical samples in the pharmaceutical field. The most widely used chemometric methods include multiple linear regression, principal component analysis/principal component regression, and partial least squares (PLS) regression.⁶ For example, chemometric NIR spectroscopy has been used to determine active pharmaceutical ingredients off-line and on-line, tablet excipient content,^7–10 drug stability,¹¹ particle size of powders,¹² tablet mechanical strength,^13,14 and dissolution rate.^14–17 However, it is costly to establish many NIR spectroscopy instruments on production lines in factories.

On the other hand, mechanical sound during chemical and/or pharmaceutical processes is useful information for evaluating the degree of product completion for veteran technicians in industry. On the basis of veteran technical experts' knowledge and skill for controlling manufacturing processes, acoustic emission (AE) technology was developed for process monitoring. AE monitoring has the advantage of being a real-time, noninvasive technique, the same as NIR spectroscopy. AE and NIR methods capture the mechanical and optical signatures of events taking place during processing, respectively.³ However, the NIR method requires installation of a microfiber probe and a line into the sample powder for measuring, but the AE method does not need a direct line-of-sight to the material of interest and therefore requires no alteration of manufacturing equipment.

To take advantage of the ease of installation of measurement devices and good measurement accuracy, the AE analysis method was developed in chemical and pharmaceutical fields as a PAT pilot study, and there are several reports on AE application studies.^19–29 The research can be divided into two types; with ultrasonic (greater than 20 000 Hz) and audible (approximately 20–20 000 Hz) AE sensors.

Ultrasonic AE sensors can be easily attached to the container wall of pharmaceutical machines to detect sounds. For example, the effects of operation conditions on the acoustic signal during tablet compression,¹⁸ the roller compaction^19,20 and table coating processes,²¹ mixing process,²² particle measurement,^23,24 and agitating granulation²⁵ have been studied.

In contrast, audible AEs differ from ultrasonic AEs in terms of the setting of measurement devices, because they propagate through air with minimal attenuation^26–29 and therefore equipment contact is not required for detection. Microphones suspended at the top of granulator air exhausts were also shown to be sensitive to granulation-based pharmaceutical formulation for identifying the granulation end-point.^26,28

On the other hand, one of the emerging on-line noninvasive PAT approaches for process characterization is acoustic chemometrics, since interpretations of these complex AE data are most effectively performed through the use of modern chemometric methods.^30,31 On-line process monitoring of the fluid-bed drying process was investigated by using acoustic chemometrics.^32,33 Ihunegbo et al.³⁴ investigated the feasibility of quantitative on-line monitoring of the drying progress and end-point determination of pharmaceuticals dried in a heated fluid-bed based on audible and ultrasonic AEs by chemometrics. They concluded that the final prediction results were satisfactory for monitoring of the drying progress and end-point determination by the PLS method. However, they did not report scientific evidence of the calibration models to predict individual pharmaceutical properties of the final products.

The present study is an attempt to adapt the audible acoustic emission sound measurement method for the on-line monitoring of the fluid-bed drying progress of pharmaceutical granules, and to clarify the scientific background of the calibration model to predict moisture content in granules by audible acoustic emission (AAE) frequency spectrum/chemometric analysis.

Materials and methods

Materials

Lactose monohydrate (Pharmatose® 200M) from DMV (Veghel, The Netherlands), potato starch from Kosakai Pharmaceutical Co., Ltd (Japan), microcrystalline cellulose (CEOLUS® PH-102) from Asahi Kasei Co., Ltd (Tokyo, Japan), and hydroxypropyl cellulose (HPC-L®) from Nippon Soda Co., Ltd (Tokyo, Japan) were used. The lactose served as a filler, potato starch as a disintegrating agent, microcrystalline cellulose as a segregation preventive agent, and HPC-L as a binding agent. Granules comprising mainly lactose, potato starch, microcrystalline cellulose, and HPC-L were prepared. This formulation was based on a standard 6.7 : 2 : 1 lactose–starch–microcrystalline cellulose mixture.

