a continuous manufacturing model for the production of...

Universidade de Lisboa

Faculdade de Farmácia

A continuous manufacturing model for the production of granules by roller compaction

Rute Sofia Fonseca Cordeiro Dias

Dissertation to obtain the Master of Science Degree in

Pharmaceutical Engineering

Work supervised by:

Professor João Almeida Lopes (Supervisor)

Professor Ossi Korhonen (Co-Supervisor)

2016

The experimental work was performed in PROMIS continuous tablet

manufacturing line (University of Eastern Finland, School of Pharmacy,

Kuopio, Finland). All the facilities, equipments, materials and support

were gently provided by


i

Abstract

Continuous manufacturing is an encouraging and a sustaining innovation in

pharmaceutical manufacturing, build upon quality-by-design principles, with a huge potential

to improve agility, flexibility and robustness to the manufacturing process. In this field, the

Roller Compaction (RC) process plays an important role since it enables continuous dry

granulation of powders. Here, the powder is densified by two counter-rotating rolls that

produce a ribbon. Then, the ribbon is milled into granules, adequate for tableting or capsule

filling. RC overcomes granulation problems with thermolabile moisture or solvents sensitive

compounds.

In this work, an experimental design was performed in order to identify the critical

process parameters (CPP) and evaluate their impact on the critical quality attributes (QCA)

of granules produced by RC. The roller compactor used was the Hosokawa Bepex

Pharmapaktor® L200/30P, with a Flake Crusher FC 200. The RC process was monitored by

a near infrared (NIR) system and a direct imaging analyzer for granules’s size (Eyecon).

For the DoE the CPP’s proposed and their respective range were: compression force

(15-35 kN), roller speed (3-8 rpm) and mill speed (50-250 rpm). The produced granules were

characterized according to their particle size, as well as their bulk and tapped density –

granules’ CQA’s.

All process variables were kept constant 2 minutes after the process onset. The

compression force fluctuated throughout the process run time. The compression force was

the variable that most affected the granules’ CQA’s: the size and the density of the granules

are directly proportional to the compression force.

The Eyecon´s measurements exhibited significant deviations when compared with the

gold standard method, thus it was not an accurate method for monitoring the granules’ size.

Two approaches were followed for the prediction of granules’ physical properties. The

first model, that used partial least squares to predict the granules’ size, was built upon near

infrared data. It returned a high RMSEP (50.54 μm) and a poor coefficient of determination

for the prediction set (0.19), so it was not acceptable for the prediction of the granules’ size.

The second approach considered process parameters data to predict the bulk density,

tapped density and size of the granules. One partial least squares model was built to predict


ii

each response. The coefficient of determination for the prediction set was high for the three

models (0.93 for granules’ size, 0.95 for tapped density and 0.96 for bulk density)

demonstrating a good prediction ability.

Key words: Continuous Manufacturing; Critical Process Parameters; Near-

infrared Spectroscopy; Partial Least Squares; Roller Compaction.


iii

Resumo

A produção contínua no contexto da indústria farmacêutica surge da contínua demanda pela

produção de medicamentos de alta qualidade combinada com formas mais eficazes de

avaliação da qualidade. Ao contrário da produção em “lote”, na produção contínua as

matérias-primas entram e saem continuamente ao longo do ciclo de produção. Com efeito,

as vantagens da adoção do processo em contínuo são diversas e podem resumir-se em três

áreas fundamentais: controlo de qualidade e desenvolvimento do produto/processo, custos e

área de produção.

O projeto de produção em contínuo pode ser apoiado nos princípios definidos pelo “quality-

by-design”, desde o início do desenvolvimento do produto até à sua entrada no mercado.

Portanto, a qualidade é erigida por desenho e não apenas avaliada no produto final. Para

isso, realizam-se delineamentos experimentais para perceber de que forma os parâmetros

críticos do processo (CPP’s) influenciam as respostas do processo. As tecnologias analíticas

de process são ferramentas fundamentais de monitorização que fornecem dados do produto

e do processo em tempo real. Estes dados alimentam os modelos multivariados, os quais

não só fornecem informações para a compreensão do processo como também estão na

base da construção dos modelos de controlo. Os controlos estão aptos a detetar anomalias

no processo e a ajustar os CPP’s de forma a garantir que os atributos críticos de qualidade

(CQA’s) do produto cumprem as especificações.

Por outro lado, no processo contínuo, os equipamentos são de menores dimensões, já que

operam em contínuo. Por conseguinte, a área de produção requer igualmente menores

dimensões, pelo reduzido tamanho dos equipamentos, pela ausência de salas de

quarentena e de operações intermediárias. Deste modo se depreende uma significativa

redução de custos pela redução de recursos humanos, pela redução de gastos com

equipamentos de grande escala, pelo menor consumo energético, pela redução de

desperdícios com produtos fora de especificação e pela redução no tempo de chegada de

novos produtos ao mercado.

A granulação é uma operação unitária com vista à obtenção de grânulos, aplicados tanto na

sua forma intermediária para a produção de comprimidos, quanto na sua forma farmacêutica

final. A granulação a seco operada num compactador de rolos é, em si, um processo


iv

contínuo que permite processar compostos termolábeis e passíveis de degradação pela

humidade e por solventes.

O processo de compactação por rolos inicia-se pela introdução da mistura de pó no

granulador, através de uma tremonha de alimentação. As partículas de pó ao chegarem à

zona de alimentação sofrem um rearranjo e densificam, devido à pressão exercida pelo

parafuso rotativo no sentido descendente e ao afunilamento desta zona, a qual se prolonga

na seguinte. A mistura de pós é então forçada a continuar pela zona de compressão, onde

numa primeira fase, devido à pressão exercida sobre as partículas, ocorre deformação ou

quebra das mesmas. À medida que as partículas avançam nesta zona, a força de

compressão aumenta até que ao atingirem o termo da zona, a força de compressão

exercida pelos dois rolos em contra-rotação provoca a fragmentação das partículas e, logo

em seguida, a sua ligação até formarem fitas compactas. Por fim, as fitas são expelidas

pelos rolos, cortadas e moídas, de modo a formarem grânulos.

O NIR é uma ferramenta PAT que devido ao facto de a interação da radiação com a matéria

ocorrer na região de 780-2500 nm, permite uma interação direta com a amostra sem causar

a sua destruição. O NIR tem a capacidade de monitorizar o processo em tempo real,

fornecendo dados para a construção de modelos multivariados. O modelo PCA ao

desvendar as combinações de variáveis que descrevem a maior tendência nos dados,

fornece informações sobre o “comportamento” de todo o processo. O modelo PLS é utilizado

para prever as propriedades do produto final.

Este trabalho pretendeu identificar os atributos críticos do processo responsáveis pelas

alterações nos atributos críticos de qualidade dos grânulos. Além disso, procurou estimar as

propriedades dos grânulos (tamanho e densidade) com base em dados espetrais e dados

do processo.

Este trabalho foi realizado utilizando a Linha de produção contínua de comprimidos do

PROMIS Centre (University of Eastern Finland, School of Pharmacy) em Kuopio, Finlândia.

Neste trabalho, recorreu-se a alimentadores gravimétricos, um misturador contínuo e um

compactador por rolos (Hosokawa Bepex Pharmapaktor® L200/30P, FC 200). A

monitorização do processo foi seguida em linha por um sistema de infravermelho próximo

(Specim RHNIR, Spectral Imaging Ltd, Oulu, Finland) e por um medidor do tamanho de

partículas (EyeconTM, Innopharma Labs, Dublin, Ireland). A densidade dos grânulos foi

medida por deslocamento do volume dentro de uma proveta e o tamanho dos grânulos foi

medido num equipamento por dispersão de luz (método padrão).


v

A relação entre os possíveis CPP’s e os CQA’s dos grânulos foi estabelecida através de um

delineamento experimental, onde o a velocidade dos rolos variou de 3 rpm a 8 rpm, a força

de compressão variou de 15 kN a 35 kN e a velocidade do moinho variou de 50 a 250 rpm.

Considerou-se a densidade (areada e batida) e a dimensão dos grânulos como CQA’s.

O processo foi monitorizado por infra-vermelho próximo tendo os espetros adquiridos sido

processados para remoção de variabilidade de linha de base e escala. A identificação de

espetros atípicos (ou outliers) e a trajetória do processo de acordo com a informação

espetral foi avaliada através de modelos de análise de componentes principais, esta última

através do primeiro componente principal.

A velocidade do parafuso alimentador, a velocidade dos rolos e a velocidade do moinho

estabilizaram cerca de 2 minutos após o início do processo. A força de compressão variou

sempre ao longo de todo o processo, independentemente da força de compressão alvo.

A densidade dos grânulos aumentou (ex. areada, 0,62 g/ml – 0,64 g/ml ) com o aumento da

força de compressão (35 kN) e diminuiu (ex. areada, 0,54 g/ml – 0,55 g/ml) com a

diminuição da força de compressão (15 kN). O valor intermédio de força de compressão (25

kN) originou grânulos com densidade intermédia (ex. areada, 0,58 g/ml – 0,60 g/ml).

O tamanho dos grânulos foi influenciado sobretudo pela força de compressão. Assim,

quanto maior a força de compressão aplicada, maior o tamanho dos grânulos (ex. 1009 μm);

e quanto menor a força de compressão aplicada, menor o tamanho dos grânulos (700 μm).

A aplicação de uma força de compressão intermédia produziu grânulos de dimensão

intermédia entre as duas anteriores (881 μm).

Na medição da dimensão das partículas, o método por análise de imagem revelou um

desvio significativo em relação ao método por dispersão de luz (método padrão), não

constituindo por isso um método preciso para a monitorização em linha do tamanho dos

grânulos.

Foram construídos dois modelos de regressão multivariada baseados em mínimos

quadrados parciais para prever a densidade e a dimensão dos grânulos. O primeiro modelo,

construído com os dados NIR, teve o intuito de prever a dimensão dos grânulos. Uma vez

que o modelo apresentou um RMSEP elevado (50.5 μm) e um um coeficiente determinação

de previsão muito baixo (0.19), considerou-se que este modelo não é aceitável para a

previsão do tamanho dos grânulos.

A segunda abordagem de regressão utilizou os parâmetros do processo para prever as

respostas do processo. Foram modeladas a densidade areada (BD), a densidade batida


vi

(TD) e a dimensão dos grânulos (GS). Pela análise estatística dos modelos, os coeficentes

de determinação de previsão foram: 0,89 para GS, 0,90 para TD e 0,91 para BD. Estes,

indicam uma boa capacidade de previsão das respetivas respostas. Cada um dos modelos

demonstrou, também, que a força de compressão foi a variável que mais influenciou, e de

forma positiva, tanto a densidade como a dimensão dos grânulos.

Palavras-chave: Compactação por Rolos; Espetroscopia de Infravermelho próximo;

Granulação; Quimiometria; Produção Contínua.


vii

Acknowledgements

I would first like to thank Prof. Ossi Korhonen, for the privilege and the wonderful opportunity

to work with his team. I also thank him for always support my work in time with patience and

wisdom. A great thank to Prof. Jarkko Ketolainen for allowing my internship at School of

Pharmacy, University of Eastern Finland.

I would also like to thank Anssi-Pekka Karttunen for the good environment we had in the Lab,

for the positive discussions concerning the Lab work and for all support on the thesis writing.

I am deeply grateful to Prof. João Almeida Lopes for providing the opportunity to develop the

experimental work at University of Eastern Finland. Besides, I am also grateful for all I have

learned from his experience and knowledge and for his valuable support on this thesis.

Special thanks to André Canas for all we have learned together, for all the support during the

internship and for our friendship.

I would also like to acknowledge my friends Sara Ramos, Joana Oliveira, Sofia Correia,

Raquel Chaves, Pedro Molero, André Glória and Miguel Vieira for unfailing support and

continuous encouragement throughout this internship.

Last but not the least, I would like to thank my family for the unconditional support and

encouragement throughout my years of study and through the process of researching and

writing this thesis.

This accomplishment would not have been possible without them. Thank you.


viii

Contents

Abstract ................................................................................................................................... i

Resumo ................................................................................................................................. iii

Acknowledgements .............................................................................................................. vii

Contents .............................................................................................................................. viii

List of figures .......................................................................................................................... x

List of tables ......................................................................................................................... xii

List of abbreviations ............................................................................................................ xiii

Chapter 1 .............................................................................................................................. 1

Motivation .......................................................................................................................... 1

Chapter 2 .............................................................................................................................. 2

Objectives .......................................................................................................................... 2

Chapter 3 .............................................................................................................................. 3

Introduction ........................................................................................................................ 3

3.1 Continuous manufacturing ........................................................................................ 3

3.2 Roller compaction ....................................................................................................11

3.3 Process monitoring tools .........................................................................................20

Chapter 4 .............................................................................................................................23

Materials and methods ......................................................................................................23

4.1 The continuous manufacturing line ..........................................................................23

4.2 Equipment ...............................................................................................................25

4.3 Raw materials..........................................................................................................32

4.4 Methods ..................................................................................................................35

Chapter 5 .............................................................................................................................40

Results and discussion .....................................................................................................40

5.1 Exploratory data analysis ........................................................................................40


ix

5.2 Analysis of process variables ..................................................................................44

5.3 Analysis of the granules ..........................................................................................49

5.4 Predictive models ....................................................................................................63

Chapter 6 .............................................................................................................................74

Conclusions ......................................................................................................................74

Future perspectives ..........................................................................................................76

References ...........................................................................................................................77


x

List of figures

Figure 1 – The integrated continuous manufacturing process concept versus the more widely adopted batch processing concept. (Adapted from (1)) .................................................................................................................. 4 Figure 2 – Representation of compaction zone with horizontal rolls, inside the roller compactor. (1) Feed Zone; (2) Compaction Zone (Nip zone + Grip Zone); (3) Extrusion Zone. (Adapted from (33)) ............................................ 13 Figure 3 – Feeder orientations. (a) Vertical, straight; (b) Inclined; (c) Vertical, tapered; (d) Horizontal. (Adapted from (30)) ............................................................................................................................................................... 14 Figure 4 – Roll orientations. (A) Horizontal; (B) Inclined; (C) Vertical. (Adapted from (30)) ................................... 15 Figure 5 – Roll surface. (a) Smooth roller; (b) Corrugated; (c) Fluted. (Adapted from (30)) .................................. 15 Figure 6 – Main components of an off-line NIR instrument. (a) radiation source; (b) sample–radiation interaction device; (c) wavelength selector; (d) detector; (e) data collector, processing, storing and control device. (Adapted from (66)) ............................................................................................................................................................... 20 Figure 7 – Depiction of complete continuous line.A – powder loss in weight feeders; B – continuous mixer; C – roller compactor; D – screw conveyor; E – vacuum conveyor; F– tableting machine. (Adapted from (74)) ........... 23 Figure 8 - Labview control software interface: monitoring of the powder mixture. ................................................. 25 Figure 9 – Labview control software interface: monitoring of the compaction force. .............................................. 25 Figure 10 – Assembly of the four LIW feeders that feed the mixer. ....................................................................... 26 Figure 11 – Continuous mixer disassembly. On the left: mixer bladed shaft; on the right: the mixer outlet. .......... 27 Figure 12 – RC Hosokawa Pharmapaktor L200/30P. On the top left: RC prepared to operate; on the top right: detail of the vertical srew; bottom, on the bottom: detail of the milling screen. ...................................................... 28 Figure 13 – The RC rollers. On the top: the two disassembled rollers of 25 Kg each; on the bottom: top view of the two horizontal assembled rollers showing the fixed gap between them. ............................................................... 29 Figure 14 – The compaction system. A) The force transducer. B) The Hexagonal nut on the right roller arm. C) The roller shoulder. The force transducer measures the pressing force between the rollers. The hexagonal nut is tightened to increase the pre-pressing force or is loosened to reduce the pre-pressing force. The roller shoulder fixes the rollers and is connected to the roller arms by eye bolts........................................................................... 29 Figure 15 – The NIR monitoring system. On the left: the NIR sphere with the 6 optic fibers; at the centre: the spectral camera and the laptop with control software; on the right: the tungsten lamp, used as light source. ....... 30 Figure 16 – Configuration of the NIR sphere and Eyecon to on-line monitoring of the granules. On the top left: the NIR sphere attached to the outlet of the RC and the glass frame through which Eyecon collects images of the granules; on the top right and on the bottom: Eyecon capturing granules’ images. .............................................. 31 Figure 17 – Malvern Mastersizer 2000, with Scirocco Unit (on the left) and Hydro Unit (on the right). (Adapted from (76)) ............................................................................................................................................................... 32 Figure 18 – Particle size distribution (PSD) of raw materials. ................................................................................ 34 Figure 19 – On the left: the RC assembled and coupled with the monitoring tools; on the right: detail of how the both NIR and Eyecon systems were assembled to collect the samples. ............................................................... 37 Figure 20 – The NIR spectra for the run N1. On the top: raw spectra; on the bottom: spectra pre-processed with first derivative. ....................................................................................................................................................... 41 Figure 21 – The score plot for run N1, depicting PC 1 versus PC 2. The total variance captured by both components is 99.04%. The ellipse’s border represents the 95% confidence limit............................................... 42 Figure 22 – The statistic plot for run N1, depicting Hotelling’s T2 statistic versus Q statistic. The dotted lines represent the 95% confidence limit for both Hotelling’s T2 statistic (horizontal dotted line) and Q statistic (vertical dotted line). ............................................................................................................................................................ 43 Figure 23 – Process Variables Variation for Run N1. The target compression force was the lowest of the DoE (15 kN). After about 2,5 min all process variables reached their target values, except the compression force, which fluctuation lasted until the end of the run. The Mill Speed stabilized at 24 s, Roll Speed at 28 s and Screw Speed at 140 s. ................................................................................................................................................................. 46 Figure 24 – Compression force fluctuation showing the roll force mean line. The mean value stood below the target, at 13 kN-14 kN. The target of 15 kN was reached at about 154 s. After that the compression force varied between 10 kN (645 s, 1030 s,…) and 20 kN (1281 s). ......................................................................................... 46 Figure 25 – Process Variables Variation for Run N12. The target compression force was the highest of the DoE (25 kN). After about 2,2 min all process variables reached their target values, except the compression force, which fluctuation lasted until the end of the run. The Mill Speed stabilized at 24 s, Roll Speed at 21 s and Screw Speed at 110 s. ..................................................................................................................................................... 47 Figure 26 – Compression force fluctuation showing the roll force mean line. The mean value stood below the target, at 22 kN-24 kN. The target of 25 kN was reached at about 131 s. After that the compression force varied between 17 kN (285 s, 992 s,…) and 35 kN (572 s). ............................................................................................. 47 Figure 27 – Process Variables Variation for Run N4. The target compression force was the intermediate value of the DoE (35 kN). After about 2 min all process variables reached their target values, except the compression


