Computers and Chemical Engineering 42 (2012) 15–29

Contents lists available at SciVerse ScienceDirect

Computers and Chemical Engineering

journal homepage: www.elsevier.com/locate/compchemeng

A perspective on PSE in pharmaceutical process development and innovation

Krist V. Gernaey ∗ , Albert E. Cervera-Padrell, John M. Woodley
Department of Chemical and Biochemical Engineering, Technical University of Denmark, Building 229, DK-2800 Kgs., Lyngby, Denmark

a r t i c l e i n f o a b s t r a c t

Article history: The pharmaceutical industry is under growing pressure to increase efficiency, both in production and in
Received 21 September 2011 process development. This paper discusses the central role of Process Systems Engineering (PSE) methods
Received in revised form 26 February 2012 and tools in pharmaceutical process development and innovation, and searches for answers to questions
Accepted 28 February 2012
such as: Which PSE methods can be applied readily? Where is more method development needed? The
Available online 8 March 2012
paper covers key subjects for development of economically and environmentally sustainable pharmaceu-
tical processes, including Process Analytical Technology in its broadest sense, continuous pharmaceutical
Keywords:
manufacturing and green processes, and is illustrated with a series of short examples taken from the
Continuous pharmaceutical manufacturing
Design space
literature and ongoing research projects.
PAT © 2012 Elsevier Ltd. All rights reserved.
Pharmaceutical production
Process development
Sustainability

1. Introduction In this paper, the focus is on production of small-molecule

Today more than ever, the pharmaceutical industry is under
pressure to deliver its end products cheaper, faster and more
efficiently (Federsel, 2003). Competition from generic drug man-
ufacturers is undoubtedly one of the main drivers, and has
already triggered dramatic changes. Indeed, many of the traditional
research-based pharmaceutical companies have started the pro-
duction of generic drugs during the past five years, and are thus
in direct competition with generic drug manufacturers. Such com-
petition demands particular focus on production cost. In addition,
there is – in particular in Europe – a growing consumer demand
for so called ‘green’ pharmaceutical products (i.e. the introduction
of greener, more environmentally friendly production methods to
manufacture pharmaceuticals). The latter demand originates from
a growing awareness that drug manufacturing processes have a
significant environmental impact (Sheldon, 1992, 2007).
Abbreviations: ANN, artificial neural network; API, active pharmaceutical
ingredient; CAMD, computer-aided molecular design; CFD, computational fluid
dynamics; CP, continuous processing; CPM, continuous pharmaceutical manufac-
turing; CPP, critical process parameter; CQA, critical quality attribute; DEM, discrete
element method; DoE, design of experiments; FEM, finite element method; iPLS,
interval PLS; LCA, life cycle assessment; LLE, liquid–liquid extraction; MSPC, multi-
variate statistical process control; MW, molecular weight; NCE, new chemical entity;
NIR, near infra-red; OED, optimal experimental design; PAT, Process Analytical Tech-
nology; PBM, population balance model; PCA, Principal Component Analysis; PDE,
partial differential equation; PLS, Partial Least Squares; PMI, process mass intensity;
PSE, Process Systems Engineering; QbD, quality by design.
∗ Corresponding author. Tel.: +45 45252970; fax: +45 45932906.
E-mail address: kvg@kt.dtu.dk (K.V. Gernaey).
(MW < 1000) drug substances – active pharmaceutical ingredi-
ents (APIs), new chemical entities (NCEs) – that are produced via
organic synthesis. A generic process flowsheet of a small-molecule
API production process is provided in Fig. 1. Aside from antibi-
otics, which are small molecules that for many decades have been
produced by fermentation, there is an emerging trend toward
bio-production methods (such as enzyme-based catalysis). Inter-
estingly, several principles and concepts that will be highlighted in
the manuscript can also be applied to large-molecular (MW > 1000),
biotechnology-based drug substances – also called biopharmaceu-
ticals – which includes for example monoclonal antibodies and
therapeutic proteins.
Traditionally, the pharmaceutical industry has been rather
reluctant to introduce innovative methods in its production pro-
cesses, despite the fact that the cost for the development of a
new drug continually increases, while the patent life time is fixed
(Federsel, 2003). Considering the typical development cycle of a
drug, two obvious ways of increasing profit are: (1) more rapid
process development, and thus maximization of the time between
product release and patent expiration; and (2) streamlining the
full-scale production system to reduce production costs, such that
traditional research-based drug manufacturers can also compete
with generic drug manufacturers when the patent of a drug has
expired. In both cases, we are convinced that increased and sys-
tematic use of Process Systems Engineering (PSE) methods (Klatt &
Marquardt, 2009; Stephanopoulos & Reklaitis, 2011) can be advan-
tageous. In this context, the introduction of the Process Analytical
Technology (PAT) guidance by the FDA (2004) is an important mile-
stone as well, since its publication has ended a long period of

0098-1354/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2012.02.022
16 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29

Fig. 1. Typical flowsheet for a pharmaceutical manufacturing process. PSE methods and tools can be used to obtain data, analyze, model, simulate, control and optimize
each unit operation. Detailed information about specific unit operations for the formulation part of the process is provided elsewhere (feeding: Boukouvala et al., 2010;
mixing/blending: Boukouvala et al., 2010; roller compaction: am Ende et al., 2007; granulation: Vervaet & Remon, 2005; drying: Mortier et al., 2011; milling: Bilgili & Scarlett,
2005; tablet compaction: Michaut et al., 2010; coating: Prpich et al., 2010).