Preparation of granules

A total of 200.0 g of lactose, 60.0 g of potato starch, 30.0 g of microcrystalline cellulose, and 10.0 g of HPC-L, were mixed in a polyethylene bag for 3 minutes by hand. Purified water was added and the mixed was then kneaded in a mortar and pestle. Granules, 1 mm in diameter, were prepared by extrusion granulation (KAR-130, Tsutsui Scientific Instruments Co., Ltd, Tokyo, Japan). After the granulation process, drying was performed in a fluid-bed dryer with a chamber made of glass (SP-15, 160 mm in diameter and 6.0 L in volume, Okada Seiko Co., Ltd, Tokyo, Japan), as shown in Fig. 1. A sampling port was located 3 cm from the bottom of a chamber of the dryer, and granular samples was withdrawn using a plastic sampling bar with a diameter of 15 mm. Fluid-bed dryer operation conditions were fixed during all processes as follows: warming up time was 15 min at 42 °C and rotor speed was 180 rpm. The granulation experiments were repeated three times in each group. Outlet air temperature was measured using a temperature sensor, and the temperature of outlet air was set at 42 °C for the groups 1, 2 and 8, and at 27 °C for group 3. Group 2 involved drying under conditions (outlet air was set at 42 °C) with noise (Japanese pop music, a vacuum cleaner, or a tableting machine) to test the robustness for the audible acoustic emission calibration model. A portable stereo radio CD player, a vacuum cleaner, and a tableting machine were placed at a distance of 30 cm from the dryer as sources of noise, respectively.

Fig. 1 Fluid-bed drying equipment for audible acoustic emission analysis.

Measurement of moisture content

The samples of approximately 3 g were collected every 60 seconds 16 times using the sampling bar during the drying process, and then the collected samples were weighed accurately by an electronic analytical balance. To determine the moisture content of the granules, drying loss at 70 °C for 24 hours was estimated. To evaluate the variability among batches, the procedures were repeated multiple times and the moisture contents of the granular samples were recorded. All batches were evaluated for the time required to reach a point when there was no change in mass of the samples over time as the drying had finished.

Acoustic signal measurements

The acoustic sensor used was a digital voice recorder (RR-XS350, Panasonic Co., Ltd, Tokyo, Japan). The recorder was placed at a distance of 0.5 cm from the wall of the lowest portion of the chamber of the fluid-bed dryer. Audible acoustic emission signals were recorded as a waveform at a sampling rate of 44.1 kHz during the drying process. The recorded AAE signals were transformed into frequency spectra every 60 seconds by using the fast Fourier transformation function of Audacity® (Audacity 2.0.5, http://audacity.sourceforge.net) as the calibration data. The AAE frequency spectra for the semi-external validation data were transformed from raw signals of the groups 1, 2 and 3 at every 61 seconds, respectively. In contrast, the spectra for the external validation data were transformed from raw signals of the group 8 at every 60 seconds.

The FT-AAE frequency spectra were calculated at intervals of 1 second and window size of 4096 by using Blackman–Harris window transformation in the frequency range between 0.01 and 22 kHz. The frequency spectra were converted from amplitude into sound pressure level L_P according to the following expression.³⁵

where P is the actual sound pressure and P₀ is the reference sound pressure which is 20 μPa in air. The actual sound pressure has a relationship with the electromotive force E described in the following equation,

E = S + 10 log₁₀ P² (2)

where S is sensitivity of the microphone. Substituting eqn (3) to (1), the following equation can be derived,

L_P = E − S + 94 (4)

Partial least squares regression

A chemometric analysis was performed using the partial least squares (PLS) regression method associated with the Pirouette software ver. 4.5 (Infometrix Corporation, Woodinville, U.S.A.). The moisture contents (the dependent variable) of the granules were estimated based on a total of 144 spectra (independent variables) involving groups 1, 2, and 3 by PLS. The PLS calibration models were constructed by cross-validation using the leave-one-out (LOOCV) method. The optimum number of factors was taken to be that leading to a minimum value in the prediction residual error sum of squares (PRESS) versus PLS component graph, the former being defined as:where ŷ_i and y_i correspond to the moisture level of each granular sample predicted by the AE method and the reference method, respectively. The goodness of calibration and prediction was assessed in terms of the root mean square error (RMSE):which was termed RESEC for calibration and RMSEP for prediction.

Results and discussion

Frequency spectra of granules during the drying process

Fig. 2 shows a typical example of an AAE waveform of audible acoustic sound during the drying process of extruded granules in the fluid-bed dryer. The AAE waveforms were transformed into AAE frequency spectra using the Fourier transformation function.