xi

force, which fluctuation lasted until the end of the run. The Mill Speed stabilized at 15 s, Roll Speed at 24 s and Screw Speed at 118s. ........................................................................................................................................... 48 Figure 28 – Compression force fluctuation showing the roll force mean line. The mean value stood close to the target, at 34 kN-36 kN. The target of 35 kN was reached at about 120 s. After that the compression force varied between 27 kN (276 s, 556 s,…) and 44 kN (388 s). ............................................................................................. 48 Figure 29 – 90 % fraction of the granules size distribution (SD) obtained off-line by light scattering. The runs are ordered from the highest target compression force to the lowest target compression force. ................................. 51 Figure 30 – 50 % fraction of the granules size distribution (SD) obtained off-line by light scattering. The runs are ordered from the highest target compression force to the lowest target compression force. ................................. 52 Figure 31 – 10 % fraction of the granules size distribution (SD) obtained off-line by light scattering. The runs are ordered from the highest target compression force to the lowest target compression force. ................................. 53 Figure 32 – 90 % fraction of the granules size distribution (SD) obtained on-line by direct imaging. The size of each sampling point corresponds to the average of sizes captured by Eyecon during the sampling time settled for each run. The runs are ordered from the highest target compression force to the lowest target compression force. .............................................................................................................................................................................. 54 Figure 33 – 50 % fraction of the granules size distribution (SD) obtained on-line by direct imaging. The size of each sampling point corresponds to the average of sizes captured by Eyecon during the sampling time settled for each run. The runs are ordered from the highest target compression force to the lowest target compression force. .............................................................................................................................................................................. 55 Figure 34 – 10 % fraction of the granules size distribution (SD) obtained on-line by direct imaging. The size of each sampling point corresponds to the average of sizes captured by Eyecon during the sampling time settled for each run. The runs are ordered from the highest target compression force to the lowest target compression force. .............................................................................................................................................................................. 56 Figure 35 – The RSD determined for 90% fraction of the granules size distribution (SD) obtained by Eyecon. .... 57 Figure 36 – The RSD determined for 50% fraction of the granules size distribution (SD) obtained by Eyecon. .... 58 Figure 37 – The RSD determined for 10% fraction of the granules size distribution (SD) obtained by Eyecon. .... 59 Figure 38 – Comparison between off-line and on-line methods for granules’ size measurement. Square Error between the both methods regarding 90% fraction of the granules size distribution (SD). .................................... 60 Figure 39 – Comparison between off-line and on-line methods for granules’ size measurement. Square Error between the both methods regarding 50% fraction of the granules size distribution (SD). .................................... 61 Figure 40 – Comparison between off-line and on-line methods for granules’ size measurement. Square Error between the both methods regarding 10% fraction of the granules size distribution (SD). .................................... 62 Figure 41 – The scores of the calibration set and the test set used in the PLS prediction model. ......................... 64 Figure 42 – The statistic plot for the PLS model, depicting Hotelling’s T2 statistic versus Q statistic. The dotted blue lines represent the 95% confidence limit for both Hotelling’s T2 statistic (horizontal dotted line) and Q statistic (vertical dotted line). Both the scores of the calibration set and the test set are represented. ............................... 65 Figure 43 – The observed versus predicted plot for the PLS model. ..................................................................... 66 Figure 44 – The values of the responses plot against experimental runs. The replicated experiments are shown in blue on the same replicate index and the other experiments in green. .................................................................. 68 Figure 45 – Summary of the basic model statistics for every response: R2, Q2, model validity and reproducibility. .............................................................................................................................................................................. 70 Figure 46 – The coefficient plot depicting the model terms for each response. RolSp, Roll Speed; ComFc, Compression Force; MilSp, Mill Speed. ................................................................................................................. 71 Figure 47 – The observed versus predicted plot for each model. Bulk Density: R2=0.96, Q2=0.91; Tapped Density: R2=0.95, Q2=0.90 Granule Size: R2=0.93, Q2=0.89. ............................................................................. 71 Figure 48 – The score plot for each model, depicting the relationship between the scores (t1) and the responses (u1). ....................................................................................................................................................................... 72 Figure 49 – The loading plot for each model depicting the first LV against the second LV. BD= Bulk Density; TD= Tapped Density; GrS= Granule Size; RolSp= Roll Speed; MilSp= Mill Speed; ComFc= Compression Force. ...... 73 Figure 50 – The residuals of each response versus the normal probability of the distribution. .............................. 73


xii

List of tables

Table 1 – Description of the equipment used in complete continuous line. ........................................................... 24 Table 2 – Some physical properties of the raw materials. ..................................................................................... 33 Table 3 – Design of experiments obtained by Modde. ........................................................................................... 35 Table 4 – Detail of the DoE performed, showing the order by which the experiments were done. The screw speed was adjusted to the roll speed in order to obtain the target compression force. The lag time was the time considered for the variables to stabilize. ................................................................................................................ 44 Table 5 – The granules’ densities determined for each run and the process throughput. The table was split in three blocks, according to the target compression force (from the highest compression force, on the top, to the lowest compression force, on the bottom). ............................................................................................................ 49 Table 6 – Description of the PLS model to predict granules’ size based on NIR spectra. ..................................... 64 Table 7 – The overview of the factors and the responses used to build the second PLS model. .......................... 67 Table 8 – Summary of the PLS Model built with the process parameters. BD, Bulk Density; TD, Tapped Density; GS, Granule Size; RS, Roll Speed; CF, Compression Force; MS, Mill Speed. MS, Mean Square; p, probability. 69


xiii

List of abbreviations

ANOVA, Analysis Of Variance

API, Active Pharmaceutical Ingredient

BA, Bonding Area

BD, Bulk Density

BS, Bonding Strength

cGMPs, Current Good Manufacturing Practises

CM, Continuous Manufacturing

CPP, Critical Process Parameters

CQA, Critical Quality Attributes

DG, Dry Granulation

DoE, Design of Experiment

EMA, European Medicines Agency

FDA, Food and Drug Administration

LV, Latent Variable

MS, Mean Square

MSPC, Multivariate Statistical Process Control

NIR, Near-Infrared

NIRS, Near-Infrared Spectroscopy

PC 1, Principal Component 1

PC 2, Principal Component 2

PCA, Principal Component Analysis


xiv

PAT, Process Analytical Technology

PLS, Partial Least Square

PMP, Pharmaceutical Manufacturing Process

PSD, Particle Size Distribution

QbD, Quality by Design

RC, Roller Compactor

RMSECV, Root Mean Square Error of Cross Validation

RMSEP, Root Mean Square Error of Prediction

RMT, Raw Materials Traceability

RSD, Relative Standard Deviation

RTD, Residence Time Distribution

RTRT, Real-Time Release Testing

SavGol, Savitzky-Golay

SD, Size Distribution

SE, Square Error

TD, Tapped Density

TRU, Traceability Resource Unit


1

Chapter 1

Motivation

The continuous manufacturing (CM) approach for medicines offers immense benefits over

traditional batch processing. Not only because it affords flexibility and economic advantages

but also because the higher knowledge-based and control level drives the process straight

into the highest quality standards.

A CM line built to produce immediate release tablets has been among the most studied

processes in the endeavor for making CM a reality. The roller compaction process plays an

important role in this context, since dry granulation is more suitable to CM processes and has

known advantages as preserving heat and moisture sensitive compounds, among others.

The CM approach for the manufacturing of drug products lays on top of the quality-by-design

(QbD) principles, which demand the implementation of process understanding

methodologies and requires full adoption of process analytical technologies (PAT). In this

context, the widely adopted near-infrared spectroscopy method, as a PAT tool, enables real-

time monitoring of the process, feeding process models with frequent and high quality data.

Statistical models can be adjusted to deal with these data to monitor and to estimate in real-

time drug product properties.

Thus, this work aims at providing knowledge on powder granulation by roller compaction, as

a continuous process, to support its future implementation in pharmaceutical industry.

The experimental work was performed in the research and development continuous tableting

line (PROMIS-line), constructed at the School of Pharmacy of the University of Eastern

Finland.


2

Chapter 2

Objectives

This written thesis is the result of the experimental work performed in the PROMIS-line

(School of Pharmacy of the University of Eastern Finland).

Therefore, the main objectives of this work are:

1) identify the critical process parameters affecting the critical quality attributes of granules

produced by RC;

2) implement an analytical method (near infrared spectroscopy) to monitor in real-time the

quality of granules in the RC process;

3) build predictive models for estimating the properties of granules (size and density) from

spectral and process data.


3

Chapter 3

Introduction

3.1 Continuous manufacturing

The pharmaceutical manufacturing sector is characterized by a traditional way of

manufacturing drug products (batch production), that has been revealed often inefficient at

several levels. Thus, the lack of agility, flexibility and robustness in pharmaceutical processes

leads to failure in production units which is in the origin of drug shortages, high production

costs and waste. Since shortages are mainly related to a supply disruption caused by non-

conformities in the quality of whole facility or drug product, regulatory authorities have been

looking for promoting the adoption of more scientific mechanistic understanding of processes

and methods applied on pharmaceutical industry.(1)(2) This endeavour is among the aims of

regulatory authorities, such as Food and Drug Administration (FDA) or the European

Medicines Agency (EMA) as they regulate pharmaceutical drug products and promote a

flexible pharmaceutical sector that reliably produces high-quality drugs without extensive

regulatory oversight.(3)

In general, in a batch process, the raw materials are loaded into the equipment at the

beginning and final products are discharged at once, at the end of the process. Nothing

comes in and nothing comes out of the equipment during the process run.(1) Some adopted

processes are indeed fed-batch as during the production cycle some components are added

in a continuous or discrete way.

In contrast, in a continuous process, the materials and products are continuously charged

and discharged over time.(1)

So far, a pharmaceutical manufacturing process (PMP) rather consists of a combination of

batch and continuous unit operations, such as in a tablet production line, where powder

blends are made in batch mode and tablets are produced continuously in a tablet press. The

PMP as a whole is globally considered to operate in batch mode.(4)

The future of PMP lies in continuous manufacturing (CM), a breakthrough over the standard

batch concept. In CM, single unit operations are all connected together in a logic sequential

order – integrated process. Process Analytical Technology (PAT) tools are implemented to


4

provide real-time acquisition data for process monitoring and control. Multivariate analysis is

performed to extract information and to gain scientific knowledge about the process.

Engineering process control systems are designed to reduce the impact of raw materials and

process variability on the quality of drug products.(1) An overview of those two

methodologies is represented in Figure 1

Figure 1 – The integrated continuous manufacturing process concept versus the more widely adopted batch

processing concept. (Adapted from (1))

Batch manufacturing has been applied for the production of pharmaceutical products since

pharmaceutical industry’s fast expansion in 80s.(5) Batch processing requires that samples

from one unit operation should be taken according to previously settled in-process controls.

Samples are tested off-line and stored, while they wait for quality control approval, and finally

they are sent to the next processing step. However, if the in-process product does not meet

quality specifications, then it may be discarded, or even reprocessed, before moving to the

next process step.(1)

In CM, materials obtained from each process step are sent directly and continuously to the

next step until the end of the process. This means that in-process material is produced within

quality specification limits, at a given time, straight through the final product.(1) Hence, CM

shows up as an integrated system approach built in with a model-based control placed

throughout the process flow. CM is, indeed, designed as a whole, so any distinction between

upstream and downstream or drug substance and drug product, as currently used, can be

eliminated.(5)

Although the aforementioned description is the best known for CM, some authors advocate

that the integration of several unit operations into a single line is a mere heterogeneous

process, since there´s still significant powder transport challenges, unnecessary process


5

steps and higher risk of process issues. Genuine CM involves “homogeneous processes”,

that is, the active-excipient combination is engineered so to have the key properties needed

with the view to directly make the final dosage form (e.g. extrusion, spray drying, thin film

formation). From this point of view, heterogeneous processes constitute the first step in the

transition from batch to homogeneous continuous processing.(5)

3.1.1 Advantages of CM over batch manufacturing

The overall advantages of CM embrace three main fields: 1) product development and

quality; 2) costs; and 3) footprint.(6) As these fields are intimately related with each other,

their benefic effects are felt as a whole over the CM concept.

CM offers a sort of opportunities to increase flexibility of PMP and to provide high quality

drug products. In fact, CM can increase production volume without scale-up concerns,

yielding a quicker response capacity. Scale-up deliberations, like process operating time,

number of parallel processing lines and flow rate set up are introduced into the process

design and control. Besides, the small volumes of raw materials needed to run the CM

process allow operating with smaller equipments, thus eliminating scale-up.(7)(4) Hence, CM

is able to reduce the costs regarding expensive active pharmaceutical ingredients (APIs) and

excipients needed for process development studies and optimization efforts. Also, CM puts

forward the streamlining of the whole process by excluding work-up unit operations.(1)(5)

Removing scale-up bottlenecks from process development, reduces time to market,

enhancing the opportunities to promote fast and efficient clinical development of new

drugs.(1)

CM is able to reduce supply chains. As production can be accomplished at different scales,

APIs, process intermediates and drug products are manufactured in-place, in a continuous

sequence operation, without any further storage or shipping steps. This is a major

improvement since removal of hold times between steps assures that sensitive materials are

not degraded; in addition, being manufactured in a small scale production line, less risks can

be associated with highly energetic or reactive materials. So, CM rises as a more flexible and

safe method even when applied to a non-specialized manufacturing facility.(4)(8)

CM is built under the regulatory authorities guidelines regarding quality-by-design (QbD)

approach for pharmaceutical development. According to the FDA, QbD is “a systematic

approach to development that begins with predefined objectives and emphasizes product

and process understanding and process control, based on sound science and quality risk

management.”(4,9,10). Over the last decade the International Conference on Harmonisation


6

(ICH) has released a few guidance documents in order to clarify how to implement this new

concept. The ICH Q8(R2) – Pharmaceutical Development, ICH Q9 – Quality Risk

Management, ICH Q10 – Pharmaceutical Quality System, ICH Q11 –Development and

Manufacture of Drug Substances, as well as ICH Quality Implementation Working Group on

Q8, Q9, Q10 Questions & Answers (R4) and Points to Consider (R2), documents consist of

high-level guidelines concerning the scope, definition and implementation of QbD in

pharmaceutical industry.(9)(11–15) A robust process is capable of producing drug products

with acceptable quality. However, only through a strategy for process control, can its

variability be controlled. Process control aims at achieving variability control by reducing

input variation and/or adjusting for input variation during production. Such a strategy should

be set up based on product and process understanding. Since PAT enhances process

understanding, a forward control strategy can afford real time analysis and control of the

output quality.(10) Thus, processes are controlled with robust and advanced process models,

leading to a much lower risk of going out of specification, when compared to batch

processes.(5)(16)

In short, CM process line is usually more efficient than its traditional batch counterpart. The

manufacturing footprint is reduced whereas intermediates flow from one processing step to

the other without concerning isolated rooms or dedicated modules. Consequently, CM

requires smaller equipment, thus on the one hand reduces both capital and operation

expenses and on the other increases throughput.(1)(17)

3.1.2 Challenges and barriers for CM

Despite of the greatest advantages of CM, it hasn´t become, yet, the gold standard for

pharmaceutical industry, mainly because of a “business as usual” perspective tied to a highly

conservative industry. Firstly, pharmaceutical industry is highly regulated: companies are

cautious regarding regulatory aspects of unconventional procedures; companies seek

approval for their products worldwide and CM may not be approved by regulators in other

different countries. Secondly, it has been found that new manufacturing concepts must be

proven both technologically and financially superior and applied to a product before

widespread implementation can occur. So, because technologies must be already adopted

for the industry to adopt them, the introduction of CM has been rather slow.(5)

However, this is starting to change. Whilst benefits of CM are being perceived by

management, more investments are made so that CM is gradually becoming more prevalent.