regulatory uncertainty. Indeed, the PAT guidance makes clear that production processes for small molecules. Finally, in the discus-
regulatory bodies are in favor of more efficient production meth- sion, we highlight several problems related to the application of
ods, as long as a safe (in-specification) product can be guaranteed. PSE methods and tools in pharmaceutical process development and
The disappearance of such regulatory uncertainty opens up new innovation, which also includes a short perspective on the function
and exciting possibilities for innovation in pharmaceutical produc- of PSE methods in the drug development cycle.
tion processes, and thereby also increased need for PSE methods
and tools. 2. Process Analytical Technology (PAT)
This paper will provide a perspective on the use of PSE methods
in pharmaceutical process development and innovation. It searches
Process Analytical Technology (PAT) (FDA, 2004) is undoubtedly
for the answers to questions such as: Which PSE methods can
one of the most influential new trends in pharmaceutical manufac-
be applied readily? Where is more method development needed?
turing in recent years. Interestingly, seen from the perspective of
The paper will be illustrated with examples taken from the liter-
the PSE community, PAT does not really bring so much new, since
ature and ongoing research projects. Rather than providing a long
the concepts described in the PAT guidance have been applied for
list of PSE methods and tools along with potential applications of
quite a long time by several other industries (e.g. petrochemical,
those methods, the manuscript has been structured according to
polymer and chemical sectors) (Kourti, 2006). As a consequence,
a number of main topics which we think are of major importance
there are quite a number of PSE methods and tools that can be used
today: implementation of PAT strategies, continuous production
in a PAT context, resulting from developments that took place in
processes and green pharmaceutical processes. It is worth not-
other industries. Here, we will first introduce the most essential PAT
ing as well that a topic such as ‘continuous production processes’
terminology, then discuss the most important PSE tools that can be
was ranked as the top priority on the list of key green engineering
used for PAT, and afterwards shift focus to PAT implementation and
research areas that has recently been published as a result of the
especially the role that PSE plays in PAT implementation.
brainstorming and prioritization exercises of the American Chem-
ical Society (ACS) Green Chemistry Institute (GCI) Pharmaceutical
Roundtable (Jiménez-González et al., 2011). This confirms that a 2.1. PAT terminology
more efficient process nearly always results in a greener process,
and that PSE methods and tools can play a prominent role in paving Several new concepts were introduced in the PAT guidance
the way toward more efficient and innovative pharmaceutical (FDA, 2004). The most important ones are:
 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
Fig. 2. Parallelism between process control loops and the definition of the design space based on CQAs of the end product.
• The design space, which is defined as “the multi-dimensional
combination of critical input variables and critical process param-
eters that lead to the right critical quality attributes” (FDA, 2004).
The term ‘critical’ should be interpreted as ‘having a significant
influence on final product quality’. The ICH Q8(R2) guideline on
pharmaceutical development provides examples of the design
space definition (ICH, 2009). Changing the process within the
design space is therefore not considered as a change, and thus
does not require a regulatory post-approval of the process. This
opens up the possibility of increased use of optimization methods
for pharmaceutical processes in the future (e.g. Nagy, Fujiwara,
& Braatz, 2008). Those methods have been used for a long time
in for example the chemical industry (e.g. Bhatia & Biegler, 1996;
Grossmann, 2005).
• A critical quality attribute (CQA) is defined as a physical, chemi-
cal, biological, or microbiological property or characteristic that
should be within an appropriate limit, range, or distribution to
ensure the desired product quality (ICH, 2009). However, we
prefer the definition given by Kourti (2006), which considers
the multivariate nature of pharmaceutical production processes.
“CQAs should be understood as the combination of measurable
properties of an intermediate or final product that are considered
critical for establishing the intended purity, efficacy, and safety of
the product. That is, the value of the multivariate statistic calcu-
lated for these properties must be within a predetermined limit
of a multivariate chart to ensure final product quality” (Kourti,
2006).
• A critical process parameter (CPP) is a process parameter whose
variability has an impact on a CQA and therefore should be mon-
itored or controlled to ensure the process produces the desired
quality (ICH, 2009).
• Real-time release means that the product quality is assessed in
real-time based on on-line quantification of the CQAs. Imple-
menting a real-time release strategy should result in a leaner
process – for example due to reduced laboratory work and
reduced need for storage of intermediates and final products
– and stands in sharp contrast to the traditional ‘lab-centric’
approach to quality, where product quality was typically assessed
off-line on the basis of often tedious laboratory analyses.
The parallelism between process control loops and the defi-
nition of the design space based on CQAs of the end product is
illustrated in Fig. 2. In this example, disturbance variables such
as variations in raw material properties x(t) are measured in real
time; output CQAs are also measured, directly or indirectly; and
process understanding has led to a predictive model f relating raw
material properties and process parameters u to output properties
y. Feed-forward and feed-back control loops ensure that output
quality attributes (CQAs) as well as CPPs remain within acceptable
17
limits. This information is used to define the design space, result-
ing in a larger space, compared to an open-loop system where raw
materials are typically rejected when they do not meet the narrow
specifications. Indeed, in principle the design space concept allows
to adjust process operation – for example by means of automatic
control – to compensate for a certain degree of variation in raw
material properties: as long as CPPs and CQAs are kept within the
design space limits that were proposed and documented by the
manufacturer when applying for regulatory approval of the pro-
cess, it is expected that the resulting product will meet the defined
quality. An optimization algorithm can in addition be established
in this large design space in order to meet process efficiency speci-
fications (e.g. low energy use, highest yield, lowest operating cost,
etc.) by proposing changes to the values of the CPPs, inspired for
example by the case studies reported by Marchetti et al. (2005) and
Nagy et al. (2008).
2.2. PAT tools
The PAT guidance (FDA, 2004) distinguishes four classes of tools:
(1) multivariate tools for design, data acquisition and analysis; (2)
process analyzers; (3) process control tools; and (4) continuous
improvement and knowledge management tools. In any of these
categories, one or several PSE methods and tools play a crucial role,
as summarized also in Table 1.
2.2.1. Multivariate tools
Similar to pharmaceutical product development – formulation
of the drug substance – pharmaceutical process development also
requires multivariate approaches in order to capture the interac-
tions between the different factors and their combined effects on
the final product quality in a satisfactory manner. The multivariate
tools category is very broad. Only the most important aspects will
be highlighted here.
2.2.1.1. Chemometrics. Chemometric methods are generally used
in the pharmaceutical industry, and can be helpful tools in acquir-
ing process knowledge, as discussed in detail by Kourti (2006).
It is important here to distinguish between: (1) basic methods
such as Partial Least Squares (PLS) (Wold, Sjöström, & Eriksson,
2001) and Principal Component Analysis (PCA); (2) slightly more
advanced methods such as interval PLS (iPLS) (Nørgaard et al., 2000)
and the combination of genetic algorithms with backwards iPLS
(Leardi & Nørgaard, 2004), methods which were both developed
for chemometric variable selection; and, (3) advanced methods,
such as non-linear versions of PCA and PLS (Dong & McAvoy, 1996;
Kruger, Zhang, & Xie, 2008; Zhang, Martin, & Morris, 1997), and
adaptive versions of PCA and PLS (Kadlec, Grbic, & Gabrys, 2011).