Fig. 2 Waveform of audible acoustic sound during the drying process of extruded granules in the fluid-bed dryer.

Fig. 3(a) shows change of raw frequency spectra of AAE sound during the drying process of extruded granules in the fluid-bed dryer. The sound pressure level below 0.1 kHz significantly decreased with increasing of the time, but that above 1 kHz increased with a lot of noise. It was considered that there were a number of noises in the high frequency range, which made it difficult to analyze the frequency spectrum.

Fig. 3 Change of (a) raw spectra, and (b) normalized spectra of AAE frequency spectra during the drying process of extruded granules in the fluid-bed dryer.

To clarify time-dependent changes in the frequency spectra of granular samples during the drying process, the raw spectra were converted to area normalized frequency spectra. Fig. 3(b) shows change of the area normalized frequency spectra of AAE sound during the drying process of extruded granules in the fluid-bed dryer. In the area normalized frequency spectra, sound pressure level at lower than 1 kHz significantly decreased with increasing time, but that above 1 kHz was almost constant.

Construction of partial least squares model

To predict the moisture content of granules, the calibration models were constructed based on frequency spectra by using the PLS method after area normalized function. Fig. 4(a) shows the correlation between the actual and predicted moisture contents of group 1 (standard drying conditions) obtained by the PLS method, and their chemometric parameters are summarized in Table 1. The relationship between the actual and predicted moisture contents shows a straight line with a slope of 0.992, y-intercept of 0.0595, and correlation coefficient of 0.992. The PRESS and the RMSECV were evaluated to be 31.5 and 2.18 by the leave-one-out method in the PLS method, and the other parameters also supported the assertion that the obtained calibration model involving the first 4 latent variables (LV) could predict the moisture content in the granular samples with sufficient accuracy.

Fig. 4 Relationship between predicted and measured moisture contents of granules based on (a) group 1, (b) group 7 of the PLS model, and (c) external validation result of semi-external group 1–3 and external group 8 data using group 7, PLS model. External frequency spectral data under G1, G2, and G3 conditions were evaluated using G1 and G7, PLS models. ◇: G1 semi-external data, △: G2 semi-external data, and ▽: G3 semi-external data, ○: G8 external data.

Table 1 Chemometric parameters for PLS calibration models based on audible acoustic sound during the drying process to predict moisture content^a

	N	LV	R^2	Slope	Intercept	PRESS	RMSEC	RMSEP	RMSECV
a N, number of experiments; LV, latent variables; R^2, coefficient of determination; PRESS, predicted residual error sum of squares; RMSEC, root mean square error for calibration; RMSEP, root mean square error for prediction; RMSECV, root mean square error for cross-validation; group 1: standard conditions (42 °C), group 2: including noise (vacuum cleaner, Japanese pop music), group 3: low-thermal air (27 °C), group 4: groups 1 + 2, group 5: groups 1 + 3, group 6: groups 2 + 3, groups 7: groups 1 + 2 + 3.
Group 1	3	4	0.992	0.992	0.0595	31.5	0.856	0.810	2.18
Group 2	3	4	0.992	0.992	0.0799	39.0	0.952	0.901	2.43
Group 3	3	4	0.992	0.992	0.116	18.7	0.692	0.651	1.79
Group 4	6	5	0.991	0.991	0.0784	80.4	0.945	0.915	2.21
Group 5	6	5	0.989	0.989	0.125	87.5	1.00	0.970	2.11
Group 6	6	5	0.987	0.987	0.158	101	1.08	1.05	2.23
Group 7	9	5	0.984	0.984	0.168	198	1.22	1.19	2.19

---

Effects of acoustic environment and drying operation conditions on the robustness of the calibration model

In order to test the robustness of the calibration model obtained under dry operating conditions of the various acoustic environments, the moisture contents of the granules were predicted based on acoustic frequency spectra containing noise. Group 2 was dried under experimental conditions involving the following type of noise. The drying experiments for the granules were performed with Japanese pop music were played on a portable radio CD player, a vacuum cleaner, and a tableting machine. As shown in Table 1, the chemometric parameters for group 2 supported the assertion that the calibration model involving the first 4 LVs could be predicted sufficient accurate the moisture content in the granular samples with sufficient accuracy. Those for group 4, involving both group 1 and group 2, also indicated that the model involving the first 5 LVs could provide accurate predictions. The results suggested that the calibration model was not affected by the typical noise in the measurement environment, and the moisture content of the granules could be predicted based on AAE frequency spectra containing noise.