In the meantime, there are some challenges for the implementation of CM:


7

1) Business and organizational. The established batch asset-base, wherein a huge

capital was invested during the 80s and 90s, is a hurdle for introduction of CM. The capital

invested in new CM facilities can be compensated by significant savings, once that

continuous process monitoring and control assure lower energy consumption, higher

production yield and shorter cycle times. Also, CM requires a smaller, highly skilled

workforce and a smaller footprint, enabling flexibility in plant settlement in order to meet local

demand for diverse products or proximity to R&D centres.(5) It´s foreseen that new

therapeutic entities or approved drugs with high market share should be the potential

candidates for the introduction of CM.(1)

2) Manufacturing and development. The CM paradigm requires a new mindset

from scientists who develop formulation and process, through quality units within the

company, until government regulators supervising the industry. Consequently, engineers,

scientists and regulators must have to upskill in statistics to better analyze and understand

process data. To reap the benefits of CM, continuous process must be adopted at the

earliest possible stage of product development. In addition, to expand CM technologies, a

new generation of equipment, sensors and automation have to be developed, as well as

system integration in order to let operation units communicate and provide process

control.(5)

3) Technical aspects. There are two main tricky situations. One is the development

of accurate process operation models of various steps in a continuous process, together with

powder characterization and handling, especially for low-dose production. Another is how to

operate start-up and shutdown as fast as possible with minimal waste. Several solutions are

being proposed for the first case, whereas for the second one, a smart sequencing of unit

operations during start-up and shutdown was suggested to decrease losses. (5)

In CM, intermediate products flow between unit operations and drug products are generated

continuously over a certain period of time. In CM, as in any other manufacturing process,

process, raw materials, quality and environmental conditions can vary over time and so those

variations have to be considered when developing the CM control strategy. In order to do

that, it´s mandatory to understand the impact of process dynamics as well as to recognize

which materials attributes and which process variables have a significant impact on final

product quality. This gained knowledge should be applied to the design of measurement

systems for process monitoring and control.(20,21) Regulators and industry will have to keep

developing knowledge and experience with those systems-approach methodologies with the

view to achieve expertise in supporting and approval of a wider implementation of CM

processes.(1)


8

3.1.3 Quality considerations for CM

Process understanding

Design of Experiments (DoE) is a powerful tool to gain process understanding since as “a

structured, organized method for determining the relationship between factors affecting a

process and the output of that process”(9), it establishes the operational boundaries of the

process. Continuous process response to changes in process parameters is swift, thus

enabling a huge collection of data in a short period of time from a small amount of in-process

materials. These data are introduced into mathematical models to get process knowledge,

whilst predictive process models are applied as a simulation tool to generate experiments

and enhance process understanding, in the course of process development.(21)

Regulation of CM requires a batch definition for continuous processes and a method to

obtain raw material traceability (RMT) – the competence to maintain and access the identity

and attributes of the raw materials throughout the process.(22) Comprehension of how raw

materials flow through the process is crucial to RMT. This knowledge can be achieved by

characterization of residence time distribution (RTD), following a tracer experiment or a

process modelling.(23,24) RTD is a probability distribution that depicts how a material moves

within different unit operations of a continuous process system. The RTD curve is very useful

to predict the diffusion of raw materials, the disturbances over the system and to determine

when substances were introduced into the manufacturing system. RTD is influenced by

several factors like processing time, equipment parameters and raw material properties. A

suggested method to report material travelling over the system is a traceability resource unit

(TRU).(25) A TRU acts as a segment of material flowing through the process together with

other raw materials and can be used as a unique identifier to the process history point of

view. If the CM process is integrated with packaging units, unique package identifiers must

connect drug product supply chain traceability to process traceability, in order to trace any

packaged drug product from marketing to its original raw materials and vice-versa.(1)

Current Good Manufacturing Practises (cGMPs) guidelines define «batch» as a specific

quantity of final or intermediate product, featured by uniform properties and within specified

quality limits, which is fabricated according to an individual manufacturing order through the

same production cycle. Likewise, it´s expected that CM process produces batches. In fact,

for CM a «lot» is equivalent to «batch» and corresponds to a specific identified amount of

final product generated in a particular time or in a certain quantity in a way that both uniform

character and quality specified limits are warranted.(1)


9

In CM, RMT is closely linked to the definition of batch. When a process demonstrates an

ongoing state of control, a batch can be determined according to production time period,

quantity of raw materials processed, equipment run time capability or process variation (e.g.

different lots of incoming raw materials). Hence, batch concept is directly associated with

process control strategy, designed to deliver uniform quality within the batch.(1)

Control strategy

A control strategy for CM is built to control the quality of final product throughout the

manufacturing process, in response to potential variations like equipment conditions,

incoming raw materials or environmental factors.(26) Usually, three levels of control strategy

implementations are described:

- Level 1 control (engineering control) takes an active process control system to

monitor, in real-time, raw materials quality attributes and then to use that information

to adjust automatically process parameters with the view of conforming those quality

attributes to the acceptance criteria;

- Level 2 control (pharmaceutical control) assumes the definition of a process design

space, in which raw materials attributes and process parameters are established, and

also implements an adequate end-product testing;

- Level 3 control is exclusively based on raw material attributes and process

parameters with extensive end-product testing.(1)

Usually, in CM, the most convenient level of control is Level 1, although a hybrid perspective

of different levels of control is possible for some CM process designs.(16)(27)

Process knowledge gained from monitoring and control strategies supports process state of

control, a condition in which a continuous process is able to ensure that a final product is

consistently delivered with the desired quality.(1) However, in the course of start up and

shutdown procedures, at some point, there may be in-process materials or final products that

do not meet the target quality specifications. Therefore, the competence to isolate and reject

materials out of specification, as a consequence of a process no longer under a state of

control, is one of the main responsibilities of a CM control strategy. Once again, RTD models

are powerful tools to track non-conforming materials over the process from deviation point.

The settlement of a disposition strategy when products provide from a process that is not

under control together with the implementation of adequate process monitoring criteria,

constitute important issues to ensure quality through the production run.(1)


10

The increased amount of total data collection during a continuous process run motivates the

adoption of multivariate process monitoring methodologies. Multivariate statistical process

control (MSPC) constitutes a process monitoring methodology applied to determine whether

the variability in the process is stable over time. MSPC is used to detect abnormal events in

the process that may potentially cause severe consequences and unveils which process

variables may be in the origin of the event. MSPC also enhances detection of abnormal

process operations by identifying changes in the relationships between process parameters

and quality attributes.(28)

Since CM process holds in-process materials and discloses a fast process dynamics, real-

time monitoring of both process parameters and intermediates quality attributes shows up as

a crucial stage for the settlement of a state of control. Moreover, PAT tools and multivariate

models are used to build process understanding. Applying PAT tools enables the measuring

of the final product quality attributes, some of which may have already been integrated into

the control strategy for process monitoring and control.(1)(5) In-line PAT tied to the control

system, explains why real-time release testing (RTRT) comes up naturally for CM, as “the

ability to evaluate and ensure the quality of in-process and/or final product based on process

data, which typically include a valid combination of measured material attributes and process

controls”.(9)(6) So, whenever process perturbations occur, real-time rejection of non-

conforming products can be performed without losing the whole batch.(5) However, a risk

analysis should be accounted to assist on PAT failure and efforts should be done to establish

contingencies for process monitoring and batch release.

Innovation in pharmaceutical development follows the direction of manufacturing a high

quality product (highly suitable for its intended use), built by a scientific-based design, whose

production process is capable of consistently providing the desired performance of the

product. The information gathered during process development (through design of

experiments, PAT tools and prior knowledge) supports process knowledge and that, along

with product knowledge, provides the scientific basis to define the design-space,

specification boundaries and production controls.(9) The design-space represents the

relationship between both material attributes and process parameters, i.e. process inputs,

and the critical quality attributes (CQA). A CQA is a specific property that should be within an

adequate range to assure the intended product quality. A critical process parameter (CPP) is

a process parameter which variability influences the CQA and thus, should be monitored and

controlled to assure the production of a product with the expected quality.(9)


11

3.2 Roller compaction

Granulation is a unit operation that consists of particle size enlargement and densification by

agglomeration of small particles into larger ones. In pharmaceutical technology, the principal

reasons to granulate powders are: to improve powder flow properties for further dosage filling

and compression purposes; to improve product stability; to prevent powder mixture from

segregation; to reduce bulk volume so that storage is minimal and transport is facilitated; to

reduce potential environmental and safety hazards.(29)

There are several methods to perform granulation, even though they are usually classified as

either dry or wet. In wet methods, powder mixture particles are agglutinated by spraying an

appropriate solution to form a mass. After that, the wet mass is dried and larger particles,

named granules, are calibrated to get the adequate size for further downstream processing.

In modern pharmaceutical manufacturing, wet granulation has been by far the most common

powder granulation technology.(29)

Dry granulation (DG) was first applied to pharmaceutical industry in the late 1940s, but has

drawn particularly attention in the last 25 years as research in new API has increased. Some

of these new therapeutic compounds are either sensitive to heat or moisture and, therefore,

cannot undergo wet granulation.(29)

DG process began by producing uniform densified granules, called slugs, during a process

named slugging. Nowadays, DG is mostly performed using roller compaction technology,

whereas it offers many advantages over other granulation methods: (29)

1) it is among the most cost-effective granulation processes, since it requires less

space, personnel, energy and time consumption;

2) it is the only efficient and safe process for formulations with drug substances

sensitive to heat, moisture or solvents, since no liquid nor additional heat are needed;

3) it is a continuous process with higher throughput and lower energy consumption;

4) it is able to produce more homogeneous products, when compared to other DG

techniques;

5) it lends itself to a higher level of control, with implementation of on-line monitoring

and control tools, as well as to automation of process settings, in order to minimise variations

between batches and to improve product quality. (29–32)


12

3.2.1 Roller compaction process

The process of dry granulation is dependent upon bond formation between particles in order

to produce granules. The aggregation of particles, stuck together due to interparticulate bond

formation, is described by four distinct stages that occur accordingly to a specific order.

1) Particle rearrangement takes place when particles start to fill void spaces and air

begins to move out from interstitial spaces inside powder mixture. Then particles start to

come closer together and powder mixture´s density increases. Spherical particles tend to

pack to one another, moving less than particles with other different shapes. Thus, particle

shape and size are crucial in rearrangement process.

2) Particle deformation is considered to be a plastic deformation, as a result of

increasing compress forces applied on powder particles. This deformation enhances points

of contact between particles, which boosts their bonding.

3) Particle fragmentation comes up at higher compression forces, creating multiple

new particle surface sites and thus potential bonding points.

4) Particle bonding ensues as a consequence of plastic deformation and

fragmentation, particularly due to van der Waals forces.(29)

When powder granules undergo an applied force, a stress force is released from granules

and they tend to return to their original form – elastic deformation. Albeit, granules may not

completely recover from deformation and so plastic deformation occurs. Elastic and plastic

deformation can occur at the same time, though only one of them predominates.(29)

Roller compaction process starts when powder mixture enters into the feeding zone, through

a hopper, where particles are rearranged and densified. At this stage, press force over

particles is too low. These powders then go through the compaction zone, beginning by nip

region, where brittle particles break and plastic particles deform as a result of a sharp

increase in compaction force. At the end of nip area, the going up further compression force

exerted by two counter-rotating rolls causes particles fragmentation and particles bond to

form ribbons (Figure 2).(30)


13

Figure 2 – Representation of compaction zone with horizontal rolls, inside the roller compactor. (1) Feed Zone; (2) Compaction Zone (Nip zone + Grip Zone); (3) Extrusion Zone. (Adapted from (33))

The higher pressure occurs at neutral nip angle, which usually is slightly before the lower roll

gap. The nip angle is the angle where pre-densified powders enter the nip zone, and it

hinges on the material friction angle and the roll surface friction angle.(34) The nip angle is

large for compressible material, but small for incompressible material.(35)

After the roll gap, the ribbons are extruded from the rolls, then chopped and milled to

produce granules with adequate particle size.(30)

3.2.2 Roller compactor features

The main components of a roller compactor are: feeder, rolls, and mill.(30)

1) Feeder. Feeding system is critical to regulate powder flow, powder densification

and powder deaeration. The head pressure, a pressure differential established between the

bottom and top of the hopper, induces continuous densification and deaeration of incoming

particles.(30)

There are two types of feeders: gravity feeder and force feeder. The gravity feeder

has an adjustable tongue and a distributor at the end of hopper, working without an external

driving force. The force feeder has a rotating screw, with a single flight, installed in the centre

of the hopper. The rotating flight pushes the powder toward the nip area. Force feeders offer

several advantages: adequate for poor flowing powders, provide continuous and consistent

flowing to the nip region and prevent powder feeding disruption from trapped compressed

air.(30)

Feeders design is important to get positive pressure in powder feeding. The

orientation of feeder can be vertical, horizontal or inclined; for feed screw only straight (most

common) and tapered designs are available (Figure 3). Vertical feeders are favoured with


14

head pressure from the hopper. Horizontal feeders minimize leakage and improve

compactibility. Inclined feeders use also gravity to introduce powder, but show less powder

leakage; they are used for multi-screw feeders in large scale equipment.(30)

Figure 3 – Feeder orientations. (a) Vertical, straight; (b) Inclined; (c) Vertical, tapered; (d) Horizontal. (Adapted from (30))

Screw feeders unveil non-uniform feeding. The cross section of the nip area is

rectangular, but the rotating screw generates a circular deliver area. As a result, the centre

region of the nip is overfed, while the edges are scarcely fed, producing ribbons with thinner

and more fragile edges.(36) Also, as single rotating flight exerts pressure on the compact in

the roller gap, the feed pressure is irregular and induces an uneven distribution of the powder

in the nip area. Hence, ribbons may be more compacted in one area than the other.(37) This

effect can be reduced, following one of these suggestions: round off the end of screw flight,

select double flight screws to get a uniform force distribution, choose multiple feeder screws

and elevate the position of the screw.(30)

Feed screw affects granules characteristics by controlling the roll gap, which further

regulates the amount of material coming between the rolls.(38) Hence, the roll gap defines

the ribbon thickness and, consequently, the product quality. It has been understood that

when the roll gap is reduced, powder is compressed at a much higher compaction force, so

more densification occurs. As a result, the average density of the ribbon is inversely

proportional to the roll gap.(39) Besides, if the two rolls are fixed, the compaction force will


15

vary enormously with the fluctuating mass flow, throughout the process. On the contrary, if

the rolls are movable, roll compaction force can be kept constant by changing the gap width

according to the mass flow.(40)

2) Rolls. Rolls are key components that exert compression force. Rolls are

characterized by their orientation, mounting, arrangement and pressurizing system. The rolls

orientation determines the feeder orientation. Therefore, rolls can be horizontal (most

common), inclined or vertical (Figure 4). Horizontally aligned rolls are equipped with vertical

or inclined feeders; vertically aligned rolls are equipped with horizontal feeders.(30)

Figure 4 – Roll orientations. (A) Horizontal; (B) Inclined; (C) Vertical. (Adapted from (30))

The surface of rolls can be smooth, corrugated or fluted (Figure 5). Smooth and

corrugated rolls are the most common. Smooth rolls are advantageous in overcoming

sticking problems and also can reduce the amount of lubricant needed. Corrugated rolls are

selected to give more gripping force to the powder, in order to solve the inadequate feeding

and uneven compact, though powder sticking may be a problem.(30)

Figure 5 – Roll surface. (a) Smooth roller; (b) Corrugated; (c) Fluted. (Adapted from (30))


16

3) Mill. The milling step intends to obtain granules from the ribbon. In general, the mill

consists of a moving rotor and a fixed sieve of a chosen mesh size. Not only the rotor speed,

but also the rotation direction (clockwise direction, counter-clockwise direction or a

combination of both) can be adjusted, according to the equipment. Some authors have

proposed that operating in oscillating mode (clockwise/counter-clockwise) yields higher

throughput than rotation (clockwise). Concerning the mill speed, it is reported that, typically,

the higher the mill speed, the higher percentage of fines is got at the end of the process.(41)

Other studies have shown that the effect of milling conditions on granules mechanical

properties, obtained from die compaction, is irrelevant.(42)

Roll compactors are built with a sealing system to prevent the loss of powder from the

sides. The most common sealing systems are: cheek plates, with a fixed side sealing, and

rimmed-roll, in which side sealing is integrated with the bottom roll. In cheek plates assembly,

roll pressure and density distribution are non-uniform, whereas for rimmed-roll, the reverse

effect is observed.(43) In order to reduce the amount of fines produced during roller

compaction, it is preferable to work at high roll force and use rimmed-roll as a sealing

system.(42)(44)

Other factors like deaeration, temperature control and feeder vibration are important as well

to keep the process feasible. In deaeration, the air trapped between particles may damage

the new-formed ribbon: when ribbon porosity is high, air continues to leave and ribbons get

weaker; when ribbon porosity is low, the trapped air may cause the ribbon to break

horizontally into pieces. The air usually escapes in three ways: between the feeder base and

the top of the roller, between the rolls and the cheek plates, against the flow of feed through

the loose bulk material. To minimize this problem, applying a vacuum at the top and bottom

level of the powder bed induces the trapped air out.(30)

The feeder vibrator can improve the powder flow, providing a continuous driving force to

break the stagnant powder bed, thus helping densification and deaeration.(30)

The screw flight can generate excessive heat when rotating in the powder bed. This heat

may elevate the local temperature and cause the powder to be partially melted and stuck to

the flight. To avoid this situation, a special flight with a cooling jacket can be used.(30)


17

3.2.3 Impact of process parameters

In roller compaction process, the critical process parameters needed to be optimized to

assure process feasibility, ribbon quality and granule tabletibility are: compaction force, roll

speed and feeder screw speed.(40)(45–47)

1) Compaction force. When subject to press force, solid particles densify, deform or

fracture and bond to form ribbons. At high compaction force, a strong ribbon with low porosity

and less fines is produced.(48–50) Over-compaction may break the ribbon and generate

poor quality granules.(30)

Once the granules from RC are to be compressed, one of the main issues regarding

a DG process is the phenomenon of loss of tabletibility of granules.(51–53) Tabletibility or

compactibility (tablet tensile strength versus pressure) describes the capacity of a powder to

be transformed into tablets with defined strengths under determined pressures. Tablet tensile

strength is the result of the interplay between bonding area (BA) and bonding strength

(BS).(54,55) Obviously, larger BA amidst granules or higher BS benefits stronger tablets.