Table 1
Overview of the most relevant PSE tools that can be applied in a PAT context. Note that the references provided should be considered as examples. In the interest of space, it is not possible in this manuscript to provide an
exhaustive overview of the literature in each of these areas.
Tools class
(1) Multivariate tools
(2) Process analyzers
(3) Process control
(4) Knowledge management
PSE method
Chemometrics, basic methods (PCA, PLS)
Chemometrics, advanced methods
Mechanistic models
Reduced order models
Hybrid models
Design of experiments (DoE)
Soft sensors, data-driven
Soft sensors, model-driven
Basic control (SISO, PID, feedforward)
Advanced methods
Ontologies, knowledge bases
Status
Well-accepted, generally used in
industry
Developed, used in academia, not
so much in industry
Well developed, widely used in
academia, increasingly used in
industry
Well developed, especially in
academia
Well developed, widely used in
academia, limited use in industry
Well developed, generally used in
academia and industry
Theory well developed, used in
academia, sporadically used in
industry
Theory well developed, used in
academia, sporadically used in
industry
Theory well developed,
well-accepted, generally used
Theory well developed, specific
applications in industry
Need for more method
development
Review, theory
Wold et al. (2001), Reich (2005), Kourti
(2006)
Kruger et al. (2008), Kadlec et al. (2011)
Asprey and Macchietto (2000), Pordal
et al. (2002), Sales-Cruz and Gani
(2003), Marquardt (2005), Kremer and
Hancock (2006) Wassgren and Curtis
(2006), Franceschini and Macchietto
(2008), Gernaey and Gani (2010),
Mortier et al. (2011), Sharma et al.
(2011)
Eriksson et al. (2008)
Nomikos and McGregor (1995),
Venkatasubramanian, Rengaswamy,
Kavuri, and Yin (2003); Bersimis et al.
(2006), Kadlec et al. (2009)
Kadlec et al. (2009),
Venkatasubramanian et al. (2003a)
Seborg et al. (2010)
18
Applications, examples
Rosas et al. (2011b), Cervera-Padrell et al. (2012)
K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
Dong and McAvoy (1996), Zhang et al. (1997), Nørgaard et
al. (2000), Leardi and Nørgaard (2004)
ODEs: Dassau et al. (2006), Sin et al. (2008), Sin et al.
(2009), Singh et al. (2009);
PBM: Bilgili and Scarlett (2005), Fujiwara et al. (2005),
Aamir et al. (2010), Abdul Samad et al. (2011), Boukouvala
et al. (2012), Capece et al. (2011), Krasnyk et al. (2012);
DEM: Remy et al. (2009, 2010), Radeke et al. (2010), Adam
et al. (2011), Dubey et al. (2011);
CFD: Wei, Zhou, and Garside (2001), Wu et al. (2002), Woo
et al. (2009)
Krasnyk et al. (2012)
Akkisetty et al. (2010)
Mandenius and Brundin (2008), Koch et al. (2009),
Boukouvala et al. (2010), Boukouvala, Ramachandran, et al.
(2011)
Kourti (2006), Zamprogna et al. (2004), Camacho et al.
(2008)
Villez et al. (2011)
Cervera-Padrell et al. (2012)
Nagy et al. (2008), Hermanto et al. (2011)
Venkatasubramanian et al. (2006), Morbach et al. (2007),
Marquardt et al. (2010), Singh et al. (2010a)
 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
The most relevant applications of these non-linear methods are in
process monitoring and fault detection (e.g. Dong & McAvoy, 1996).
2.2.1.2. Mechanistic models. A dynamic mechanistic model is an
excellent tool for mapping process knowledge in the context of PAT
applications. The model equations can indeed represent a series of
input–output relations, and can be extended whenever new pro-
cess knowledge becomes available. In this context, it is important to
formulate the dynamic models in a structured way (Gernaey & Gani,
2010; Sin, Ödman, Petersen, Eliasson Lantz, & Gernaey, 2008), since
this also facilitates knowledge transfer. Once formulated in a struc-
tured way, a dynamic model can then be supplemented by a set
of well-established model analysis tools (e.g. Asprey & Macchietto,
2000; Sales-Cruz & Gani, 2003; Marquardt, 2005), also including
uncertainty and sensitivity analysis to assess the statistical qual-
ity (reliability) of the simulated scenarios (Sin, Gernaey, & Eliasson
Lantz, 2009). Most important, however, is that these dynamic mod-
els are actively used for in-silico testing of a set of potential process
operating strategies, e.g. by comparing different control strategies
in a series of dynamic simulations, without disturbing the full-scale
process operation. In this way implementation of more advanced
control can be established more efficiently.
The term “mechanistic model” is rather broad, since it covers
models consisting of systems of ordinary differential equations
(ODEs), differential algebraic equations (DAEs) and partial differ-
ential equations (PDEs).
Examples of mechanistic models based on ODEs can for example
be found in Sin et al. (2008), Singh, Gernaey, and Gani (2009) and
Zimmermann, Hennemann, Daußmann, and Kragl (2007), while
McMullen and Jensen (2011) demonstrate how to discriminate
between different model candidates for such a system of ODEs.
PDEs are increasingly applied to pharmaceutical production
processes as well. Population balance models (PBM) form one very
important class of mechanistic models based on PDEs. They have
been applied frequently for the description of the dynamics of the
crystal size distribution in crystallization processes (e.g. Aamir,
Nagy, & Rielly, 2010; Abdul Samad, Singh, Sin, Gernaey, & Gani,
2011; Nagy et al., 2008), for the description of the particle size dis-
tribution and the binder content in granulation processes (e.g. Poon
et al., 2009; Ramachandran, Immanuel, Stepanek, Litster, & Doyle,
2009) and blending operations (e.g. Boukouvala et al., 2012), and
for the description of the particle size distribution in milling pro-
cesses (e.g. Bilgili & Scarlett, 2005; Capece, Bilgili, & Dave, 2011).
Other relevant PBM applications are for example the description of
the moisture content of particles during a drying process (see also
Mortier et al., 2011), or even the description of the properties of a
population of microorganisms used for production of pharmaceu-
ticals via fermentation (Lencastre Fernandes et al., 2011).
Computational fluid dynamics (CFD), based on the
Navier–Stokes equations describing the motion of fluids, is
another application of PDEs that is increasingly used in the phar-
maceutical industry as well, and has for example been reviewed
by Kremer and Hancock (2006) and Wassgren and Curtis (2006).
Mortier et al. (2011) review specific CFD applications for fluid
bed drying processes. Pordal, Matice, and Fry (2002) suggested
already 10 years ago that CFD would be useful in the pharma-
ceutical industry for applications within mixing, solids handling,
separations (e.g. Wu, Wang, & Ching, 2002) and drying processes.
Furthermore, they specifically pointed toward the use of CFD
to model equipment for heat transfer and heat generation, in
order to optimize performance. In addition to those areas, the
rapid development of microreactor technology – very relevant
for the pharmaceutical industry as well with the current focus
on continuous production and process intensification – has also
benefitted substantially from the insights gained by using CFD for
the description of fluid flow, mass and heat transfer (e.g. Kashid,
19
Agar, & Turek, 2007; Mills, Quiram, & Ryley, 2007). CFD is also
increasingly used in bio-based manufacturing in the frame of QbD
projects to demonstrate detailed process knowledge (Sharma,
Malhotra, & Rathore, 2011). It is quite common to combine a CFD
model with a model describing chemical reactions (e.g. Kashid
et al., 2007). However, nowadays the trend is clearly toward the
development of combined CFD-PBM models which can describe
the change of distributed properties as a function of spatial coor-
dinates within (part of) a unit operation. As an example, Woo, Tan,
and Braatz (2009) combined CFD and PBM for the modeling of an
impinging jet crystallizer. Combining CFD and PBM is clearly an
area where the PSE community can make a major contribution to
development of detailed process understanding.
The discrete element method (DEM) is often used in the parti-
cle technology field for tracking of individual particles. Interactions
between the particles and the fluid are neglected, and instead the
focus is on description of particle–particle interactions (Radeke,
Glasser, & Khinast, 2010). In order to produce useful results, a large
number of particles need to be tracked, and as a consequence the
method is computationally rather expensive. Radeke et al. (2010),
for example, simulated several millions of particles during the
operation of a mixer. Similarly, Remy, Khinast, and Glasser (2009),
Remy, Glasser, and Khinast (2010) and Dubey, Sarkar, Ierapetritou,
Wassgren, and Muzzio (2011) also applied the DEM method while
studying granular blending processes. Adam, Suzzi, Radeke, and
Khinast (2011) give a specific QbD example on the use of the DEM