It is well known that the drying process is dependent on change of the outer air temperature, so the effect of outlet air temperature on the drying process was investigated. Group 3 was dried under lower-temperature conditions at 27 °C, and the chemometric parameters indicated that the calibration model involving the first 4 LVs could predict the moisture content with sufficient accuracy, as shown in Table 1. Those for group 5, involving temperature variations (27 and 42 °C), also indicated that the model involving the first 5 LVs could provide accurate predictions.

Finally, the drying process in a fluid-bed dryer is affected by various process operating conditions, such as outer air temperature, humidity, and noise. Therefore, the combined effects of drying temperature variations and environmental noise on the robustness of the calibration model were investigated. Groups 6 and 7 underwent drying conditions involving both temperature variations and environmental noise. Fig. 4(b) shows the relationship between the predicted and actual moisture contents for group 7, with a straight line with a slope of 0.984, y-intercept of 0.168, and correlation coefficient constant of 0.984. The calibration model for group 7 consisted of the first 5 LVs involving 72.6% cumulative variance, and the parameters indicated that the model could predict the moisture content with sufficient accuracy, as shown in Table 1.

Validation of the fitted calibration models based on external validation data

For validation of the PLS calibration models to predict the moisture content of the granules, the other frequency spectra as an external validation data set were applied to the obtained calibration models. Fig. 4(c) and Table 2 show the suitability of each calibration model of the semi-external validation data for groups 1–3 and external validation data for group 8. The semi-external validation data of G1, G2, and G3 were evaluated using all calibration models, and then the best R²s were obtained by using the calibration models that were created using the individual data sampling time at every 61 seconds. However, there was no model that could be applied to the other data sets, except for the model based on group 7. The R² for groups 1, 2, and 3 were 0.985, 0.994, and 0.928 by using the calibration model for group 7, respectively. These results suggest that the calibration model based on group 7 could be applied to any group data set.

Table 2 Validation result of semi-external group 1–3 and external group 8 data using G1 or G7 PLS models^a

Batch	PLS model	R^2	Slope	Intercept
a **-best-fitted, *-second best-fitted, and frequency spectra of the external granular samples obtained under G1, G2, and G3 conditions were evaluated using G1, G2, G3, and G7, PLS models.
Semi-external group 1	Group 1	0.991	0.985	0.479	**
	Group 2	0.980	1.21	0.142
	Group 3	0.913	0.685	4.04
	Group 7	0.985	1.03	−0.906	*
Semi-external group 2	Group 1	0.929	0.768	3.50
	Group 2	0.992	0.987	0.117	*
	Group 3	0.848	0.593	6.50
	Group 7	0.994	0.920	0.630	**
Semi-external group 3	Group 1	0.827	0.844	−0.845
	Group 2	0.931	0.989	−0.129
	Group 3	0.958	0.992	−0.273	**
	Group 7	0.938	0.940	0.988	*
External group 8	Group 1	0.892	0.983	0.800
	Group 2	0.944	1.12	0.345	**
	Group 3	0.820	0.673	4.85
	Group 7	0.928	0.988	0.396	*

---

Scientific background of the PLS calibration model to predict the moisture content in the granules based on the AAE frequency spectrum

PLS regression is effective in the extraction of features and regularity, and modeling of unstable, large, and complex numerical data. However, the disadvantages of PLS regression are the difficulty of interpretation of the factors, and it is also necessary to determine the number of factors to be used. Therefore, in order to provide the scientific evidence of the ability of the PLS models based on AAE frequency spectra to predict the moisture content of the granules, relationships between the loading or regression vector and information on the formulation powder during drying were examined.