Several mechanisms have been proposed to explain the loss of tabletibility in DG.

One of the earliest hypotheses put forward is the “work hardening”, defined as the production

of harder granules with markedly increased resistance to deformation.(56) Since work

hardening is difficult to measure in granules, some authors have proposed its replacement by

“granule hardening”(57) (under a compaction force, the granules porosity is reduced and the

granule strength increases). Because of granule hardening, higher compaction forces lead to

harder ribbons, harder granules and lower tensile strength tablets.(49)

It has been concluded that BA and BS influence mechanisms that determine

tabletibility. In the case of plastic materials, the main mechanisms are: lubrication, granule

size enlargement and granule hardening. Adding more lubricant, especially when blended in

the mixture, leads to lower BS and reduced tabletibility. Thus, granules with a higher porosity

are more deformable when subjected to compaction force and promote large BA and a

stronger tablet. In brittle materials, granule hardening is the most important mechanism for

tabletibility. However, the dominating mechanism for each situation depends upon material

properties and process parameters.(58)

Some authors have pointed out that compaction force is the most important factor

that affects the quality of the compact,(59) mostly for plastic materials. As for brittle materials,

they seem to be less susceptible to compactibility loss. Thereby, since powder mixtures often

contain plastic and brittle materials, an optimum operating range of compaction force should

be determined on a case-by-case base throughout process development.(30)


18

2) Screw speed. The rotating flight generates a force which induces a downward

compression force, pushing the powder into the compaction zone and causing its pre-

densification.(36) Screw speed operating range should be determined based on powder flow,

aeration condition and roll speed. Low screw speed may provoke scant feeding to the nip

zone and poor ribbon strength. High screw speed may cause a highly densified zone in the

nip area and eventually induce melting of particles on the flight, which may stop powder flow.

A proper formulation modification, adequate deaeration or addition of feeder vibrator may be

necessary to improve process feasibility.(30)

3) Roll speed. The roll speed controls the dwell time of the ribbon and the throughput

of the roller compactor. Roll speed is defined according to the flowability, plasticity and

elasticity of the powder.

For plastic materials, sensitive to dwell time, a low roll speed is prone to produce

granules with better flow and lower friability.(60) But longer dwell time may also cause the

material to lose compactibility, generating tablets with low hardness and high friability. The

impact of dwell time on plastic or partially plastic deforming materials can be reduced with

high roll speed and low screw speed.(59)

In case of elastic recovery materials, the compact strength depends on the dwell time

in the compaction zone, as elastic recovery may occur upon the release of the ribbon. High

roll speed or short dwell time may provoke cracking, weakening or even the destruction of

the ribbon. Hence, for highly elastic materials, the overall process throughput is hedged by

physical properties of raw materials.(30)

For brittle materials, the compact strength reveals independence from dwell time,

once fragmentation occur in short time and so exposure to compaction force tends to have a

limited effect on the ribbon strength.(61–63)

4) Relationship between roll speed and screw speed. The production of granules

with desirable compression capacity is achieved by controlling the ratio of roller speed to

screw speed. At constant roll speed, a low screw speed may lead to insufficient feeding and

thus to thinner ribbons and weaker ribbon strength. On the contrary, high screw speed may

induce overfeed and produce thicker ribbons.(48) When roller speed and screw speed ratios

are kept constant, both the ribbon density and strength are independent of roller speed.(60)

The definition of roll speed to screw speed ratios hinges mostly on the properties of

powder blend. Generally, for plastic materials, a higher roll speed to screw speed ratio gives

better quality granules and tablets (best tablet friability, hardness and dissolution rate).(30)


19

Finally, the most important critical quality attributes are:

1) Granule particle size distribution. Small particle size of both raw materials and

granules have a positive effect on enhanced tablet strength.(51) Large granules may induce

weight variation during compression and because less BA is exposed, tablets exhibit low

tensile strength. (45)(58)

2) Granule/ribbon density. Ribbon is not homogeneous in terms of density

distribution, showing lower density at the edges and higher density at the centre.(31)(45)(39)

3) Granule/ribbon porosity. Recent studies have highlighted the importance of

granule porosity. In fact, granules with high porosity, under a compression force, breakdown

and above a critical porosity value disintegrate completely into primary particles, producing a

tablet which microstructure resemble that formed from the original powder.(64,65) For that

reason, critical porosity should be lower at higher compaction force.(58)

Harder ribbons (resulted from higher compaction force) exhibit low porosity, and leads

to harder granules and lower tensile strength tablets.(45)(49)

4) Amount of fines produced. When there are too much fines, problems regarding

poor flow, weight variation and picking may occur during tablet compression. Moreover, fines

generated during RC are undesirable as they frequently cause material loss and reduce

granule flow properties. Also, it is strongly discouraged to recycle fines in the process, since

the regranulation of fines has a negative influence on API-conformity.(42,45)(51)


20

3.3 Process monitoring tools

3.3.1 Near-infrared spectroscopy

Near infrared spectroscopy (NIRS) refers to the interaction of light with wavelength from 780

to 2500 nm with matter. The observed spectral bands are overtones and combinations of

fundamental bands occurred in the mid-infrared region. The NIR bands are combinations of

hydrogen atoms with larger atoms, mainly, C–H, N–H, O–H bonds. The overtones of these

fundamental bands are close to NIR region, and that´s why they can be seen. However, NIR

bands are 10-100 times weaker than the corresponding mid-IR bands. Thus, NIR spectra

have been extensively used to identify raw materials as they are able to discriminate

compounds with resembling structures.(66)

Diffuse reflectance measurements derive from interaction of NIR radiation with solid

samples. The sample is illuminated directly, without any pre-treatment, and the incoming

radiation is absorbed or scattered by the particles in many angles and then collected through

mirrors. The scattered radiation that returns to the detector is called remitted fraction. The

obtained spectrum is a result of the intensity of the radiation measured by the detector. An

adequate software processes the detector measured intensities, generating a final spectrum

and stores the spectral data in a convenient format (Figure 6).(66)

Figure 6 – Main components of an off-line NIR instrument. (a) radiation source; (b) sample–radiation interaction device; (c) wavelength selector; (d) detector; (e) data collector, processing, storing and control device. (Adapted from (66))

Overview of the main NIR components

(a) radiation source – the most common is a tungsten filament source with a trace of

iodine inside a quartz bulb. Typical power and temperatures are in the range 25–100 W and

2000–3000 K, respectively.

(b) sample-radiation interaction device – promotes the interaction of the

polychromatic or monochromatic NIR radiation with the sample.

(c) wavelength selector – determines the resolution of the NIR spectra.


21

(d) detector – measures the intensity of the radiation scattered by the sample.

InGaAs devices (an alloy of gallium arsenide and indium arsenide) demonstrate a quick

response and a high detectivity over the range of 1000-2500 nm.

(e) data collector, processing, storing and control device – consists of a

microcomputer running a program able to control the spectrophotometer, to collect and to

process the detector measured intensities in order to build the final information in terms of a

spectrum (absorption intensities versus wavelength).(66)

In a continuous process, the use of a single or a bundle of fibers to carry the NIR radiation

toward the sample and back to the NIR detector is a smart choice to perform monitoring at

long distances. The optical fiber has the ability to conduct electromagnetic radiation between

two points through a nonlinear path with low losses at distances of 10-100 m.(66)

3.3.2 Chemometrics tools

Chemometric techniques are generally required to the treatment of spectral data for process

data analysis. Due to the capability of processing vast amounts of data, algorithms are able

to extract the multivariate information needed to better understand the process.

NIRS has the ability not only to identify a certain compound, by comparing it with those

existent in the “spectral library” (qualitative method), but also to determine the amount of

compound present in a sample (quantitative method). One of the most common multivariate

quantitative methods is the partial least squares (PLS), as it´s been widely used to predict

product properties from NIR spectra.(66–72) PLS estimates its model components (latent

variables) by maximising the covariance between two matrices: the system inputs and the

system response.

Principal component analysis (PCA) is a powerful tool for data compression and information

extraction. It provides an interpretable model of a data set by finding combinations of

variables that describe major trends in data. PCA model is the result of compressed data in

few components (principal components) in terms of the scores, which contain information on

how the samples relate to each other, and loadings, that contain information on how the

variables relate to each other.(73)


22

3.3.2 Process analytical technology

Accordingly to the ICH Q8, PAT is “a system for designing, analyzing, and controlling

manufacturing through timely measurements (i.e., during processing) of critical quality and

performance attributes of raw materials and in-process materials and processes with the goal

of ensuring final product quality”. The objective of PAT is to design and develop dynamic

manufacturing processes able to compensate for variability in both raw materials and

equipment. Thus, generating process and product quality information in real-time, operations

can be immediately adjusted. Through PAT any source of variability affecting a process is

identified, explained and managed. This relates to the key principle that quality cannot be

tested, but should rather be built into the product by design. Therefore, production under

QbD principles involves PAT strategies to reduce identified manufacturing risks linked to

product quality.(9)

The successfully establishment of PAT requires a careful selection of methods with the ability

to measure the CPP, preferably in an in-line or on-line mode – PAT monitoring tools. In this

context, NIR is one of the most flexible tools, since it´s a fast and non-sample-destructive

technique that measures samples without previous preparation in a real-time basis.(66)

Process monitoring within PAT should use samples, in their native state, ideally taken on-line

or in-line, through continuous measurements, in order to get a faithful representation of the

real process. Thus, NIRS is capable of describing process trajectory trend through the whole

spectra treated by multivariate analysis.(66)


23

Chapter 4

Materials and methods

4.1 The continuous manufacturing line

The equipment used to produce the granules was part of the PROMIS continuous tablet

manufacturing line (University of Eastern Finland, School of Pharmacy, Kuopio, Finland)

(Figure 7).

Figure 7 – Depiction of complete continuous line.A – powder loss in weight feeders; B – continuous mixer; C – roller compactor; D – screw conveyor; E – vacuum conveyor; F– tableting machine. (Adapted from (74))

This line is able to operate in complete line configuration, double blending/direct compression

and direct compression configuration. In complete line configuration, up to four powder loss-

in-weight (LIW) feeders feed API and excipients to the continuous mixer. From the

continuous mixer, a screw conveyer transfers the powder mixture to the RC. From the RC, a

vacuum conveyer transfers granules to the granule LIW feeder. This last feeder together with

LIW microfeeder (for lubricant) feed the second continuous mixer. At last, a vacuum

conveyer transfers granules to the tableting machine.


24

Table 1 – Description of the equipment used in complete continuous line.

Equipment Brand and

manufacturer Specifications

3 Loss-In-Weight powder feeders

K-Tron, K-ML-D5-KT20 500 g/h - 24 kg/h

Loss-In-Weight granule feeder

K-Tron, K-CL-24-KT24 300 g/h - 40 kg/h

Loss-In-Weight micro feeder

K-Tron, K-CL-SFS-MT12 32 g/h - 300 g/h

Modified Loss-In-Weight feeder for

lubricant and low dose API

K-Tron, K-CL-24-KT24 modified

X - 150 g/h

Two continuous blenders

Hosokawa, Modulomix 300 - 1450 rpm

Roller compactor

Hosokawa, Pharmapaktor L200/30P, with flake crusher FC 200

Screw speed: 0 - 53 rpm; Roll speed: 0 - 19 rpm; Roll pressure: 0 - 50 kN; Flake crusher: 32 - 313

rpm

Tableting machine PTK, PT-100 with

PISCon 96 000 tablets/h

Screw conveyer Entecon Spiral Screw Constant speed

Vacuum conveyer K-Tron, P10-BV-100-VE Constant speed

Vacuum conveyer Volkmann, VS200 Eco Constant speed

The optimum throughput tested in complete continuous line is 20 kg/h. It was realized that

the tableting is the rate limiting equipment in the line. Moreover, it is known that upper and

lower limits of the line are dependent on the formulation (flowability, cohesiveness and

adhesiveness of powder/granule mixture).

Each equipment unit of the line (except the tableting machine) is connected to the control-

PC. The Labview control software (University of Eastern Finland, School of Pharmacy,

Kuopio, Finland) controls, collects and stores the data from every equipment unit (Figure 9

and Figure 9). Data are also stored in the Datain Server (Kuava, Finland). Data can be

shown in graphs for monitoring purposes.


25

Figure 8 - Labview control software interface: monitoring of the powder mixture.

Figure 9 – Labview control software interface: monitoring of the compaction force.

4.2 Equipment

The equipment used to perform the experimental work is described below.

1) Production of granules

- Three LIW powder feeders K-Tron, K-ML-D5-KT20 (for excipients and API) (Figure 10);

- One LIW micro feeder K-Tron, K-CL-SFS-MT12 (for lubricant) (Figure 10);

- One continuous mixer, Hosokawa Modulomix (Figure 11);

- One Roller Compactor Hosokawa, Pharmapaktor L200/30P, with a flake crusher FC 200

(Figure 12, Figure 13, Figure 14).


26

The LIW powder feeder K-Tron has a feeding device with a hopper placed on a platform

scale. The weight of the feeding device and hopper is electronically tared. The feeder is filled

through its hopper and the raw materials are discharged from the hopper by twin-screws.

The resultant weight loss per unit of time is determined by weighing and control system. The

actual weight loss per unit of time is compared with the desired weight loss per unit of time.

Any difference between those two values induces a correction in the speed of the feeding

device.(75)

Figure 10 – Assembly of the four LIW feeders that feed the mixer.

continuous mixer, Hosokawa Modulomix, is a tubular blender with a horizontal cylinder and a

bladed shaft that rotates along its central axis. Powder is fed in one end and the impeller

moves the powder to the other end of the cylinder and out of the mixer. The homogeneity of

the mixture depends on the axial and radial movements of the particles in the mixer.(74)


27

Figure 11 – Continuous mixer disassembly. On the left: mixer bladed shaft; on the right: the mixer outlet.

The RC Hosokawa Pharmapaktor L200/30P, with a flake crusher FC 200, consists of: a

feeding unit with a vertically oriented screw; a compaction unit with two counter-rotating

rollers, horizontally oriented, with a ribbed surface and a fixed gap between the rollers and a

milling unit with a mesh screen of 1.25 mm.

The force transducer (Figure 14 A) measures and regulates the pressing force between the

rollers. If the maximum allowable pressing force is exceeded, the machine switches off. A

pre-stressing force must be adjusted in order to approximate the pressing force to the target

value (i.e, if the target pressing force is 35 kN, the pre-stressing force should be adjusted to

30 kN). Usually, the pre-stressing force (called actual value) is 5 kN lower than the target

pressing force.


28

Figure 12 – RC Hosokawa Pharmapaktor L200/30P. On the top left: RC prepared to operate; on the top right: detail of the vertical srew; bottom, on the bottom: detail of the milling screen.