method within the definition of the design space for a blending unit
operation.
2.2.1.3. Reduced order models and hybrid models. In many cases the
implementation of a detailed mechanistic model comes with a high
computational burden, for example in the case of a detailed 3D
CFD model of a unit operation. Of course, setting up such a detailed
model is beneficial since it also leads to a deeper understanding of
the system. However, for many applications, for example related to
control of a specific operation, solving the problem in a computa-
tionally efficient way is of major importance. Here, development of
a reduced order model is an appropriate solution. In fluid dynam-
ics problems, for example, it is quite common to derive a reduced
order model from a CFD model in order to reduce simulation times
(e.g. Alonso, Velazquez, & Vega, 2009; Krasnyk, Mangold, & Kienle,
2010; Lang et al., 2009). Krasnyk, Mangold, Ganesan, and Tobiska
(2012) demonstrated how to replace a detailed multi-dimensional
PBM of a lab-scale crystallizer by a reduced order model. We
expect that methods for generation of reduced order models will
be increasingly applied within QbD projects in order to generate
fast predictions.
Hybrid models also form an attractive alternative in those
cases where a full mechanistic model is not available, and rely
on the combination of a mechanistic model with a data-driven
model component (e.g. Doyle, Harrison, & Crowley, 2003). It is
indeed too optimistic to assume that a detailed mechanistic model
can be developed for each pharmaceutical process. Often, in the
interest of time, a hybrid approach will be preferred, where the
mechanistic part of the model is gradually extended when more
process knowledge becomes available, e.g. during process develop-
ment. Akkisetty, Lee, Reklaitis, and Venkatasubramanian (2010), for
example, report on the use of a neural network model for represent-
ing the breakage function in a population balance model describing
the particle size distribution resulting from the processing of a par-
ticulate material – three types of API ribbons – in a milling unit.
2.2.1.4. Design of experiments. Statistical design of experiments
(DoE) is generally used in pharmaceutical process and product
development, and is well-documented (e.g. Eriksson, Johansson,
Kettaneh-Wold, Wikström, & Wold, 2008). Similar to chemometric
20 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
methods, the general acceptance of DoE is promoted by the fact
that there are several commercial software products available that
can be used to perform a DoE. The traditional way of using DoE is
based on an off-line design, followed by a series of experiments to
test a set of conditions of factors proposed in the DoE. However,
current challenges in this area are especially related to the on-line
application of such a DoE in continuous flow systems, for example
at the microscale (Koch et al., 2009).
2.2.2. Process analyzers
2.2.2.1. Spectroscopy. Methods such as PLS were developed in the
1970s and 1980s, and can be considered as generally accepted tools
in the pharmaceutical industry. The introduction of PAT has led
to a tremendous increase in the number of spectroscopic appli-
cations in the pharmaceutical industry, and as a consequence the
use of basic and slightly more advanced chemometric methods has
grown considerably. One of the factors explaining this growth is
that several software products – commercial and open source –
have become available. Another factor is that spectroscopic meth-
ods such as near infrared (NIR) spectroscopy, to give one example,
is described in both the American and the European Pharmacopeia.
Typical applications of NIR spectroscopy are (Reich, 2005): (1) iden-
tification and characterization of raw materials and intermediates;
(2) analysis of solid dosage forms such as tablets; and (3) prediction
of one or more variables in process streams or final product streams
(composition) on the basis of on-line, in-line or at-line spectro-
scopic measurements. Other types of spectroscopy can of course
be applied as well, but this manuscript is clearly not the place for
an extensive overview on spectroscopic methods.
2.2.2.2. Soft sensors. Soft sensors can be divided into two classes:
data-driven and model-driven soft sensors (Kadlec, Gabrys, &
Strandt, 2009). Soft sensors can potentially be used for prediction
of unknown variables, as well as for fault detection and diag-
nosis. An extensive overview of methods for fault detection and
diagnosis can be found in Venkatasubramanian, Rengaswamy, Yin,
and Kavuri (2003), Venkatasubramanian, Rengaswamy, and Kavuri
(2003), Venkatasubramanian, Rengaswamy, Kavuri, and Yin (2003).
Chemometric models clearly belong to the first class of soft sen-
sors, and can indeed be applied to traditional process data as well,
for example for prediction of the composition of mixtures or con-
centrations that are difficult to measure on-line (e.g. Camacho, Pico,
& Ferrer, 2008; Kadlec et al., 2009; Zamprogna, Barolo, & Seborg,
2004). Of course, there are other data-driven soft sensors as well,
and an exhaustive overview of the data-driven methods frequently
used in soft sensors in the process industries was recently provided
by Kadlec et al. (2009). Specifically for application of data-driven
soft sensors to pharmaceutical production processes, it is important
to realize that software sensors based on PLS or PCA are in fact to be
preferred, since these methods are well-known in the pharmaceu-
tical industry which facilitates validation. Both methods are based
on latent variables, and examining the scores and loadings (PCA,
PLS) or the weights (PLS) is excellent for data exploration as well
as to identify unusual or abnormal process behavior (Kourti, 2006).
Multivariate statistical process control (MSPC) (Bersimis, Psarakis,
& Panaretos, 2006; Kourti, 2006), often based on PCA or PLS, cov-
ers a series of well-established methods and tools for monitoring
a process that is influenced by several variables of interest. Other
types of data-driven soft sensors, for example based on artificial
neural networks (ANN), can also be useful, but in practice the gen-
eral opinion in industry is that it is difficult to have such an ANN
accepted by the regulatory bodies.
Monitoring of batch and continuous processes is fundamen-
tally different, in the sense that batch processes have an additional
dimension due to batch to batch variability (Kadlec et al., 2009).
Multi-way methods, for example multi-way PLS (Nomikos &
McGregor, 1995), were specifically developed for monitoring batch
processes, and are quite frequently used due to the fact that
most traditional pharmaceutical production processes are based
on batch operations.
Model-driven soft sensors can be based on mass or energy
balances (first principles approach). The main drawback is that
the development of such a model-driven soft sensor is gener-
ally considered to be rather time consuming (Kadlec et al., 2009).
When developing a pharmaceutical production process, time is
usually one of the most critical factors, and that is in our view
one of the main reasons why such model-driven soft sensors
are not more widespread in the pharmaceutical industry. Alter-
natively, an algorithm such as a Kalman filter or an extended
Kalman filter can also be used as a soft sensor, for example for
fault detection (Villez, Srinivasan, Rengaswamy, Narasimhan, &
Venkatasubramanian, 2011).
In conclusion, soft sensors have tremendous potential, for exam-
ple to exploit available on-line data to the maximum extent
possible, but practical implementation on pharmaceutical produc-
tion processes is not straightforward. One major challenge is the
time needed to develop a soft sensor. A second challenge might
be validation by regulatory bodies, especially for more advanced
data-driven or model-driven soft sensors. Last but not least, the lack
of convincing or easily accessible examples does not promote the
general acceptance of soft sensors in the pharmaceutical industry
either. Especially, with respect to this last point, we are convinced
that an advanced benchmarking tool for such algorithms, similar
to the Tennessee–Eastman challenge (Downs & Vogel, 1993) but
more focused on pharmaceutical production processes, could be
useful as a testbed to compare the performance of different algo-
rithms (benchmarking), and to demonstrate the benefits related to
the use of the different algorithms to potential end-users (training).
2.2.3. Process control tools
Similar to chemometric methods (see Section 2.2.1.1), one can
also distinguish between basic control methods (single input – sin-
gle output, PID controller) (Seborg, Mellichamp, Edgar, & Doyle,
2010) and advanced methods such as adaptive control or model
predictive control. Basic control methods can be assumed to be