Fig. 5 shows the loading vectors for first and second LVs of the calibration model to predict moisture content in the granules. The loading vectors for the first and second LV contained 58.6% and 7.5% of the total variance, respectively. The loading vector for the first LV had the positive broad peaks at 10–100 Hz and positive specific peaks at 484 and 1216 Hz. It had negative peaks at 656, 1442, and 15 000–22 000 Hz. The vector for the second LV had positive broad peaks at 86, 204, and 355 Hz and negative peaks at 667, 947–1141, and 2000–3500 Hz. The result of the first loading vector indicates that the sound with lower frequency than 100 Hz was converted into sound higher than 15 000 Hz. The vector for the second LV was due to mid-frequency range sound transformation, which means that sound of 60–200 Hz was converted into sound of 1000–4000 Hz.

Fig. 5 Loading vectors for group 7 of the PLS model based on normalized frequency spectra of audible acoustic sound during the drying process of extruded granules in the fluid-bed dryer.

Fig. 6 shows the relationship between the scores of first LV and second LV for the calibration model based on group 7 to predict the moisture content in the granules. In the first half of the drying process, the first LV decreased, in the second half, it gradually reached a constant value of −2. In contrast, the second LV increased in the first half and decreased in the second half.

Fig. 6 Score plot of group 7, PLS model, based on normalized frequency spectra of audible acoustic sound during the drying process of extruded granules in the fluid-bed dryer.

Fig. 7 shows the regression vector plot as a weighting function of the calibration model for group 7, involving temperature variations and environmental noise. In the regression vector, positive peaks were observed at a relatively low frequency range at 100 Hz, 200 Hz, 340 Hz, 840 Hz, and 1570 Hz. In contrast, the negative peaks were observed at a higher range at around 2030 Hz and 3440 Hz. These results indicated that the peaks at lower frequency decreased during the drying process, but the peaks at higher frequency increased. The sound in the low frequency range might have been caused by the contact of the granules upon over-hydration at the initial stage of the drying process. In contrast, the sound at high frequencies might have been caused by friction of the dried granules later in the drying process.

Fig. 7 Regression vector for group 7, PLS model, based on normalized frequency spectra of audible acoustic sound during the drying process of extruded granules in the fluid-bed dryer. Solid line is regression vector and gray line is frequency spectra of group 7.

Kinetic evaluation of fluid-bed drying process of the extruded granules

Fig. 8 shows the changes of the moisture contents in the granules for groups 1, 2, and 3 predicted by using the best-fitted PLS model. Predicted moisture content profiles were generally consistent with measured values. It is well known that there are three phases (i.e. pre-heating period, constant drying rate period, and falling drying rate period) during the drying process, as reported previously.³⁶ The pre-heating period is the time required to reach a certain dynamic equilibrium temperature determined by the drying conditions of the initial temperature. The constant drying rate period is a period during which the cooling rate due to evaporation of free water is equal to heating by hot air, and the drying rate is constant. In other words, as long as free water is present on the granule surface, the constant rate drying period continues. The falling drying rate period is the time required to dry the water present inside the granules. As shown by the results for various drying conditions, the drying process of the granules could be separated into pre-heating period, constant drying rate period, and falling drying rate period by the AAE chemometrics. The drying process of the granules could also be divided into two processes; the former process might be due to the sound caused by the collision of wetted granules, and the latter process might be the sound caused by friction of dried granules.

Fig. 8 External validation result of semi-external group 1–3, and external group 8 data using groups 1 and 7, PLS model. Closed circle is measured moisture under G1, G2, and G3 conditions, dashed line is predicted by individual best-fitted PLS model, and solid line is predicted based on G7 PLS model.

Conclusion

The present study demonstrated the usefulness of real-time monitoring using AAE analysis to predict the moisture content of granules and product quality parameters during the fluid-bed drying process in real time. To determine the parameters, a PLS model based on AAE frequency spectra and loss on drying measurements was constructed under different dry operating conditions with various acoustic environments. This technique facilitated the construction of a robust model with no variability from batch to batch. This technique provides for better understanding and control of the drying process in a less expensive manner.

Acknowledgements

This research was supported in part by a Grant for Musashino-Jyoshi Gakuin.