29

Figure 13 – The RC rollers. On the top: the two disassembled rollers of 25 Kg each; on the bottom: top view of the two horizontal assembled rollers showing the fixed gap between them.

Figure 14 – The compaction system. A) The force transducer. B) The Hexagonal nut on the right roller arm. C) The roller shoulder. The force transducer measures the pressing force between the rollers. The hexagonal nut is tightened to increase the pre-pressing force or is loosened to reduce the pre-pressing force. The roller shoulder fixes the rollers

and is connected to the roller arms by eye bolts.


30

2) Monitoring of produced granules

- The NIR-sphere system (Specim RHNIR, Spectral Imaging Ltd, Oulu, Finland) (Figure 15).

consists of an integrated sphere with a tube inside, through which powder flows, surrounded

by six fibers placed in different angles. The fibers collect the signals to the Specim´s Spectral

Camera, which consists of an ImSpector N17E imaging spectrograph for the wavelength

region 900-1700 nm and a temperature stabilized InGaAs detector and a monochrome

camera. This system collects 100 spectra per second. The image resolution and rate are 320

pixel and 35 Hz, respectively. The light source is a halogen lamp.

Figure 15 – The NIR monitoring system. On the left: the NIR sphere with the 6 optic fibers; at the centre: the spectral camera and the laptop with control software; on the right: the tungsten lamp, used as light source.

- EyeconTM system (EyeconTM, Innopharma Labs, Dublin, Ireland) (Figure 16) is a particle

sizing system, based on direct imaging. It provides real-time particle size and shape

information, with a maximum material speed of 10 m/s and a measurement time of 2

seconds per image. The light source is 15 x high intensity/low energy LEDs (red, green and

blue). The measured size range is 50 μm – 3000 μm.


31

Figure 16 – Configuration of the NIR sphere and Eyecon to on-line monitoring of the granules. On the top left: the NIR sphere attached to the outlet of the RC and the glass frame through which Eyecon collects images of the granules; on

the top right and on the bottom: Eyecon capturing granules’ images.

3) Measurement of granules’ physical properties

- Malvern Mastersizer 2000 equipment (Figure 17) measures off-line the particle size

through the principle of light scattering. A laser beam impinges on the particles and they

scatter light at an angle that is inversely proportional to their size. The measured size range

is 0.02 μm – 2000 μm. It is able to measure wet (Hydro Unit) or dry materials (Scirocco Unit).


32

The light source is composed of a helium-neon laser (red light) and a solid-state light source

(blue light).

Figure 17 – Malvern Mastersizer 2000, with Scirocco Unit (on the left) and Hydro Unit (on the right). (Adapted from (76))

4.3 Raw materials

The powder mixture consisted of:

- Microcrystalline cellulose (MCC) (AVICEL PH 102, FMC, Cork, Ireland) is an excipient,

with the chemical name cellulose, the chemical formula (C6H10O5)n where n ≈ 220 and the

molecular weight 36 000 g/mol. In tablet manufacturing MCC is used as diluent and

disintegrant and leads itself to direct compression. MCC is a purified, partially depolymerised

cellulose, commercially available in different particle sizes and moisture grades. MCC PH

102 is suitable for direct compression. MCC is slightly soluble in sodium hydroxide solution,

and practically insoluble in water, dilute acids and most organic solvents.(77) MCC deforms

mainly by plastic deformation.(30)

- Lactose α-monohydrated (Lactochem, DOMO, The Netherlands) is an excipient, with the

chemical name O-β-D-Galactopyranosyl-(14)- α-D-glucopyranose monohydrate, the

chemical formula C12H22O11.H2O and the molecular weight 360.31 g/mol. Lactose α-

monohydrated is a natural disaccharide, obtained from milk and it is composed by one

galactose and one glucose moiety. In tablet manufacturing it is used as binder, diluent and

filler and is able to undergo direct compression. It is practically insoluble in chloroform,

ethanol and ether and soluble in water.(77) Lactose is mostly brittle and partially plastic.(30)

- Paracetamol (Xiamen Forever Green Source Biochem Tech. Co., Ltd, Xiamen, China) is

an API with analgesic and antipyretic therapeutic effects. Its chemical name is N-(4-

Hydroxyphenyl)-acetamide, with chemical formula C8H9NO2, and molecular weight of 151.16

g/mol. It is very slightly soluble in cold water, considerably more soluble in hot water; soluble


33

in methanol, ethanol; practically insoluble in petroleum ether, pentane, benzene. (Index

Merck) It is a highly brittle material that deforms mainly by elastic deformation.

- Magnesium stearate (Mg.Stearate) (Peter Greven, Venlo, The Netherlands) is an

excipient with the chemical name octadecanoic acid magnesium salt, the chemical formula

C36H70MgO4, and the molecular weight 591.24 g/mol. Mg.Stearate is used in tablet

manufacturing as lubricant. It flows poorly and is a cohesive powder. It is practically insoluble

in ethanol, ether and water; and slightly soluble in warm benzene and warm ethanol.(77)

The raw materials were characterized according to the physical properties described in Table

2 and Figure 18. The size, the bulk density and the tapped density were measured according

to the procedure described in Methods (points 4.3.5 and 4.3.6). The angle of repose was

measured in Hosokawa Micron Powder Characteristics Tester PT-X equipment. The Carr

index was determined by the formula:

(4.1)

Lactose α-monohydrated shows the highest bulk density and poor flow properties (Carr Index

24.9). Paracetamol has the worst flowability (Carr Index 28). MCC is the material that flows

the best (Carr Index 16.7).

Table 2 – Some physical properties of the raw materials.

Raw material

Amount used in

formulation (%)

Size (90%)/(μm)

Bulk density (g/ml)

Tapped density (g/ml)

True density

(Literature) (g/ml)

Angle of repose

Carr Index

Avicel PH 102

60 275.30 0.36 0.43 1.440(78) 39.3 16.7

Lactose Lactochem

24.5 148.62 0.49 0.65 1.545(77) 55.8 24.9

Paracetamol 15 353.6 0.33 0.46 1.293(77) 60.0 28.0

Mg Stearate 0.5 17.83 0.26 0.35 1.092(77) 53.3 25.2


34

Figure 18 – Particle size distribution (PSD) of raw materials.


35

4.4 Methods

4.4.1 Design of experiments

In order to determine the relationship between the possible CPP´s and granules’ CQAs, a

design of experiments (DoE) was performed in Modde (MKS, US). In this DoE, the following

three CPP’s were selected and their corresponding variation was: roll speed (from 3 rpm to 8

rpm), compression force (from 15 kN to 35 kN) and mill speed (from 50 rpm to 250 rpm)

(Error! Reference source not found.). The DoE yielded 13 experiments. The experiments

order assured that the three centre points (run 11, run 12 and run 13) took place in three

different moments of the experimental design: in the beginning, in the middle and in the end

of the DoE. Besides, the experiments’ run order also interchanged in terms of target

compression force, to avoid having two consecutive runs with the same compression force

(see detail in Table 4, given in Chapter 5).

Table 3 – Design of experiments obtained by Modde.

Experience Name Run Order Roll Speed

(rpm) Compression

Force (KN) Mill Speed (rpm)

N1 2 3 15 50 N2 3 3 35 250 N3 5 8 15 250 N4 10 8 35 50 N5 11 3 15 250 N6 12 3 35 50 N7 4 8 15 50 N8 8 8 25 150 N9 9 5.5 35 150

N10 7 5.5 25 250 N11 1 5.5 25 150 N12 13 5.5 25 150 N13 6 5.5 25 150


36

4.4.2 Preparation of the mixture

Each one of the four LIW feeders was manually filled in with its raw material. The feeders

were remotely controlled by the Labview control software.

After ensuring that the feeders had been filled in with enough amount of raw materials, the

mixing process was ready to start. The rotating hopper (that helps the powder flow

continuously from the feeders into the mixer) was turned on, followed by the feeders and the

mixer. Each feeder flow rate was adjusted according to both the mixture throughput (20kg/h)

and the proportion of the respective raw material in the mixture. The mixer speed was set to

1300 rpm. The first operating minutes of the mixer were considered a purge and that mixture

was wasted. After that, the mixture was collected to a big plastic bag attached to the outlet of

the mixer.

4.4.3 Granulation process

The RC was remotely controlled by the Labview control software. This software commanded

the switching on/off operations of the RC. However, the parameters’ settings were introduced

manually by the RC onsite control panel. The RC feeding was done manually throughout the

process. At the beginning of each run, the parameters were set gradually, following the

order: mill speed, roll speed and screw speed. On the other hand, when the process was

about to end, the process parameters were slowed down gradually by the opposite order:

screw speed, roll speed and mill speed. The process variables variation was followed by

graphs exhibited by the control software and the data (measured once per second) were

stored by the same software.

4.4.4 On-line monitoring of the properties of granules

The granules’s size was monitored on-line by NIRS and direct imaging (EyeconTM system).

The NIR-sphere was attached to the outlet of the RC, collecting the in-coming granules. The

NIR signals were sent to the detector and stored by the software. The granules then fell onto

a tilted ramp where they were exposed to the Eyecon LED’s and photographed (Figure 19).

The camera information was sent to the software and data were stored.


37

Figure 19 – On the left: the RC assembled and coupled with the monitoring tools; on the right: detail of how the both NIR and Eyecon systems were assembled to collect the samples.

4.4.5 Granules’ size measurement

During the run, four granules samples were collected (one sample at the beginning, two

samples in the middle and one sample at end of the process) onto a tray and then stored in a

labeled plastic bag for future measurement. These samples were weighed, one by one, and

given the total run time of each experiment, the respective throughput was computed.

The size of the granules was measured off-line by light scattering in the Scirocco unit of the

Malvern Mastersizer 2000.The diameters of the particles shown in the reports of the light

scattering method are displayed as d(0.1), d(0.5) and d(0.9). These results correspond to the

mass median diameter of the volume of distribution and are expressed in microns. For

instance, d(0.9) indicates that 90 % of the sample has a size smaller than that value,

whereas 10% has a larger size. To facilitate the understanding of this terminology, in next

chapters, d(0.9) was translated by 90% fraction of the granules’ size distribution.


38

4.4.6 Granules’s density measurement

The granules’ samples collected throughout the process were used to determine the bulk

density (BD) and the tapped density (TD). The BD was determined by reading the bulk

volume of the granules in a graduated cylinder and measuring the corresponding mass on an

analytical scale.

(4.2)

In order to determine the TD, the bulk volume inside the beaker was tapped by an Erweka

tapped density tester apparatus. The tapped volume was read and the corresponding mass

was measured on an analytical scale.

(4.3)

4.4.7 Exploratory data analysis

Data collected from NIR spectrometer were introduced in Matlab (The MathWorks Inc.

Natick, MA). The spectra obtained were pre-processed with Savitzky-Golay (first derivative,

first order, 15 points). Before PCA or PLS modelling all datasets were subjected to mean-

centring. After that, a PCA model was built in PLS toolbox software (Eigenvector Research,

Inc., Manson, WA).

4.4.8 Prediction models

The prediction of density and size of the granules was performed by two models: one built

upon NIRS data and the other built upon process variables and process responses. For the

first NIRS based model, the spectra were used to calibrate a model against the granule size.

Thus, the NIR data that matched with their respective granules’ sampling times was selected.

Those data were introduced in Matlab and in PLS toolbox software. The spectra were

preprocessed with Savitzky-Golay (first derivative, second order, 15 points) and mean

centring. The response data matrix consisted of information regarding 90% fraction of the


39

granules size distribution (SD) obtained off-line by light scattering. This matrix was

preprocessed with mean centring. Besides that, many other models were built and several

different preprocessing were tested. The number of latent variables was optimized with

cross-validation. The model error is given as the root mean square error of prediction

(RMSEP), and is relative to an unseen part of the data (the prediction set).

The second PLS model was built using Modde. The selected variables were: compression

force (or roll force), roll speed, mill speed and flow rate. The selected responses were: bulk

density, tapped density and granule size (90% fraction of the granules SD). The value of

each response was determined by the aforementioned methods. So, the values of variables

and their corresponding responses were introduced in Modde and a PLS model was built.

The capability of the model in predicting the granules’ physical properties was evaluated by

the analysis of variance (ANOVA). A similar strategy to compute model errors was followed

in Modde.


40

Chapter 5

Results and discussion

5.1 Exploratory data analysis

The data collected by the NIR spectrometer was introduced in Matlab, where the mean of

100 spectra per second was calculated, in order to have 1 average spectrum per second.

Then the spectra were plotted against the wavelength. The spectral region 927 nm–1085 nm

was cut off, because they exhibit essentially noise and didn´t have any relevant information

(Figure 20, top). The spectra were preprocessed with derivative (Savitzky-Golay (SavGol)

algorithm), to remove irrelevant baseline drift. The SavGol was used with the first derivative.

This subtraction removes the same signal between the two variables and leaves the part of

signal that is different, removing any offset from the sample. Derivatives tend to accentuate

noise, that´s why SavGol algorithm smoothes the data. Smoothing is used to remove high-

frequency noise from samples and it´s an operation that acts separately on each row of the

data matrix. Smoothing fits single polynomials to windows around each point in the spectrum.

Then the size of the window (filter width) and the order of the polynomial are selected. In this

case, the filter width has 15 points and the polynomial order is 1 (Figure 20, bottom). At last,

the spectra were mean-centered, in order to build the PCA model. Mean-centring calculates

the mean of each column in the data matrix and subtracts it from the column.

The following discussion will make use of the first run (run N1) of the DoE as an example for

considerations on this subject over the other runs. As it can be seen, the region that reveals

more information goes from around 1320 nm to 1540 nm and from around 1580 nm until the

end of the spectra.


41

Figure 20 – The NIR spectra for the run N1. On the top: raw spectra; on the bottom: spectra pre-processed with first derivative.

After preprocessing the spectra, a PCA model was built. Figure 21 shows the score plot with

the PC 1, with 64.99% of variance captured, versus PC 2, with 34.05% of variance captured.

The PC 1, the first component, describes the direction of the major variation in the data set,

which is the greatest axis of the ellipse. The scores were colored according to the Hotelling’s

T2 statistic and labeled according to time. So, it can be seen a clear trajectory of the

granulation process from its beginning (score 1, outside the ellipse, on the bottom, at right)

until the end (score 2005, inside the ellipse). Following the scores along the PC 1 axis, the

process starts with positive scores, out of the ellipse, then approaches the centre of the

model with some negative scores deviated from that central cluster (intermediate process

stage), and finally returns to the positive domain, where some of the last scores remain

outside the ellipse (end of the process). The scores retained on the second and on the forth

quadrants are negatively correlated, and refer to different stages in the granulation process.

The Hotelling’s T2 statistic is the sum of normalized squared scores and a measure of the

variation in each sample inside the PCA model. The limit defines the ellipse seen on the

plane within which the data are projected (assuming that data is normally distributed which is

not the case). The samples colored by the Hotelling’s T2 value represent their distance to the

centre of the model. So, the blue cluster of scores inside the ellipse is the core that better

describes the model. The scores painted in yellow, orange or red, outside the ellipse, are far

away from the centre of the model, which means that the model is not able to explain these


42

samples. As it will be explained later, these last scores are related to specific times of the

process, where the variables varied more.

-3 -2 -1 0 1 2 3 4

x 10-3

-3

-2

-1

0

1

2

3

4x 10

-3

Scores on PC 1 (64.99%)

Score

s o

n P

C 2

(34.0

5%

)

1 2

4 5

7 9 11 12

19

25

31

38

46

47 50

51

52 55

57

68

69

76

78

84 86

87

88

91 134

144

155 157

165 171

198

199

236

240

246 260

268

271

272

273

275

287

298

358

368 371

383

384

389

394

407

415 418

422

444

450

471

472

478

506

533 549

554

558

566

567

624

649

650 653

655

681

705 709

738

739

742

755

773

774

815

817

830

832

833 838

882

893

938

948

984

1001

1002

1024

1028

1030 1065

1073

1098 1157

1188

1200

1203

1288

1303

1309

1327

1334

1341

1356

1361

1383

1387

1389 1420

1448

1453

1472

1485

1509 1514

1516 1583

1585

1594

1610

1661 1665

1681 1682

1683

1735

1739

1764

1840

1856

1863

1926

1955 1956

1962

1966 1969

1985

1994 1996

2002 2004

2005

Samples/Scores Plot

5

10

15

20

25

30

35

40

Figure 21 – The score plot for run N1, depicting PC 1 versus PC 2. The total variance captured by both components is 99.04%. The ellipse’s border represents the 95% confidence limit.

The error matrix (E) is obtained by the difference between the original data and the model

predictions (residuals). Q is the sum of squares of each row in matrix E. The Q statistic

points out how well each sample conforms to the PCA model. In Figure 22 the plot

represents the Hotelling’s T2 statistic versus the Q statistic. The dotted lines are their

respective 95% confidence limit. The intersection of the confidence limits defines the region

beyond which the scores distributed over this area are the most distant from the model and

behave like possible outliers. Likewise the score plot, the statistic plot reveals the same

samples as possible outliers: 1-12, 17, 19, 271, 893, 1485 and 1926.