industry standard, and are generally used. However, a new chal-
lenge here with the introduction of PAT is the incorporation of
advanced on-line measurements, for example spectroscopic mea-
surements (e.g. Cervera-Padrell et al., 2012), in such a basic control
loop rather than a standard temperature, flow rate or pressure mea-
surement.
Advanced control methods are increasingly used, and the devel-
opment of new applications and case studies is often driven by
academia (e.g. Hermanto et al., 2011). What industry needs in order
to also adopt advanced control is a number of convincing case stud-
ies which clearly demonstrate the economical benefit of including
more advanced control.
2.2.4. Continuous improvement and knowledge management
tools
Due to the complexity of the problems to be addressed in
pharmaceutical product-process design, an efficient and system-
atic knowledge base coupled with an inference system is essential.
Morbach, Yang, and Marquardt (2007) presented an overview of
OntoCAPE, a general ontology for structuring knowledge in the
chemical process engineering field. Details of this ontology can be
found in Marquardt, Morbach, Wiesner, and Yang (2010). Singh,
Gernaey, and Gani (2010a) described an ontological knowledge
based system for selection of process monitoring and analysis tools
in the frame of PAT system design. An ontological framework to
support product and process development in the pharmaceutical
 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
Fig. 3. The central role of PSE to simultaneously address process design for contin-
uous pharmaceutical manufacturing, design space definition and monitoring and
control using process analytical technologies. Synergy is obtained when the three
design problems are tackled using central mechanistic modeling.
manufacturing domain was developed by Venkatasubramanian et
al. (2006).
Turning data into knowledge and managing that knowledge will
remain one of the major challenges for the future. In the pharma-
ceutical industry, storage of historical data is usually managed by an
appropriate software product from an external supplier, and such
software products are nowadays well established. However, even
with such a software product in place, one does not necessarily
possess the required process knowledge, for example in order to
improve the performance of a process. Indeed, extraction of knowl-
edge from data is the bottleneck, and should be one of the focus
points of future research.
2.3. Implementation of PAT
It is up to the drug producer to define the design space of a
particular pharmaceutical production process, and this definition
should be based on process understanding, i.e. a detailed – prefer-
ably mechanistic – understanding of the way in which critical input
variables and process parameters will influence product quality.
Tools such as DoE and chemometric models are frequently used for
the mapping of a design space (Boukouvala, Muzzio, & Ierapetritou,
2010; Rosas, Blanco, González, & Alcalá, 2011a, 2011b) of a pro-
cess. Clearly, mapping of the design space of a PAT-assisted process
increasingly justifies the development of mechanistic models of
the process as well, to capture the often non-linear relationship
between process variables and final product quality (Dassau, Zadok,
& Lewin, 2006; Sin et al., 2008; Singh et al., 2009). Addressing
the uncertainties in the process, either on the basis of a mecha-
nistic model description (Sin et al., 2009) or using experimental
approaches (Paakkunainen et al., 2009), will also gain importance
in order to obtain a prediction of the risk of not achieving final prod-
uct quality. Methods such as uncertainty analysis combined with
local and global sensitivity analysis (Sin et al., 2009) can be helpful
to direct experimental efforts, for example in designing experi-
ments that have focus on the collection of data that lead to those
parameters that are most influential on the model outputs.
Thus, according to us, PSE methods and tools will in general
play an increasingly important role in the definition of the design
space, which is schematically illustrated in Fig. 3 as well. At the
recent AIChE Annual Meeting 2010, several major pharmaceuti-
cal companies indeed confirmed this trend by indicating that they
are developing mechanistic models of unit processes in order to
demonstrate process understanding. The drawback of the devel-
opment of a mechanistic model is that it is considered to be time
consuming in comparison to data-driven approaches. Here, the key
challenge is to build and extend the model along the life cycle of the
product, whenever new process knowledge becomes available. For
example, detailed information on the kinetics of a reaction should
be available from lab-scale tests, and can be used to establish a
21
simple kinetic model. Assuming that the kinetics are transferable
across scales, they can be extended with appropriate constitutive
equations to describe the relevant transport processes at each scale
of operation (lab, pilot, production). Advanced on-line data, for
example collected during pilot-scale experiments with the process
to be developed, often yields dynamic data which can be used for
model validation.
Crystallization, a key operation in most pharmaceutical pro-
cesses, forms a good example to illustrate how process knowledge
incorporated into models has been exploited to further improve
the process. The increased understanding of the crystallization
process has led to the use of first-principles based approaches
for crystallizer design (Fujiwara, Nagy, Chew, & Braatz, 2005),
and first-principles based models are increasingly used in the
design of advanced control strategies for crystallization (Nagy et
al., 2008). Within the crystallization field, an important trend is
now to actively influence the crystal size distribution and the crys-
tal morphology. As an illustration of that, Chen and Trout (2010)
recently reported the use of a computer-aided solvent selection
procedure with the specific purpose of improving the morphol-
ogy of needle-like crystals. The method was successfully applied to
2,6-dihydroxybenzoic acid (DHB) to improve the crystal morphol-
ogy. This type of applications opens up the possibility of rational
design of crystals, in order to improve the quality of the formulated
drug.
Real-time release, one of the key concepts in PAT, requires that
real-time monitoring tools are available and that they are used
efficiently. When implementing PAT on a production process, it
is thus important to decide which variables should be monitored,
and which control loops are necessary in order to obtain a prod-
uct that has the required quality. Here also, PSE methods and tools
can point toward a solution. Indeed, Singh et al. (2009) provide an
example of the application of a model-based framework for design
and analysis of PAT systems. Singh, Gernaey, and Gani (2010b) have
then created a systematic computer-aided framework to develop a
software (ICAS-PAT) for design, validation and analysis of PAT sys-
tems. The design of a PAT system involves the identification of the
CQAs, selection of economical and reliable on-line measurement
tools and integration of these on-line sensors with the control sys-
tem (Singh et al., 2009, 2010b). The objective could for example be
to obtain the predefined end product quality consistently, with the
minimum instrumentation cost.
3. Development of continuous production processes
Conventionally, when transferring a production process for an
API from lab-scale to production scale, the chemistry of the process
has been modified to fit the process into existing batch reac-
tor equipment, eventually leading to limited design possibilities
(Hessel, 2009). Despite adopting such a strategy, batch process-
ing has demonstrated its flexibility and suitability, for example in
accommodating very slow and/or heterogeneous chemistries. Nev-
ertheless, it has been shown as well that continuous processing
(CP), applied throughout all the stages of the life cycle of a drug
(Kang, Chung, Langer, & Khademhosseini, 2008; Webb & Jamison,
2010), provides an opportunity to reduce costs, footprint, energy
consumption, solvent utilization and environmental impact while
improving quality and control of manufacturing processes and final
products (LaPorte & Wang, 2007; Plumb, 2005; Pollet et al., 2009;
Schaber et al., 2011). Compared to batch processes, continuous
production typically implies a significantly smaller size of reactor,
ideally leading to flexible unit operation building blocks that can
be combined to create modular plants with advanced on-line sen-
sors and process control in order to guarantee product quality in
real-time (Singh et al., 2011).
22 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29