References

1. Process Analytical Technology (PAT) Initiative, U.S. Food and Drug Administration Center for Drug Evaluation and Research Home Page, http://www.fda.gov/AboutFDA/CentersOffices/OfficeofMedicalProductsandTobacco/CDER/ucm088828.htm, accessed August 2009.
2. International Conference on Harmonization of Technical Requirements for Registration of Pharmaceuticals for Human Use, Pharmaceutical Development Q8 (R2), http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryinformation/Guidances/ucm073507.pdf, accessed August 2009.
3. S. Matero, F. V. Den Berg, S. Poutiainen, J. Rantanen and J. Pajander, Towards Better Process Understanding: Chemometrics and Multivariate Measurements in Manufacturing of Solid Dosage Forms, J. Pharm. Sci., 2013, 102(5), 1385–1403.
4. L. X. Yu, Pharmaceutical Quality by Design: Product and Process Development, Understanding, and Control, Pharm. Res., 2008, 25(10), 781–791.
5. T. De Beer, A. Burggraeve, M. Fonteyne, L. Saerens, J. P. Remon and C. Vervaet, Near infrared and Raman spectroscopy for the in-process monitoring of pharmaceutical production processes, Int. J. Pharm., 2011, 417(1–2), 32–47.
6. H. Martens and T. Naes, Multivariate calibration, John Wiley & Sons, New York, 1989.
7. M. Blanco, A. Eustaquio, J. M. González and D. Serrano, Identification and quantitation assays for intact tablets of two related pharmaceutical preparations by reflectance near-infrared spectroscopy: validation of the procedure, J. Pharm. Biomed. Anal., 2000, 22(1), 139–148.
8. L. Alvarenga, D. Ferreira, D. Altekruse, J. C. Menezes and D. Lochmann, Tablet identification using near-infrared spectroscopy (NIRS) for pharmaceutical quality control, J. Pharm. Biomed. Anal., 2008, 48(1), 62–69.
9. C. Bodson, E. Rozet, E. Ziemons, B. Evrard, P. Hubert and L. Delattre, Validation of manufacturing process of Diltiazem HCl tablets by NIR spectrophotometry (NIRS), J. Pharm. Biomed. Anal., 2007, 45(2), 356–361.
10. A. S. Zidan, M. J. Habib and M. A. Khan, Process Analytical Technology: Nondestructive Evaluation of Cyclosporine A and Phospholipid Solid Dispersions by Near Infrared Spectroscopy and Imaging, J. Pharm. Sci., 2008, 97(8), 3388–3399.
11. J. K. Drennen and R. A. Lodder, Nondestructive near-infrared analysis of intact tablets for determination of degradation products, J. Pharm. Sci., 1990, 79(7), 622–627.
12. P. Frake, I. Gill, C. N. Luscombe, D. R. Rudd, J. Waterhouse and U. A. Jayasooriya, Near-infrared mass median particle size determination of lactose monohydrate, evaluating several chemometric approaches, Analyst, 1998, 123(10), 2043–2046.
13. M. Otsuka and I. Yamane, Prediction of Tablet Hardness Based on Near Infrared Spectra of Raw Mixed Powders by Chemometrics, J. Pharm. Sci., 2006, 95(7), 1425–1433.
14. M. Blanco, M. Alcala, J. M. Gonzalez and E. Torras, A Process Analytical Technology Approach Based on Near Infrared Spectroscopy: Tablet Hardness, Content Uniformity, and Dissolution Test Measurements of Intact Tablets, J. Pharm. Sci., 2006, 95(10), 2137–2144.
15. M. Donosso and E. S. Ghaly, Prediction of Drug Dissolution from Tablets Using Near- Infrared Diffuse Reflectance Spectroscopy as a Nondestructive Method, Pharm. Dev. Technol., 2004, 9(3), 247–263.
16. M. P. Freitas, A. Sabadin, L. M. Silva, F. M. Giannotti, D. A. do Couto, E. Tonhi, R. S. Medeiros, G. L. Coco, V. F. T. Russo and J. A. Martins, Prediction of drug dissolution profiles from tablets using NIR diffuse reflectance spectroscopy: A rapid and nondestructive method, J. Pharm. Biomed. Anal., 2005, 39(1–2), 17–21.
17. C. Gendre, M. Boiret, M. Genty, P. Chaminade and L. M. Pean, Real-time predictions of drug release and end point detection of a coating operation by in-line near infrared measurements, Int. J. Pharm., 2011, 421(2), 237–243.