43

0 5 10 15 20 25 30 35 40 450

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8x 10

-6

Hotelling T 2̂ (99.03%)

Q R

esid

uals

(0.9

7%

)

1 2 3 4 5 6 7 8 9 10 11 12 17 19 20 34 37 38 40 41 45 57 61 63 65 68 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 190 191 195 196 197 198 199 200 202 205 206 207 208 209 210 211 212 214 215 216 218 219 220 221 222 223 224 225 226 227 228 229 230 231 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 259 260 261 262 263 264 265 266 267 268 269 270

271

272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892

893

894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484

1485

1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925

1926

1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Samples/Scores Plot

Figure 22 – The statistic plot for run N1, depicting Hotelling’s T2 statistic versus Q statistic. The dotted lines represent the 95% confidence limit for both Hotelling’s T2 statistic (horizontal dotted line) and Q statistic (vertical dotted line).


44

5.2 Analysis of process variables

Table 4 shows the experiments’ order performed. The screw speed was adjusted to the roller

speed target value, in order to get the target compression force. The lag time ended when all

variables (roll speed, screw speed, mill speed and compression force) stabilized. Even

though the compression force has not kept constant, it was considered “stable” when the

mean variation around the target compression force was about 10%.

Table 4 – Detail of the DoE performed, showing the order by which the experiments were done. The screw speed was adjusted to the roll speed in order to obtain the target compression force. The lag time was the time considered for the

variables to stabilize.

Experience

identification

Variables

Experience time Target variables and Target values

Adjusted variable

Experience

name

Run order

Roll speed (rpm)

Compression force (kN)

Mill speed (rpm)

Screw speed (rpm)

Lag time (min)

Total run time

aprox (min)

N1 2 3 15 50 7.5 3.5 33 N2 3 3 35 250 10.5 3 32 N3 5 8 15 250 23.5 2 13 N4 10 8 35 50 31.5 3 14 N5 11 3 15 250 7.5 4 33 N6 12 3 35 50 10.5 4 33 N7 4 8 15 50 23.5 2 13 N8 8 8 25 150 28.5 3 14 N9 9 5.5 35 150 20.5 2 18

N10 7 5.5 25 250 18.5 2 18 N11 1 5.5 25 150 18.5 1.5 16 N12 13 5.5 25 150 17.5 3 19 N13 6 5.5 25 150 18.5 2 18

The variation of the process variables is depicted in Figures 23-28. The graphs selected are

representative of the process behaviour within each target compression force. The graphs

describe the process variables behaviour when the compression force is set to 15 kN

(Figures 23-24), 25 kN (Figures 25-26) and 35 kN (Figures 27-28). The roll force mean line

exhibited in Figures 23, 25 and 27 smooths the fluctuations in the data to show a trend more

clearly. Each point of the line is the mean of the roll force values every 50 seconds.

Generally, the mill speed was the first variable to stabilized (between 15 s and 24 s after the

process onset), followed by roll speed (between 21 s and 28 s after the process onset),

screw speed (between 110 s and 140 s after the process onset) and compression force. In


45

fact, the compression force was the variable that most varied in all runs, regardless of its

target value. The applied compression force is expressed in kN/cm and is the result of the

interaction of both screw speed and roll speed over the materials, within a fixed gap width

between the rollers. However, this force merely represents the pressure within the hydraulic

system and so it´s not a precise measure of the force applied onto the powder. Moreover, as

the gap width is fixed, the powder flow rate drawn into the gap is not constant, which results

in the variation of the force applied to the powder.

The compression force variation can be associated with the “outliers” identified in Figures 21-

22. Actually, in Figure 24, the compression force varied significantly up to 153 s. This

variation was exposed by the outliers numbered from 1 to 19. In addition, the compression

force peaks that occurred at 265 s (18 kN), 893 s (11 kN), 1940 s (17 kN) can be linked,

respectively, to the outliers labelled with 271/272, 893 and 1926.

The screw speed, roll speed and mill speed stabilized at about the first 2 min of the process

(Figure 23) and after that they were kept constant until the process finish. On the other hand,

the compression force took more time to approach the target value, though it kept fluctuating

throughout the process run time.

Therefore, these results demonstrate that the compression force is one of the CPP’s that

most affects the granulation process. So it is expected that the compression force fluctuation

will affect the granules’ physical properties.

.


46

Figure 23 – Process Variables Variation for Run N1. The target compression force was the lowest of the DoE (15 kN). After about 2,5 min all process variables reached their target values, except the compression force, which fluctuation

lasted until the end of the run. The Mill Speed stabilized at 24 s, Roll Speed at 28 s and Screw Speed at 140 s.

Figure 24 – Compression force fluctuation showing the roll force mean line. The mean value stood below the target, at 13 kN-14 kN. The target of 15 kN was reached at about 154 s. After that the compression force varied between 10 kN

(645 s, 1030 s,…) and 20 kN (1281 s).


47

Figure 25 – Process Variables Variation for Run N12. The target compression force was the highest of the DoE (25 kN). After about 2,2 min all process variables reached their target values, except the compression force, which fluctuation

lasted until the end of the run. The Mill Speed stabilized at 24 s, Roll Speed at 21 s and Screw Speed at 110 s.

Figure 26 – Compression force fluctuation showing the roll force mean line. The mean value stood below the target, at 22 kN-24 kN. The target of 25 kN was reached at about 131 s. After that the compression force varied between 17 kN

(285 s, 992 s,…) and 35 kN (572 s).


48

Figure 27 – Process Variables Variation for Run N4. The target compression force was the intermediate value of the DoE (35 kN). After about 2 min all process variables reached their target values, except the compression force, which

fluctuation lasted until the end of the run. The Mill Speed stabilized at 15 s, Roll Speed at 24 s and Screw Speed at 118s.

Figure 28 – Compression force fluctuation showing the roll force mean line. The mean value stood close to the target, at 34 kN-36 kN. The target of 35 kN was reached at about 120 s. After that the compression force varied between 27 kN

(276 s, 556 s,…) and 44 kN (388 s).


49

5.3 Analysis of the granules

Table 5 displays the values of the granules’ density (BD and TD). The table was divided in

three blocks to make the interpretation easier. In fact, sorting the densities by target

compression force, it is realized that when the compression force is higher (35 kN), the

density of the granules is also higher (0.62 g/ml – 0.64 g/ml); when the compression force is

lower (15 kN), the density of the granules is also lower (0.54 g/ml – 0.55 g/ml). The

intermediate compression force value (25 kN) gave rise to granules with and intermediate

density (0.58 g/ml – 0.60 g/ml). The variation on density seems not to be so affected by

single changes on roll speed, screw speed or mill speed. Thus, the compression force is the

variable that most influences the density of the granules.

Once almost all of the granules produced were collected, it was possible to calculate the

process throughput. As the Table 5 shows the process throughput depends upon the screw

speed: when the screw speed increases, the process throughput increases (see run 12 and

run 13).

Table 5 – The granules’ densities determined for each run and the process throughput. The table was split in three blocks, according to the target compression force (from the highest compression force, on the top, to the lowest

compression force, on the bottom).

Run Bulk

density (g/ml)


Process throughput

(kg/h)

Roll speed (rpm)

Screw speed (rpm)

Target compression

force (kN)

Mill speed (rpm)

2 0.63 0.72 9.24 3 10.5 35 250 4 0.64 0.74 20 8 31.5 35 50 6 0.62 0.73 9.12 3 10.5 35 50 9 0.62 0.72 17.16 5.5 20.5 35 150

8 0.59 0.68 20.17 8 28.5 25 150 10 0.59 0.69 15.52 5.5 18.5 25 250 11 0.58 0.67 15.52 5.5 18.5 25 150 12 0.58 0.69 14.37 5.5 17.5 25 150 13 0.60 0.70 15.55 5.5 18.5 25 150

1 0.55 0.64 6.77 3 7.5 15 50 3 0.55 0.65 18.4 8 23.5 15 250 5 0.54 0.64 6.85 3 7.5 15 250 7 0.54 0.65 18.3 8 23.5 15 50


50

The size of the granules collected during each run was measured by light scattering. This

method was considered the gold standard for subsequent methods’ comparisons. Figures

29-31 exhibit the PSD for each run. Regardless of its distribution, the size profile is quite

similar for this set of runs. Runs where the target compression force was higher (35 kN), runs

2, 4, 6 and 9, correspond to higher granules’ size. Runs where the target compression force

was lower (15 kN), runs 1, 3, 5 and 7, show lower granules’ size. Runs with the intermediate

compression force (25 kN), runs 8, 10, 11, 12 and 13 revealed intermediate granules’ sizes.

The effect of the roll speed and the mill speed on granules’ size is not clear (Figure 29).

Keeping the target compression force and the mill speed constant, the size of granules

decreases when roll speed increases (see run 6 and run 4, run 13 and run 8, run 5 and run

3). This result may be because the powder stays longer in the nip zone (when the roll speed

is lower), where particles rearrange and start to deform as a consequence of the pre-

compression stage. Once the particles are closer together for more time, they feel the pre-

compression force for longer time and that´s why their compressed ribbon is denser and the

correspondent granules are harder and larger. However, in the case of runs 1 and 7, that

logic reasoning is true for the first sample, but in the last three samples the size decreases

as the roll speed decreases.

Concerning the mill speed, now keeping the target compression force and roll speed

constant (see run 6 and run 2, run 13 and run 10, run 1 and run 5), the size of the granules

decreases when the mill speed increases. As the rotor speed increases the impact force of

the granules against the sieve is higher which results in smaller granules. However, in the

case of the run 13 and run 10, that logic reasoning is true for the first three samples, but in

last sample the size increases as the mill speed increases.

Therefore, the compression force is the variable the most affects the size of the granules

produced by roller compaction.


51

Figure 29 – 90 % fraction of the granules size distribution (SD) obtained off-line by light scattering. The runs are ordered from the highest target compression force to the lowest target compression force.


52



53


The results from the on-line method (Eyecon) used to monitor the size of granules are

represented in Figures 32-34. The run 5 is missing because an error prevented the collection

of data. The size profiles are identical within each run, despite of their distribution.

In Figure 32 the runs with the lowest target compression force are the ones with the highest

granules’ size. The run 6 has the lower granules’ size. The run 4 and the run 9 are among

the ones that show lower sizes too. The runs with intermediate target compression force

appear in middle of the graph, with intermediate sizes. The run 2 is the one that looks close

to its real size, measured by light scattering.


54

Figure 32 – 90 % fraction of the granules size distribution (SD) obtained on-line by direct imaging. The size of each sampling point corresponds to the average of sizes captured by Eyecon during the sampling time settled for each run.

The runs are ordered from the highest target compression force to the lowest target compression force.


55




56



The Eyecon results show a great deviation from their mean value. This deviation is

represented by RSD value. Thus, the granules’ sizes obtained with the on-line method

exhibit a great RSD when considering 90% of granules (Figure 35). In 50% fraction of the

granules SD, the RSD is higher for runs 2, 3 and 7 (Figure 36). In 10% fraction of the

granules SD, the RSD is lower for runs 1, 2, 6 and 7, but run 3 keeps its high RSD (Figure

37).


57

Figure 35 – The RSD determined for 90% fraction of the granules size distribution (SD) obtained by Eyecon.


58



59


The granules’ size obtained on-line by Eyecon were compared to the gold standard off-line

method. Each sample was individually compared within the both methods by determining the

square error (SE) (Figures 38-40). In Figure 38 it can be seen that run 2 has the lowest SE,

in accordance with the closest results to the off-line method, and the run 6 has the highest

SE, once its sizes are greatly deviated from the true ones. Run 3 has the second highest SE,

which corroborates with the high sizes captured by Eyecon during this run.


60

Figure 38 – Comparison between off-line and on-line methods for granules’ size measurement. Square Error between the both methods regarding 90% fraction of the granules size distribution (SD).


61



62


The direct imaging method demonstrated a great deviation in granules’ size, comparing to

the ones measured by light scattering. These results might be due to the inappropriate flow

rate of the granules that reached the Eyecon camera and its image acquiring speed. The

flow rate was inconstant, with periods of high flow rate interchanging with periods of a very

low flow rate. Besides, the direction of the granules fall did not always favor the camera, and

sometimes it was not possible the collect focused images. These acquired blurred images

might have contributed for the wrong size measurements. As the images were captured each

2 seconds, the information lost between the collection times might have improved the size

measurements.

So, the on-line method did not reveal to be accurate enough for granule’s size monitoring.


63

5.4 Predictive models

Two predictive models were built in view of estimating the granules’ physical properties. The

first PLS model was built with the NIR data collected throughout the process to predict the

granules’ size.

A summary of the performance of the first model is displayed in Table 6. Different models

were built testing several pre-processings and applying a cross-validation method. The best

model obtained was chosen by minimizing both the number of latent variables and the RMSE

of cross-validation.

The chosen model is simple, with only one LV. The pre-processing applied differed from the

one used in the all spectra for the PCA model by selecting a polynomial of second order in

the Savitzky-Golay filter. The performance of the model was assessed by an internal

validation (cross-validation) and an external validation. The model was built with NIR data

from runs 1, 3, 4, 5, 7, 8, 10, 11, 12 and 13. Spectral data from runs 2, 6 and 9 had some

sort of problems regarding data processing and that is why those were not used in the

model.

The 40 samples (because each run had 4 granules’ samples) of the model were divided in 2

groups: one with 32 samples was used to calibrate the model and the other with 8 samples

was used to test the model. The calibration set underwent an internal validation method,

cross-validation (contiguous-blocks). The calibration set was split into 8 blocks, where 1

block was used to “test” the subset and the remaining blocks were used to build a subset

calibration model. This internal validation provides an estimate of the model’s prediction

performance, which can be assessed by the RMSECV (Root Mean Square Error of Cross

Validation) and the coefficient of determination for cross validation (R2(cv)).

In the external validation, the calibration model was tested using data that was not used to

build the model: the test set with 8 samples. This method provides a reasonable assessment

of the model’s prediction performance on new other samples. The external validation is

evaluated by the RMSEP and the coefficient of determination for Prediction (R2(P)).


64

Table 6 – Description of the PLS model to predict granules’ size based on NIR spectra.

The calibration model has a high RMSECV (146.3 μm) and a poor R2 of CV (0.14), both

foreseeing that the model will not work well when predicting the granules’ size. The

prediction model has also a high RMSEP (50.5 μm), even though with a lower magnitude

order than RMSECV, and a very low R2 of prediction (0.19), thus the model is not acceptable

for predicting the granules’ size.

The figures 41-42 show the scores of the calibration set and the test set. Only one sample of

the test set is out of the 95% confidence limit and cannot be described by the model, so it is

probably an outlier.

10 20 30 40 50 60 70 80

-0.04

-0.02

0

0.02

0.04

0.06

0.08

Sample

Score

s o

n L

V 1

(98.7

3%

)

Samples/Scores Plot of c & Test,

Scores on LV 1 (98.73%)

Calibration

Test

95% Confidence Level

Figure 41 – The scores of the calibration set and the test set used in the PLS prediction model.

Spectral range

Pre-processing Cross-

validation method

LV RMSECV

(μm) R2

(cv) RMSEP

(μm) R2

(P)

950-1650 nm

1st derivative (order 2; 15

points); mean centring

contiguous blocks (8)

1 146.3 0.14 50.5 0.19


65

0 1 2 3 4 5 6 7 80

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

x 10-5

Hotelling T 2̂ (98.73%)

Q R

esid

uals

(1.2

7%

)


Q Residuals (1.27%)

Calibration

Test


Figure 42 – The statistic plot for the PLS model, depicting Hotelling’s T2 statistic versus Q statistic. The dotted blue lines represent the 95% confidence limit for both Hotelling’s T2 statistic (horizontal dotted line) and Q statistic (vertical

dotted line). Both the scores of the calibration set and the test set are represented.

Figure 43 exhibits the correlation between the observed and the predicted data. It is seen the

great dispersion over the regression line, which explains the weak link between the

experimental and the predicted data. The green line shows the ideal regression line (a strong

association between observed and predicted data) and the red line shows the real regression

line.

The model’s lack of accuracy in predicting the granules’ sizes is due to the inadequate

quality of NIR data. Despite the missing three runs, which would contribute to give more

information on the process, the interaction between the sample and the radiation was not

ideal. The granules flow rate into the glass tube might influence the ability to gather

information with better quality. Maybe the diameter of the glass tube should be reduced to

provide a better flow rate. Moreover, the spectral range from 950 nm to 1680 nm used is too

narrow, even if searching only the physical information. A wider spectral range, from 1000

nm to 2500 nm would benefit the collection of better process information.


66

500 600 700 800 900 1000 1100 1200650

700

750

800

850

900

950

1000

1050

1100

Y Measured 1

Y P

redic

ted 1


Y Predicted 1

Calibration

Test

1:1

fit


Figure 43 – The observed versus predicted plot for the PLS model.