Table 2
Summary of specific challenges for continuous pharmaceutical manufacturing (CPM) and related PSE methods and tools which may be used as drivers for innovation. Note
that many PSE methods and tools and CPM specific challenges are inter-related.

Specific challenges PSE methods and tools References

• Batch vs. continuous selection  Reaction kinetics Hartman et al. (2011)

 Heat transmission Schaber et al. (2011)
 Mass transfer Gerogiorgis and Barton (2009)
 Process flowsheets Cervera et al. (2011)
 Economic and environmental assessment

• Economic and environmental assessment  Process modeling Schaber et al. (2011)

 Process flowsheets
 Simulation
 Property prediction
• Improved property prediction  Property prediction models and tools for
complex molecules and solutions
 Integration with process simulators
• New reactor and separation technologies  Mechanistic modeling for process
understanding
 Property prediction
Jimenez-Gonzalez et al. (2011)
Carvalho et al. (2008, 2009)
O’Connell et al. (2009)
Jensen (2001), Roberge et al. (2005)
Hartman et al. (2009), Kralj et al. (2007)
Hartman and Jensen (2009)
Hessel (2009), Kockmann and Roberge, 2009

• Scale-up, scale-out  Reaction kinetics Kockmann et al. (2011)

 Heat transmission Barthe et al. (2008)
 Mass transfer Jensen (2001)
 Fluid dynamics Mendorf et al. (2010)
 CFD, DEM, and FEM Pordal et al. (2002)

• Process optimization  Mechanistic modeling Gerogiorgis and Barton (2009)

 Data-based (black-box) models Boukouvala et al. (2012), Boukouvala, Muzzio,
 Hybrid models et al. (2011), Boukouvala, Ramachandran, et al.
 Process flowsheets (2011)
 CFD, DEM, and FEM Dubey et al. (2011)

• Real-time monitoring  Multivariate analysis Kourti (2006), Kadlec et al. (2009)

 Multivariate statistical process control McMullen and Jensen (2010a, 2010b, 2011)
 Optimal experimental design
• Design space and real-time release  Mechanistic modeling García Muñoz et al. (2010)
 Data-based (black-box) models MacGregor and Bruwer (2008)
 Hybrid models Kourti (2006), Kadlec et al. (2009)
 Multivariate statistical process control
 Process control
 Model predictive control
• Formulation operations  Improved process understanding Schaber et al. (2011)
 Modeling Boukouvala et al. (2012), Boukouvala, Muzzio,
 Multivariate analysis et al. (2011), Boukouvala, Ramachandran, et al.
 CFD, DEM, FEM (2011)
 Process flowsheets for operations involving Dubey et al. (2011)
solids Singh et al. (2009, 2010b)
 Property prediction Mortier et al. (2011)

• Start-up, shut-down, operational issues (clogging)  Dynamic modeling Hartman, Naber, Buchwald, et al. (2010),
 Process control Hartman, Naber, Zaborenko, et al. (2010)
 Risk assessment Boukouvala, Ramachandran, et al. (2011)

• Continuous bio(catalytic)processes  Improved process understanding Sin et al. (2008, 2009)

 Mechanistic modeling Gernaey, Eliasson Lantz, Tufvesson, Woodley,
 Property prediction and Sin (2010)

• Multipurpose and multistep continuous production lines  Dynamic modeling Singh et al. (2011)
 Process flowsheets Kockmann et al. (2011)

• Plantwide dynamic models  Mechanistic models Boukouvala, Ramachandran, et al. (2011)

 Data-based (black-box) models
 Hybrid models
 Process flowsheets
 Property prediction