18. M. J. Waring, M. H. Rubinstein and J. R. Howard, Acoustic emission of pharmaceutical materials: the effect of compression speed, ejection, lubrication and tablet weight, Int. J. Pharm., 1987, 40(1–2), 15–22.
19. A. Hakanen and E. Laine, Acoustic Characterization of a Microcrystalline Cellulose Powder During and After Its Compression, Drug Dev. Ind. Pharm., 1995, 21(13), 1573–1582.
20. J. Salonen, K. Salmi, A. Hakanen, E. Laine and K. Linsaari, Monitoring the acoustic activity of a pharmaceutical powder during roller compaction, Int. J. Pharm., 1997, 153(2), 257–261.
21. T. Yoshida, T. Tanino, Y. Sumi and F. Nogami, Coating and polishing methods, and the surface treatment apparatus to do by the methods, JP Pat., Tokkai 2001-252548, P2001-252548A, 2001.
22. P. J. Tilly, S. Porada, C. B. Scruby and S. Lidington, Monitoring of mixing processes using acoustic emission, in Fluid Mixing III, ed. N. Harnby, H. Benkreira, K. J. Carpenter and R. Mann, Institution of Chemical Engineers, Rugby, 1988.
23. M. F. Leach and G. A. Rubin, Size Analysis of Particles of Irregular Shape from their Acoustic Emissions, Powder Technol., 1978, 21(2), 263–267.
24. M. F. Leach, G. A. Rubin and J. C. Williams, Particle Size Distribution Characterization from Acoustic Emissions, Powder Technol., 1978, 19(2), 157–167.
25. M. Whitaker, G. R. Baker, J. Westrup, P. A. Goulding, D. R. Rudd, R. M. Belchamber and M. P. Collins, Application of acoustic emission to the monitoring and end point determination of a high shear granulation process, Int. J. Pharm., 2000, 205(1–2), 79–91.
26. L. Briens, D. Daniher and A. Tallevi, Monitoring high-shear granulation using sound and vibration measurements, Int. J. Pharm., 2007, 331(1), 54–60.
27. H. E. Bass, A. J. Campanella, J. P. Chambers, R. B. Lindsay, Sound absorption, AccessScience@McGraw-Hill, http://www.accessscience.com.proxy2.lib.uwo.ca:2048, retrieved May 2008,  DOI:10.1036/1097-8542.637300.
28. D. Daniher, L. Briens and A. Tallevi, End-point detection in high-shear granulation using sound and vibration signal analysis, Powder Technol., 2008, 181(2), 130–136.
29. L. D. Martin, J. V. Briongos, J. M. Aragon and M. C. Palancar, Can low frequency accelerometry replace pressure measurements for monitoring gas-solid fluidized beds?, Chem. Eng. Sci., 2010, 65(13), 4055–4064.
30. M. Halstensen and K. Esbensen, New developments in acoustic chemometric prediction of particle size distribution – ‘the problem is the solution’, J. Chemom., 2000, 14(5–6), 463–481.
31. E. M. Hansuld, L. Briens, J. A. B. McCann and A. Sayani, Audible acoustics in high-shear wet granulation: Application of frequency filtering, Int. J. Pharm., 2009, 378(1–2), 37–44.
32. Y. Li, J. R. Grace, R. B. Gopaluni, H. Bi, C. J. Lim and N. Ellis, Characterization of gas–solid fluidization: A comparative study of acoustic and pressure signals, Powder Technol., 2011, 214(2), 200–210.
33. K. H. Esbensen and P. Geladi, Principles of Proper Validation: use and abuse of re-sampling for validation, J. Chemom., 2010, 24(3–4), 168–187.
34. F. N. Ihunegbo, C. Ratnayake and M. Halstensen, Acoustic chemometrics for on-line monitoring and end-point determination of fluidised bed drying, Powder Technol., 2013, 247, 69–75.
35. Sound & Vibration Technologies – about the sensor and sound, Onosokki technical report, http://www.onosokki.co.jp/HP-WK/c_support/newreport/sound/soundsensor.pdf, accessed December 2013.
36. Y. Hayashi, T. Sato and M. Otsuka, Real-time monitoring of the drying of extruded-granules in a fluidized bed using near infrared spectroscopy and kinetic evaluation of the drying process, J. Near Infrared Spectrosc., 2013, 21(4), 107–115.

---

Footnote

† These authors contributed equally to this work.

---