The second PLS model was built with the values of the process variables (factors) and their

respective responses. The variables chosen were the same as in the DoE’s: the roll speed,

the mill speed and the compression force. This second approach used the three process

responses (the process results) – BD, TD and GS – to build three PLS models capable of

predicting each of those responses (

Table 7).


67

Table 7 – The overview of the factors and the responses used to build the second PLS model.

Factors Responses

Experience name Run order Screw speed

(rpm) Roll

speed (rpm)

Target compression

force (kN) Mill speed

(rpm) Bulk

density (g/ml)


Granules size/90%

fraction/last sample (μm)

N1 2 7.5 3 15 50 0.55 0.64 699.67 N2 3 10.5 3 35 250 0.63 0.72 1009.40 N3 5 23.5 8 15 250 0.55 0.65 655.53 N4 10 31.5 8 35 50 0.64 0.74 1120.33 N5 11 7.5 3 15 250 0.54 0.64 648.58 N6 12 10.5 3 35 50 0.62 0.73 1057.05 N7 4 23.5 8 15 50 0.54 0.65 729.22 N8 8 28.5 8 25 150 0.59 0.68 881.36 N9 9 20.5 5.5 35 150 0.62 0.72 1028.87 N10 7 18.5 5.5 25 250 0.59 0.69 890.76 N11 1 18.5 5.5 25 150 0.58 0.67 854.89 N12 13 17.5 5.5 25 150 0.58 0.69 996.03 N13 6 18.5 5.5 25 150 0.60 0.70 860.53

The plots depicted in Figure 44 show the variation of the responses so that the raw data may

be seen as a whole. The values of the responses are plotted against the experimental runs

and also the variation in the response for replicated experiments is displayed. The replicates

are experiments with the same factor values plus or minus the replicate tolerance of 10%.

Ideally, the variability of the repeated experiments should be much less than the overall

variability. The variability of the replicates in BD and TD models is low, but in the case of the

GS the replicates 11 and 12 show a greater variability. Nevertheless, this variation is not

higher than the overall variation for the responses (see reproducibility in Figure 45).

When a set point is defined for the responses, the plot shows the upper bound and the lower

bound (dotted red line), as well as the target value (dotted black line). Therefore, the

experiment points should not be above the upper limit or under the lower limit. In this case,

however, those limits are merely indicative, so it is not important whether the points are

above or under the marked limits.


68

Figure 44 – The values of the responses plot against experimental runs. The replicated experiments are shown in blue on the same replicate index and the other experiments in green.

The Table 8 exhibits the analysis of variance (ANOVA) regarding the three models, as well

as their regression equations. The ANOVA splits the total variation of the response (sum of

squares corrected for the mean) into a component due to the regression model and a

component due to the residuals. The residual sum of squares is further divided into Replicate

Error and Model Error. The model is evaluated by comparing the Mean Square (MS) Model

Error to the MS Replicate Error.

Only two components were used to build the model, so its absence of complexity favors the

prediction performance.

The R2 denotes the fraction of the response that is explained by the model. Once it is an

optimistic indicator of the model prediction ability, R2 could be considered an upper bound of

the estimate for how well the model predicts the outcomes of new experiments. In these

three models the R2 is high varying from 0.93 (for GS) to 0.96 (for BD), which means a strong

association between the variables and the predicted response.

The Q2 estimates the predictive ability of the model and is used to validate the regression

models. A Q2 value higher than 0.70 indicates a very good predictive ability and a very low

prediction error on new samples. The computed Q2 for the three models ranged from 0.89

(GS) to 0.91 (BD), which suggests a model with good prediction ability.

The p-value for the regression model should be inferior to 0.05 and the three models satisfy

this criterion. The p-value for the lack of fit (the model error, i.e, the non modeled part of the

model) should be greater than 0.05, which is also true for all of the three models. The model


69

is also evaluated by the MS for the model error and the MS for the replicate error. The

prediction performance of the model is good if the MS for model error is lower than the MS

for replicate error. All of the three models satisfy this criterion.

Table 8 – Summary of the PLS Model built with the process parameters. BD, Bulk Density; TD, Tapped Density; GS, Granule Size; RS, Roll Speed; CF, Compression Force; MS, Mill Speed. MS, Mean Square; p, probability.

The Figure 45 presents the summary statistics for each response in four parameters: R2, Q2,

model validity and reproducibility, where 1 is the perfect outcome.

The first two columns, R2, Q2, should be close in size, which is true for the three models. The

model validity usually indicates statistically significant model problems, such as the presence

of outliers, when its value is inferior to 0.25. However if the replicates are identical (the pure

error is rather small) the model validity can be very low, even though the model is good and

complete. As Figure 45 shows, in the case of the three models, the model validity is higher

than 0.25.

The reproducibility is the variation of the replicates compared to overall variability and should

be greater than 0.5. The three models also satisfy this criterion.

Response PLS Model equation Components R2

Q2

p

(Regression) p (Lack of Fit)

MS (variance) Model error

(μm2

)

MS (variance) Replicate error

(μm2

)

BD BD=0.0034RS+0.0343C

F+0.59 2 0.96 0.91 < 0.001 0.898 3.9x10

-5 1.3 x10

-4

TD TD=0.0306CF-0.0051MS+0.69

2 0.95 0.90 < 0.001 0.843 5.1 x10-5

1.3 x10-4

GS GS=139.383-

27.8055+894.379 2 0.93 0.89 < 0.001 0.918 1342.4 4912.2


70

Figure 45 – Summary of the basic model statistics for every response: R2, Q2, model validity and reproducibility.

The significance of the regression model coefficients is represented in Figure 46. A model

term is deemed to be significant when it exhibits a large distance from y=0 (either positive or

negative) and has an uncertainty level that does not extend across y=0.

All three models included the three coefficients concerning the three variables: roll speed,

mill speed and compression force. However, in BD model, the mill speed was considered

non significant and in both TD model and in GS model, the roll speed was considered non

significant. Thus, only the model terms displayed in the coefficient plot were considered

significant for the prediction performance of the model. The compression force coefficient is

the most significant model term for each model.


71

Figure 46 – The coefficient plot depicting the model terms for each response. RolSp, Roll Speed; ComFc, Compression Force; MilSp, Mill Speed.

The observed values versus predicted values for each model plot are displayed in Figure 47.

All three plots show their points close to the straight line, which means a strong match

between observed and predicted data. These plots indicate the capacity of the model in

predicting the new values. The best model is the BD with Q2=0.91, followed by TD, with

Q2=0.90 and finally GS with Q2=0.89.

Figure 47 – The observed versus predicted plot for each model. Bulk Density: R2=0.96, Q2=0.91; Tapped Density: R2=0.95, Q2=0.90 Granule Size: R2=0.93, Q2=0.89.


72

The plots represented in Figure 48 illustrate the relation between scores for the factors (t1)

and the responses (u1) for the first component. Within each model, it can be seen the

presence of three groups: scores 1, 3 and 5 (with compression force equal to 15 kN), scores

8, 10, 11 and 12 (with compression force equal to 25 kN) and scores 2, 4 and 6 (with

compression force equal to 35 kN). So the score plot can clearly reveal the existence of three

clusters within each model. However, the scores 7, 8 and 9 are not well described by the

model, as the plot shows.

Figure 48 – The score plot for each model, depicting the relationship between the scores (t1) and the responses (u1).

The loading plot correspondent to each of the three models is represented in Figure 49. This

plot explains how the responses vary in relation to each other and which ones provide similar

information.

Similarly to the information gathered in the Figure 46, for the BD model the compression

force is the variable that most affects the BD of the granules in a positive way while the roller

speed affects less the BD of granules and in a negative way. For the TD model the

compression force is the variable that most affects the TD of the granules in a positive way

while the mill speed affects less the BD of granules and in a negative way. And finally, for the

GS the compression force is the variable that most affects the GS of the granules in a

positive way while the mill speed affects less the BD of granules and in a negative way.


73

Figure 49 – The loading plot for each model depicting the first LV against the second LV. BD= Bulk Density; TD= Tapped Density; GrS= Granule Size; RolSp= Roll Speed; MilSp= Mill Speed; ComFc= Compression Force.

The Figure 50 represents the standardized residuals from each model. The standardized

residuals are the raw residuals (difference between the observed and the predicted values)

divided by the residual standard deviation. The residuals account for the non modeled part of

the model. The residuals have to demonstrate a normal distribution in order to ground the

predictive ability of the model. If all the points of the residual plot are on a straight line on the

diagonal, the residuals are normally distributed noise. For the BD and the TD models,

residuals tend to be on a straight line, so they are considered normally distributed. For the

GS model, a diagonal can also be drafted through the points.

Figure 50 – The residuals of each response versus the normal probability of the distribution.


74

Chapter 6

Conclusions

In this experimental work, the CPP’s proposed and their respective variations were: roll

speed, from 3 to 8 rpm; mill speed, from 50 to 250 rpm; and compression force, from 15 kN

to 35 kN. The granules’ CQA´s measured were: bulk density, tapped density and granule’s

size.

The NIR sphere, as a PAT tool, looked inside the process and unveiled not only the process

trajectory, but also the sources of variability that influenced the process. Thus, the PCA

model built with NIR data revealed that the variance of the scores far from the centre of the

model was due to abnormal variation in the compression force, during the process.

Although the roll speed, the screw speed and the mill speed were kept constant few seconds

after the process onset, the compression force fluctuated throughout the process run time.

The compression force variation was due to the unstable flow rate coming through the fixed

gap between the rollers. Since this fixed gap could not be adjusted to the flow rate variation,

the compression force exerted by the rollers over the powder varied significantly.

The bulk and tapped density of the granules were strongly influenced by the compression

force. The higher the compression force achieved, the higher the bulk and tapped density of

the granules; and the lower the compression force achieved the lower the bulk and the

tapped density of the granules. Likewise, for granules’ size, the compression force was the

variable that most affected this response. The granules’ size increased when compression

force increased and the granules’ size decreased as the compression force decreased.

The granules’ size on-line monitoring with Eyecon showed a great deviation from its off-line

measurement and therefore did not provide accurate results. The Eyecon system was

strongly influenced by the granules flow rate. As it was not constant, Eyecon’s camera was

not able to capture a clear real-time image of the granules’s shape and therefore the size

was wrongly determined.

Two PLS prediction models were built in order to predict the size and the density of granules.

The first PLS model, built upon NIR data, had a high RMSEP (50.54 μm) and a very low R2 of

Prediction (0.19), thus the model was not acceptable for predicting the granules’ size. Data

collected through the NIR-sphere was sensitive to granules flow rate variations, and that


75

affected the quality of spectral data obtained. In addition, the spectral range used (950 nm –

1700 nm) was a drawback concerning information on the granules’s properties.

The second PLS model, built upon the values of the process parameters and the process

responses predicted three responses: the bulk and the tapped density of the granules and

the size of the granules. In these three models the R2 was high, yielding 0.93 for GS, 0.95 for

TD and 0.96 for BD. The Q2 was also high: 0.89 for GS, 0.90 for TD and 0.91 for BD.

Together, both values account for three models with good prediction ability.

Besides, these last three models pointed out that the compression force was the variable that

most influenced each model response. While the compression force exerted a strong and a

positive influence on the responses, the roll speed and the mill speed, on the opposite,

performed a slight and negative influence over the responses.

Thus the compression force was the variable that should be kept under control in order to

produce granules with the desired density and size.


76

Future perspectives

One of the main variables that affects the granulation process is the formulation. In this

experimental work, the formulation was kept constant in terms of its composition and the

proportion of raw materials. So, in a first trial varying the composition by exchanging one

component of each kind and in a second trial keeping the composition and varying the

amount of one or two components would provide more information regarding the influence of

the formulation on the final product.

The flow rate had a strong impact on the compression force variation and on the quality of

NIR data. Thus, in the next DoE, including the flow rate as a CPP would yield a better

understanding of how this variable affects the process responses. Therefore, taking the flow

rate into consideration as an input variable, might improve the prediction models, especially

in the case of the prediction of the granule size.

Moreover, in what concerns NIR spectroscopy, improving the interaction between the sample

and the radiation, as well as providing a wider spectral range, would yield more and better

NIR data and so better prediction models. As a result, the changes in the granules’ density

could also be followed by the shift in the NIR spectra baseline.(46)

The ribbon, as an intermediate product of the granulation process, should be evaluated

mostly by its porosity. A NIR probe could be attached to the ribbon sampling window so that

data could be collected in order to follow the porosity variation by the shift in the NIR spectra

baseline.(46)(72)

Finally, the granules produced should be compressed at different compression forces and

the tablets produced should be characterized by their tensile strength and friability. This

study should clarify which granules yield the higher tensile strength tablets with lower

friability. This information should be used in the granules prediction models in order to adjust

the CPP’s for the production of the desired granules, i.e, with the intended CQA’s (the most

suitable for future compression into tablets).


77

References

1. Lee SL, O’Connor TF, Yang X, Cruz CN, Chatterjee S, Madurawe RD, et al.

Modernizing Pharmaceutical Manufacturing: from Batch to Continuous Production. J

Pharm Innov. 2015;191–9.

2. Boukouvala F, Niotis V, Ramachandran R, Muzzio FJ, Ierapetritou MG. An integrated

approach for dynamic flowsheet modeling and sensitivity analysis of a continuous

tablet manufacturing process. Comput Chem Eng [Internet]. 2012;42:30–47. Available

from: http://dx.doi.org/10.1016/j.compchemeng.2012.02.015

3. FDA. Pharmaceutical CGMPs for the 21s Century - A risk-based approach.

2004;(September):32. Available from:

http://www.fda.gov/Drugs/DevelopmentApprovalProcess/Manufacturing/Questionsand

AnswersonCurrentGoodManufacturingPracticescGMPforDrugs/UCM071836

4. Allison G, Cain YT, Cooney C, Garcia T, Bizjak TG, Holte O, et al. Regulatory and

Quality Considerations for Continuous Manufacturing May 20-21, 2014 Continuous

Manufacturing Symposium. J Pharm Sci [Internet]. 2016 Nov 10;104(3):803–12.

Available from: http://dx.doi.org/10.1002/jps.24324

5. Byrn S, Futran M, Thomas H, Jayjock E, Maron N, Meyer RF, et al. Achieving

continuous manufacturing for final dosage formation: Challenges and how to meet

them May 20-21, 2014 continuous manufacturing symposium. J Pharm Sci.

2015;104(3):792–802.

6. Fonteyne M, Vercruysse J, De Leersnyder F, Van Snick B, Vervaet C, Remon JP, et

al. Process Analytical Technology for continuous manufacturing of solid-dosage forms.

TrAC - Trends Anal Chem [Internet]. 2015;67:159–66. Available from:

http://dx.doi.org/10.1016/j.trac.2015.01.011

7. Mascia S, Heider PL, Zhang H, Lakerveld R, Benyahia B, Barton PI, et al. End-to-end

continuous manufacturing of pharmaceuticals: Integrated synthesis, purification, and

final dosage formation. Angew Chemie - Int Ed. 2013;52(47):12359–63.

8. Malet-Sanz L, Susanne F. Continuous flow synthesis. a pharma perspective. J Med

Chem. 2012;55(9):4062–98.


78

9. ICH. Pharmaceutical Development Q8. ICH Harmon Tripart Guidel. 2009;8(August):1–

28. Available from:

.http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Quality/Q8_R

1/Step4/Q8_R2_Guideline.pdf

10. Services H. Guidance for Industry Process Validation : General Principles and

Practices Guidance for Industry Process Validation : General Principles and Practices.

Quality. 2011;(January):22.

11. ICH Expert Working Group. Quality Implementation Working Group on Q8, Q9 and

Q10 Questions & Answers (R4). 2010;41(22):18. Available

from:http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Quality/

Q8_9_10_QAs/QIWG_QAs_Step4/Q8_Q9_Q10_Question_and_Answer_R4_step_4_

November_2010.pdf

12. ICH Quality Implementation Working Group Points To Consider ( R2 ) ICH-Endorsed

Guide for ICHQ8/Q9/Q10 Implementation. Quality. 2011;41(December):20. Available

from:

http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Quality/Q8_9_

10_QAs/PtC/Quality_IWG_PtCR2_6dec2011.pdf

13. ICH Expert Working Group. Quality Risk Management Q9. ICH Harmon Tripart Guidel

[Internet]. 2005;(November):1–23. Available from:

http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:Quality+Risk+Manag

ement+Q9#0\nhttp://www.cls.co.at/downloads/ich-q9---quality-risk-management.pdf

14. Pogány J. ICH pharmaceutical quality system Q10. WHO Drug Inf. 2008;22(3):177–

81. Available from: http://www.fda.gov/downloads/.../ucm073517.pdf

15. ICH. ICH Q11 - Development and Manufacture of DSs. Guideline [Internet].

2012;(May):30. Available from:

http://www.ich.org/products/guidelines/quality/article/quality-guidelines.html

16. Singh R, Ierapetritou M, Ramachandran R. System-wide hybrid MPC-PID control of a

continuous pharmaceutical tablet manufacturing process via direct compaction. Eur J

Pharm Biopharm [Internet]. 2013;85(3 PART B):1164–82. Available from:

http://dx.doi.org/10.1016/j.ejpb.2013.02.019

17. Schaber SD, Gerogiorgis DI, Ramachandran R, Evans JMB, Barton PI, Trout BL.

Economic analysis of integrated continuous and batch pharmaceutical manufacturing:

A case study. Ind Eng Chem Res. 2011;50(17):10083–92.