3.1. The role of PSE in the design of continuous (pharmaceutical)
processes
When developing such continuous plants, it is evident that
property prediction and process design methods, which are both
well-established in the PSE academic community (as well as in
industry), will be required. Cervera, Gani, Kiil, Skovby, and Gernaey
(2011) propose a methodology for the development of a contin-
uous pharmaceutical manufacturing process, and apply it to the
transformation of a batch to a continuous manufacturing process.
The main benefits for a case study with zuclopenthixol, an API, are
a reduction of solvent use and the number of solvent exchange
operations. Computer-aided methods and tools form the corner-
stone of the methodology, and can be used for flowsheet generation
(Chakraborty & Linninger, 2002; Jaksland, Gani, & Lien, 1995),
solvent selection and replacement (Gani, Gómez, Folić, Jiménez-
González, & Constable, 2008), as well as solvent integration (Ahmad
& Barton, 1999). Ideally, dynamic models of the system should be
constructed as well, and used in simulation-based scenario anal-
ysis for reactor design, for model-based economic evaluation and
for environmental impact assessment. These models should ide-
ally be mechanistic and constructed on the basis of experimental
 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
data, where possible. However, if no detailed reaction mechanism
is known, they can also be data-driven (Kourti, 2006; Kadlec et al.,
2009) or hybrid models (Kahrs & Marquardt, 2008). In our opinion,
a key requirement here is that all the available process knowledge
should, whenever possible, be incorporated in mechanistic models
(Gernaey & Gani, 2010). One of the most important benefits offered
by computer-aided methods and tools relying on those models are
indeed the fact that one can obtain a rapid indication of the eco-
nomic performance of different process options, which is helpful
in decision-making.
We believe that PSE should have a central role in the design of
new continuous pharmaceutical processes or the retrofit of exist-
ing batch processes (see Table 2). However, CP is a new field for the
pharmaceutical industry, where new challenges and opportunities
are constantly arising in different areas: reaction engineering, sep-
aration technology, monitoring and control, and solids handling are
only some of the engineering challenges that have been intensively
studied in recent years. Thus far, research has focused primarily in
enabling these technologies, while a holistic view over the design of
CP is still under development (an introduction to the design prob-
lem can be found in Boukouvala, Ramachandran, Vanarase, Muzzio,
& Ierapetritou, 2011; Cervera et al., 2011; Gerogiorgis & Barton,
2009; Schaber et al., 2011). In the following we discuss some of
the specific challenges that PSE will need to address as a driver for
future innovation.
3.2. Specific challenges and opportunities for PSE
3.2.1. Batch or continuous?
The first question that arises when designing a pharmaceuti-
cal process is whether a specific unit operation should operate
in batch mode or in continuous mode. Hartman, McMullen, and
Jensen (2011) provide some clues to answer this question in
the context of discovery and development in a laboratory. They
propose a decision roadmap where the goal of the study (discov-
ery/screening/process development), the inherent characteristics
of the chemistry (fast/slow, exothermic/endothermic), and the
desired process conditions (easy to handle/extreme conditions)
are all taken into account to decide what the best experimental
setup should be. The decision is made based on non-dimensional
numbers summarizing information on reaction, mixing and heat
transfer rates as well as on axial dispersion in flow reactors. While
the decision tool they developed focuses on experimental studies, it
would be highly desirable to extend it to industrial scale processes.
Such decision-making methodologies should further include a
detailed economic (Schaber et al., 2011) and environmental assess-
ment, while considering monitoring and control, automation,
solvent recovery and heat integration. For example, Gerogiorgis and
Barton (2009) demonstrated with a steady-state simulation of an
upstream API synthesis process, that by using continuous process-
ing (in this case using microreactors and microseparators) it was
possible to reduce the flowsheet size, thereby also reducing fixed
investment cost and suppressing feedstock and solvent invento-
ries (operating cost). The total effluent waste was found to be an
order of magnitude lower than that generated by a more tradi-
tional batch manufacturing process. The waste could potentially be
reduced even further by solvent recycling, but this was not consid-
ered, since an extra unit operation would be needed which would
increase the fixed cost.
3.2.2. Continuous API synthesis – new paths enabled
Microreactors and microstructured reactors (Jensen, 2001;
Roberge, Ducry, Bieler, Cretton, & Zimmermann, 2005) as well
as microseparators (Hartman, Sahoo, Yen, & Jensen, 2009; Kralj,
Sahoo, & Jensen, 2007) have received a lot of attention in
recent years, enabling continuous flow multistep organic synthesis
23
(Hartman & Jensen, 2009; Hartman, Naber, Buchwald, & Jensen,
2010; Sahoo, Kralj, & Jensen, 2007). Mass transfer and heat trans-
mission phenomena are enhanced thanks to a large surface area
to volume ratio, while capillary and surface tension effects may be
exploited in many applications (Hartman & Jensen, 2009; Jensen,
2001; Kralj et al., 2007; Roberge et al., 2005). Improvements in mix-
ing and heat transmission as well as lower reaction volumes lead to
novel process windows (Hessel, 2009), where reactions can be car-
ried out under extreme conditions (Kockmann & Roberge, 2009),
for example at high temperature and pressure. Furthermore, new
chemical transformations (or already known chemistries which are
incompatible with batch processing), reactions at high concentra-
tion or under solvent-free conditions, and reactions in the explosive
or thermal runaway regime may become feasible (Hessel, 2009).
PSE methods and tools will need to be developed further to address
these problems, where mechanistic models should play a central
role. For example, surface tension and capillary effects, which are
the basis for microscale liquid–liquid extraction (LLE) and distilla-
tion, should be thoroughly described and understood, assisting for
example in membrane material selection or evaluating the feasibil-
ity of membrane based separations. While surface tension data is
usually available for pure solvents, mixtures of solvents with par-
tial miscibility and/or with dissolved solutes need to be evaluated
experimentally on a “case to case” basis. Development of prop-
erty prediction tools and rigorous models describing diffusion in
membrane assisted LLE and distillation would be highly beneficial.
3.2.3. Scaling-up and scaling-out of microscale CP
According to the microreaction paradigm as originally proposed,
the beneficial features of micro- reaction and separation technolo-
gies should be conserved across different scales of operation using
a scaling-out (or numbering-up) approach, i.e. replicating small
units working in parallel (Jensen, 2001; Plumb, 2005; Roberge
et al., 2005). However, fluid distribution and parallelization of
microscale units is not such a straightforward task as is often
claimed (Mendorf, Nachtrodt, Mescher, Ghaini, & Agar, 2010), and
according to some authors it should even be kept as a last option
(Kockmann, Gottsponer, & Roberge, 2011). Furthermore, solids
handling is still a major limitation for microreaction technology
(Hartman, Naber, Zaborenko, Buchwald, & Jensen, 2010). Therefore,
scaling-up microreactions has been proposed as a way to maintain
the advantages of operating at small length scales while being able
to process large flow rates in realistic (challenging) scenarios with a
low capital cost. The design across different scales should be driven
by a thorough description of mass and heat transfer characteristics
(Barthe et al., 2008; Kockmann et al., 2011). Computational Fluid
Dynamics (CFD) could be useful to describe mixing behavior across
different scales (Pordal et al., 2002). Indeed, assuming that reaction
kinetics can be determined at small scale and are independent of
scale, combining the reaction kinetics obtained for a specific pro-
cess with a CFD code should allow in silico investigation on how
non-ideal mixing influences reaction performance at larger scales.
3.2.4. Monitoring and control – integration of flow chemistry and
monitoring
Continuous-flow reactions require on-line monitoring and feed-
back control in order to take full advantage of the real-time
release PAT concept (FDA, 2004). Monitoring tools should provide
information about both quality specifications and overall system
performance. McMullen and Jensen (2010b) have reviewed the
integration of microreactors with in-line physical sensors and
analytical chemistry techniques, focusing on reaction monitoring
and development of algorithms for automated reaction screen-
ing and optimization (McMullen & Jensen, 2010a, 2010b, 2011).
The integration of continuous flow reactors with in-line analyti-
cal tools provides an opportunity to perform optimal experimental
24 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
design (OED) (Atkinson & Donev, 1992) for model discrimination
and kinetic parameter fitting in mechanistic models (McMullen &
Jensen, 2011). OED is performed in an automated fashion, reduc-
ing the number of experiments to carry out and obtaining the
maximum amount of information. However, while this method
has been successfully applied to relatively simple kinetic models
(McMullen & Jensen, 2011), algorithms for more complex reac-
tion mechanisms (typical of many pharmaceutical processes) need
to be further developed, as well as unified methods to quantify
information content (Franceschini & Macchietto, 2008).
3.2.5. Definition of design space
The introduction of PAT and QbD in the pharmaceutical indus-
try encourages the design and validation of processes to be done
not only at one set of fixed processing conditions, but rather for a
range of processing conditions known as the design space. Defining
a design space offers the possibility of responding to disturbances
entering the process (e.g. variability in raw materials) through
establishment of appropriate control loops, maintaining the pro-
cess operation under optimal conditions with a satisfactory product
quality, without the need to revalidate the process when certain