79

18. Mit CC, Fda TG, Fda RM, Vertex EM. Regulatory and Quality Considerations for

Continuous Manufacturing. 2014; [Internet]. 2016 Nov 10;104(3):803–12. Available

from: http://dx.doi.org/10.1002/jps.24324

19. Byrn S, Futran M, Thomas H, Jayjock E, Maron N, Meyer RF, et al. Achieving

Continuous Manufacturing for Final Dosage Formation: Challenges and How to Meet

Them May 20-21 2014 Continuous Manufacturing Symposium. J Pharm Sci [Internet].

2016 Nov 10;104(3):792–802. Available from: http://dx.doi.org/10.1002/jps.24247

20. Baxendale IR, Braatz RD, Hodnett BK, Jensen KF, Johnson MD, Sharratt P, et al.

Achieving Continuous Manufacturing: Technologies and Approaches for Synthesis,

Workup, and Isolation of Drug Substance May 20-21, 2014 Continuous Manufacturing

Symposium. J Pharm Sci [Internet]. 2016 Nov 10;104(3):781–91. Available from:

http://dx.doi.org/10.1002/jps.24252

21. Ferreira LS, Trierweiler JO. Modeling and simulation of the polymeric nanocapsule

formation process. IFAC Proc Vol. 2009;7(PART 1):405–10.

22. Engisch W, Muzzio F. Using Residence Time Distributions (RTDs) to Address the

Traceability of Raw Materials in Continuous Pharmaceutical Manufacturing. J Pharm

Innov. 2016;11(1):64–81.

23. Gao Y, Vanarase A, Muzzio F, Ierapetritou M. Characterizing continuous powder

mixing using residence time distribution. Chem Eng Sci [Internet]. 2011;66(3):417–25.

Available from: http://dx.doi.org/10.1016/j.ces.2010.10.045

24. Reitz E, Podhaisky H, Ely D, Thommes M. Residence time modeling of hot melt

extrusion processes. Eur J Pharm Biopharm [Internet]. 2013;85(3 PART B):1200–5.

Available from: http://dx.doi.org/10.1016/j.ejpb.2013.07.019

25. Moe T. Perspectives on traceability in food manufacture. Trends Food Sci Technol.

1998;9(5):211–4.

26. Singh R, Ierapetritou M, Ramachandran R. An engineering study on the enhanced

control and operation of continuous manufacturing of pharmaceutical tablets via roller

compaction. Int J Pharm [Internet]. 2012;438(1–2):307–26. Available from:

http://dx.doi.org/10.1016/j.ijpharm.2012.09.009


80

27. Yu LX, Amidon G, Khan MA, Hoag SW, Polli J, Raju GK, et al. Understanding

pharmaceutical quality by design. AAPS J [Internet]. 2014;16(4):771–83. Available

from:http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=4070262&tool=pmcent

rez&rendertype=abstract

28. Kourti T. The Process Analytical Technology initiative and multivariate process

analysis, monitoring and control. Anal Bioanal Chem. 2006;384(5):1043–8.

29. Mehta KA, Rekhi GS, Parikh DM. Extrusion/Spheronization as a Granulation

Technique. Handb Pharm Granulation Technol [Internet]. 2005;333–63. Available

from: http://books.google.com/books?id=at26LZUP9MIC

30. Teng Y, Qiu Z, Wen H. Systematical approach of formulation and process

development using roller compaction. Eur J Pharm Biopharm [Internet].

2009;73(2):219–29. Available from: http://dx.doi.org/10.1016/j.ejpb.2009.04.008

31. Miguélez-Morán AM, Wu CY, Dong H, Seville JPK. Characterisation of density

distributions in roller-compacted ribbons using micro-indentation and X-ray micro-

computed tomography. Eur J Pharm Biopharm [Internet]. 2009;72(1):173–82.


32. Lim H, Dave VS, Kidder L, Neil Lewis E, Fahmy R, Hoag SW. Assessment of the

critical factors affecting the porosity of roller compacted ribbons and the feasibility of

using NIR chemical imaging to evaluate the porosity distribution. Int J Pharm [Internet].

2011;410(1–2):1–8. Available from: http://dx.doi.org/10.1016/j.ijpharm.2011.02.028

33. Rana A, Khokra SL, Chandel A, Prasad G. ewsletter Rana et al . 2011;298:286–98.

34. Bindhumadhavan G, Seville JPK, Adams MJ, Greenwood RW, Fitzpatrick S. Roll

compaction of a pharmaceutical excipient: Experimental validation of rolling theory for

granular solids. Chem Eng Sci [Internet]. 2005 Jul;60(14):3891–7. Available from:

http://www.sciencedirect.com/science/article/pii/S0009250905001223

35. Chang CK, Alvarez–Nunez FA, Jr JVR, Magnusson L-E, Sueda K. Roller Compaction,

Granulation and Capsule Product Dissolution of Drug Formulations Containing a

Lactose or Mannitol Filler, Starch, and Talc. AAPS PharmSciTech [Internet].

2008;9(2):597–604. Available from: http://link.springer.com/article/10.1208/s12249-

008-9088-y\nhttp://link.springer.com/content/pdf/10.1208/s12249-008-9088-y.pdf


81

36. Pietsch W. Agglomeration Technologies. In: Agglomeration Processes [Internet].

Wiley-VCH Verlag GmbH; 2008. p. 133–8. Available from:

http://dx.doi.org/10.1002/9783527619801.ch6

37. Simon O, Guigon P. Correlation between powder-packing properties and roll press

compact heterogeneity. Powder Technol [Internet]. 2003 Feb 19;130(1–3):257–64.

Available from: http://www.sciencedirect.com/science/article/pii/S0032591002002024

38. Falzone AM, Peck GE, Mccabe GP. Effects of changes in roller compactor parameters

on granulations produced by compaction. Drug Dev Ind Pharm [Internet]. 1992 Jan

1;18(4):469–89. Available from: http://dx.doi.org/10.3109/03639049209043864

39. Acevedo D, Muliadi A, Giridhar A, Litster JD, Romañach RJ. Evaluation of three

approaches for real-time monitoring of roller compaction with near-infrared

spectroscopy. AAPS PharmSciTech [Internet]. 2012;13(3):1005–12. Available from:

http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=3429687&tool=pmcentrez&r

endertype=abstract

40. Kleinebudde P. Roll compaction/dry granulation: Pharmaceutical applications. Eur J

Pharm Biopharm. 2004;58(2):317–26.

41. Samanta AK, Ng KY, Heng PWS. Cone milling of compacted flakes: Process

parameter selection by adopting the minimal fines approach. Int J Pharm. 2012;422(1–

2):17–23.

42. Perez-Gandarillas L, Perez-Gago A, Mazor A, Kleinebudde P, Lecoq O, Michrafy A.

Effect of roll-compaction and milling conditions on granules and tablets properties. Eur

J Pharm Biopharm [Internet]. 2016; Available from:

http://linkinghub.elsevier.com/retrieve/pii/S0939641116302004

43. Mazor A, Perez-Gandarillas L, de Ryck A, Michrafy A. Effect of roll compactor sealing

system designs: A finite element analysis. Powder Technol [Internet]. 2016;289:21–30.

Available from: http://dx.doi.org/10.1016/j.powtec.2015.11.039

44. Perez-Gandarillas L, Perez-Gago A, Mazor A, Kleinebudde P, Lecoq O, Michrafy A.

Effect of roll-compaction and milling conditions on granules and tablet properties. Eur

J Pharm Biopharm [Internet]. 2016;106:38–49. Available from:



82

45. Khorasani M, Amigo JM, Sun CC, Bertelsen P, Rantanen J. Near-infrared chemical

imaging (NIR-CI) as a process monitoring solution for a production line of roll

compaction and tableting. Eur J Pharm Biopharm [Internet]. 2015;93:293–302.


46. Khorasani M, Amigo JM, Bertelsen P, Sun CC, Rantanen J. Process optimization of

dry granulation based tableting line: Extracting physical material characteristics from

granules, ribbons and tablets using near-IR (NIR) spectroscopic measurement.

Powder Technol [Internet]. 2015; Available from:

http://dx.doi.org/10.1016/j.powtec.2016.03.004

47. Feng T, Wang F, Pinal R, Wassgren C, Carvajal MT. Investigation of the Variability of

NIR In-line Monitoring of Roller Compaction Process by Using Fast Fourier Transform

(FFT) Analysis. Aaps Pharmscitech [Internet]. 2008;9(2):419–24. Available from:

http://dx.doi.org/10.1208/s12249-008-9054-8

48. Bacher C, Olsen PM, Bertelsen P, Kristensen J, Sonnergaard JM. Improving the

compaction properties of roller compacted calcium carbonate. Int J Pharm.

2007;342(1–2):115–23.

49. Freitag F, Reincke K, Runge J, Grellmann W, Kleinebudde P. How do roll

compaction/dry granulation affect the tableting behaviour of inorganic materials?

Microhardness of ribbons and mercury porosimetry measurements of tablets. Eur J

Pharm Sci. 2004;22(4):325–33.

50. Inghelbrecht S, Remon JP. Roller compaction and tableting of microcrystalline

cellulose/drug mixtures. Int J Pharm [Internet]. 1998 Feb 23;161(2):215–24. Available

from: http://www.sciencedirect.com/science/article/pii/S0378517397003566

51. Herting MG, Kleinebudde P. Roll compaction/dry granulation: Effect of raw material

particle size on granule and tablet properties. Int J Pharm. 2007;338(1–2):110–8.

52. Heiman J, Tajarobi F, Gururajan B, Juppo A, Abrahmsén-Alami S. Roller compaction

of hydrophilic extended release tablets-combined effects of processing variables and

drug/matrix former particle size. AAPS PharmSciTech [Internet]. 2015;16(2):267–77.

Available from: http://www.ncbi.nlm.nih.gov/pubmed/25273028


83

53. Dumarey M, Wikstr??m H, Fransson M, Spar??n A, Tajarobi P, Josefson M, et al.

Combining experimental design and orthogonal projections to latent structures to

study the influence of microcrystalline cellulose properties on roll compaction. Int J

Pharm [Internet]. 2011;416(1):110–9. Available from:

http://dx.doi.org/10.1016/j.ijpharm.2011.06.018

54. Sun CC. Decoding Powder Tabletability: Roles of Particle Adhesion and Plasticity. J

Adhes Sci Technol [Internet]. 2011;25(4–5):483–99. Available from:

http://dx.doi.org/10.1163/016942410X525678

55. Sun CC. Materials Science Tetrahedron 2014;A Useful Tool for Pharmaceutical

Research and Development. J Pharm Sci [Internet]. 2016 Nov 10;98(5):1671–87.


56. Malkowska S, Khan KA. Effect of Re-Conpression on the Properties of Tablets

Prepared by Dry Granulation. Drug Dev Ind Pharm [Internet]. 1983 Jan 1;9(3):331–47.

Available from: http://dx.doi.org/10.3109/03639048309044678

57. Patel S, Dahiya S, Calvin Sun C, Bansal AK. Understanding Size Enlargement and

Hardening of Granules on Tabletability of Unlubricated Granules Prepared by Dry

Granulation. J Pharm Sci [Internet]. 2016 Nov 10;100(2):758–66. Available from:


58. Sun CC, Kleinebudde P. Mini review: Mechanisms to the loss of tabletability by dry

granulation. Eur J Pharm Biopharm [Internet]. 2015; Available from:


59. Inghelbrecht S, Paul Remon J. The roller compaction of different types of lactose. Int J

Pharm [Internet]. 1998 May 18;166(2):135–44. Available from:

http://www.sciencedirect.com/science/article/pii/S0378517398000222

60. Gupta A, Peck GE, Miller RW, Morris KR. Effect of the variation in the ambient

moisture on the compaction behavior of powder undergoing roller-compaction and on

the characteristics of tablets produced from the post-milled granules. J Pharm Sci.

2005;94(10):2314–26.

61. Roberts Rj, Rowe Rc. The effect of punch velocity on the compaction of a variety of

materials. J Pharm Pharmacol [Internet]. 1985 Jun 1;37(6):377–84. Available from:

http://dx.doi.org/10.1111/j.2042-7158.1985.tb03019.x


84

62. David ST, Augsburger LL. Plastic flow during compression of directly compressible

fillers and its effect on tablet strength. J Pharm Sci [Internet]. 1977 Feb 1;66(2):155–9.


63. Rees Je, Rue Pj. Time-dependent deformation of some direct compression

excipients*. J Pharm Pharmacol [Internet]. 1978 Sep 1;30(1):601–7. Available from:

http://dx.doi.org/10.1111/j.2042-7158.1978.tb13340.x

64. Nordström J, Alderborn G. The Granule Porosity Controls the Loss of Compactibility

for Both Dry- and Wet-Processed Cellulose Granules but at Different Rate. J Pharm

Sci [Internet]. 2016 Nov 10;104(6):2029–39. Available from:


65. Nordström J, Alderborn G. Degree of compression as a potential process control tool

of tablet tensile strength. Pharm Dev Technol [Internet]. 2011 Dec 1;16(6):599–608.

Available from: http://dx.doi.org/10.3109/10837450.2010.502177

66. Blanco M, Alcal M, Menezes JC, Felizardo PM, Garrido A, Dolores P, et al. Near-

infrared Spectroscopy in Laboratory and Process Analysis. Encyclopedia of Analytical

Chemistry. 2012. 1-46 p.

67. Dave VS, Fahmy RM, Hoag SW. Near-infrared spectroscopic analysis of the breaking

force of extended-release matrix tablets prepared by roller-compaction: Influence of

plasticizer levels and sintering temperature. Drug Dev Ind Pharm [Internet].

2015;41(6):898–905. Available from: http://www.scopus.com/inward/record.url?eid=2-

s2.0-84930514587&partnerID=40&md5=904eeca9fc5433c093689bd0e0f69c21

68. Gupta A, Peck GE, Miller RW, Morris KR. Nondestructive Measurements of the

Compact Strength and the Particle-Size Distribution after Milling of Roller Compacted

Powders by Near-Infrared Spectroscopy. J Pharm Sci. 2004;93(4):1047–53.

69. Gupta A, Peck GE, Miller RW, Morris KR. Real-time near-infrared monitoring of

content uniformity, moisture content, compact density/tensile strength, and young’s

modulus of roller compacted powder blends. J Pharm Sci. 2005;94(7):1589–97.

70. Gupta A, Austin J, Davis S, Harris M, Reklaitis G. A novel microwave sensor for real-

time online monitoring of roll compacts of pharmaceutical powders online - A

comparative case study with NIR. J Pharm Sci. 2015;104(5):1787–94.


85

71. Järvinen K, Hoehe W, Järvinen M, Poutiainen S, Juuti M, Borchert S. In-line

monitoring of the drug content of powder mixtures and tablets by near-infrared

spectroscopy during the continuous direct compression tableting process. Eur J

Pharm Sci. 2013;48(4–5):680–8.

72. Khorasani M, Amigo JM, Sonnergaard J, Olsen P, Bertelsen P, Rantanen J.

Visualization and prediction of porosity in roller compacted ribbons with near-infrared

chemical imaging (NIR-CI). J Pharm Biomed Anal [Internet]. 2015;109:11–7. Available

from: http://dx.doi.org/10.1016/j.jpba.2015.02.008

73. Wise BM, Gallagher NB, Windig W. Chemometrics Tutorial for PLS _ Toolbox and

Solo. 2006. 420 p.

74. Simonaho S-P, Ketolainen J, Ervasti T, Toiviainen M, Korhonen O. Continuous

manufacturing of tablets with PROMIS-line – Introduction and case studies from

continuous feeding, blending and tableting. Eur J Pharm Sci [Internet]. 2016;90:38–46.

Available from: http://www.sciencedirect.com/science/article/pii/S0928098716300367

75. Hosokawa Ktron [cited 2016 Oct 13]. Available from:

http://www.coperionktron.com/process-equipment/feeders/feeding-equipment/loss-in-

weight-feeders-overview.cfm?

76. Malvern [cited 2016 Sep 12]. Available from:

http://www.malvern.com/en/support/product-support/legacy-products/mastersizer-

2000/

77. Rowe R, Sheskey P, Quinn M. Handbook of Pharmaceutical Excipients. Handb Pharm

excipients, Sixth Ed. 2009;549–53.

78. Prasad K V, Sheen DB, Sherwood JN. Fracture property studies of paracetamol single

crystals using microindentation techniques. Pharm Res. 2001 Jun;18(6):867–72.

a continuous manufacturing model for the production of...

Documents