process parameters are manipulated. Hence, the acceptance of a
particular material is conditional on the values of the manipulated
variables calculated by the controller, if and only if the controller is
able to compensate for the variability in a stable and robust manner
(García Muñoz, Dolph, & Ward, 2010; MacGregor & Bruwer, 2008).
Representing the complex relationships arising from an
integrated materials-process-product design space requires a
model-centric approach. García Muñoz et al. (2010) propose a
mathematical framework for feed-forward control using latent
variable regression methods. A PLS model may be used to describe
the relationships between materials and products, under all
expected processing conditions. To perform the control calcula-
tions, first a regression model is built. This model uses a regressor
that is composed of known variables (i.e. raw material proper-
ties) and unknown variables (i.e. process manipulated variables),
and the final product quality as a response. When a new batch
of material is received, a control calculation is executed and the
combination of the best process conditions for the properties of
the new batch of material is found. The performance of the con-
trol strategy can be back-propagated to illustrate that an improved
controller implies the acceptance of a broader range of materials
(García Muñoz et al., 2010).
3.2.6. Plant-wide dynamic models
A special challenge related to the operation and control of phar-
maceutical plants working in continuous mode is how to obtain
a strictly continuous flow throughout unit operations with very
different characteristic times, and still be able to respond to dis-
turbances in the system in each operation. Another challenge
is how to introduce on-line operation into an existing process.
Clearly, it is very difficult to replace a complete production line
at once. The alternative is to integrate one continuous opera-
tion at a time in a process that is otherwise based on batch
processing, thus slowly turning the batch process into continu-
ous operation. In general, there is a need to develop plant-wide
dynamic models able to describe start-up, shut-down, how to deal
with clogging problems, and changing product lines in a mul-
tipurpose plant (Jiménez-González et al., 2011; Schaber et al.,
2011). Gerogiorgis and Barton (2009) performed a steady-state
simulation of an upstream continuous pharmaceutical process
including microreaction and microseparation steps. Boukouvala,
Ramachandran, et al. (2011) built a dynamic flowsheet model and
simulated an integrated continuous downstream pharmaceutical
process, including feeding, blending and granulation. Integrated
modeling and simulation tools for microreaction, microseparation
and solids processing requires further development and will even-
tually enable the simulation, design and optimization of continuous
pharmaceutical processes without the need to perform a consider-
able number of expensive trial and error experiments.
3.2.7. Property prediction
Pharmaceutical process design needs predictive thermody-
namic models to reduce the experimental search and focus
laboratory work in the most promising areas. Predictive models
like UNIFAC usually cannot be applied reliably to chemicals with
complex functionalities such as pharmaceuticals (O’Connell et al.,
2009). The prediction of crystal structure and solid-state properties
(crucial for many of the most common pharmaceutical unit oper-
ations) is still computationally demanding and far from generally
applicable (O’Connell et al., 2009). Furthermore, the pharmaceuti-
cal industry usually deals with complex chemistry, phase equilibria
involving organic salts and aqueous electrolytes. Pharmaceutical
systems are thus still challenging for property prediction tools. A
property prediction package would also be desirable, integrating
a database, a model library and a tool for parameter adoption. A
review of currently available property prediction tools is available
(O’Connell et al., 2009), including many examples from the phar-
maceutical industry. A more detailed review of property prediction
methods is out of the scope of this manuscript. Specific examples
of property prediction methods applied to pharmaceuticals can be
found in Taskinen and Yliruusi (2003), Matsuda, Kaburagi, Kurihara,
Tochigi, and Tomono (2010) and Diedrichs and Gmehling (2011).
3.2.8. Drug product formulation operations
A new trend in continuous manufacturing of pharmaceuticals is
the introduction of continuous processes in the final formulation of
drug substances as well (Boukouvala, Ramachandran, et al., 2011;
Plumb, 2005; Schaber et al., 2011; Vervaet & Remon, 2005). Indeed,
there is increasing interest in for example continuous tabletting
lines, both in academia and industry, where processes such as gran-
ulation, drying, compression and coating are integrated (Fig. 1).
Tablets are one of the most common dosage forms of pharma-
ceuticals. During the formulation of a tablet (see also Fig. 1 for
a schematic overview of the operations involved) the API is usu-
ally combined with excipients such as water, lactose, starch, sugar,
and coloring agents. Here also, advanced mechanistic models can
be helpful in equipment design and tuning, in defining the design
space as well as in the development of control strategies. However,
modeling of these processes is not straightforward, and devel-
opment of a detailed process model often requires a multi-scale
approach. For a drying process, to give one example, a detailed
description of the process requires a combination of a population
balance model (PBM) to track the dynamics of the distribution of the
moisture content of the particles, combined with a computational
fluid dynamics (CFD) code to describe the detailed spatio-temporal
behavior of the moisture content in the gas phase of the system.
The combined CFD-PBM should then allow the simulation of the
spatio-temporal dynamics of the moisture content in the parti-
cles throughout the dryer. Development of such a PBM starts at
the individual particle level, i.e. one first needs to investigate the
drying of individual particles in order to develop an understand-
ing of the most important phenomena involved. A review on the
state of the art in modeling of fluidized bed drying processes of
pharmaceutical granules can be found in Mortier et al. (2011). The
granular dynamics of continuous blending processes have recently
been studied by Dubey et al. (2011) using discrete element methods
(DEM), and were compared with less computationally expensive
PBM and data-based methods by Boukouvala et al. (2012), includ-
ing high-dimensional representations, response surface methods
and kriging (Boukouvala, Muzzio, & Ierapetritou, 2011). The objec-
tive was to study the effect of a number of process parameters
 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
(e.g. impeller speed) on the mixing quality while ensuring the
right residence time distribution and amount of shear, which may
affect properties such as hydrophobicity and flowability that could
adversely affect downstream unit operations such as tabletting
(Dubey et al., 2011). There is in general still a lot of work to be
done to further develop models, modeling strategies and solution
methods for multi-phase systems (gas, solid, and liquid). Reklaitis
(2007) identified specific process design problems that need to
be addressed, such as the development of predictive models and
design spaces for a set of common unit operations used in pharma-
ceutical production: multiphase batch reactors, antisolvent based
crystallization, solid–liquid separation, size reduction, granulation,
lyophilization, and a variety of separation types, including sim-
ulated moving beds. There is a need for systematic and reliable
methods for scale-up/scale-down based on rigorous CFD, DEM and
finite element method (FEM) simulation models. Multivariate con-
trol systems need to be adapted, and different operational issues
need to be solved (e.g. trend monitoring, incipient fault detection,
etc.). For continuous processing, process-wide automatic control
and real time process optimization methods need to be developed
further (Reklaitis, 2007).
4. Development of greener pharmaceutical processes
4.1. Design of sustainable pharmaceutical processes
In the last 5 years or so it has been clear that the pharmaceutical
industry mostly requires synthetic routes to be environmentally
sustainable in addition to still fulfilling the demands of high
product quality and process economics (Tucker, 2006). Different
approaches to pollution prevention are possible, from at-source
reduction to end-of-pipe treatment (Linninger & Malcolm, 2006).
Pollution prevention should be considered already in the concep-
tual design phase, since the regulatory (e.g. FDA) re-approval of a
process is very expensive. There are nowadays different software
tools for pollution prevention which help to quantify the envi-
ronmental implications of a process and generate suggestions for
process modifications (e.g. SustainPro developed by Carvalho, Gani,
& Matos, 2008; Carvalho, Matos, & Gani, 2009; Shonnard, Kicherer,
& Saling, 2003). A list of on-line manuals, databases of chemi-
cal impacts, decision-analysis tools, design options ranking tools,
expert systems, process analysis tools (simulators), waste reduc-
tion algorithms and finance tools can be found in Linninger and
Malcolm (2006).
Several metrics have been proposed in order to quantify
environmental sustainability and assist in the design of greener
processes (Constable, Curzons, & Cunningham, 2002; Curzons,
Constable, Mortimer, & Cunningham, 2001; Saling et al., 2002).
Among them, the process mass intensity (PMI) (or its inverse, the
mass efficiency) was selected by the ACS Green Chemistry Institute
(GCI)’s Pharmaceutical Roundtable as the key metric for evaluation
and benchmarking of sustainability (Jimenez-Gonzalez, Ponder,
Broxterman, & Manley, 2011), since it provides a good estimation
of the life cycle assessment (LCA) and footprint of a process, while
being fairly correlated to CO2 emissions.
4.2. Solvent selection and recovery
One way of achieving a greener production process is by eval-
uating solvent use, and by replacing solvents currently used in
a pharmaceutical production process with greener, more envi-
ronmentally friendly solvents where possible (Jiménez-González,
Curzons, Constable, & Cunningham, 2005). As an example, solvent
use at GlaxoSmithKline was reported to account for between 80%
and 90% of the mass utilization in a typical process (Constable,
25
Jiménez-González, & Henderson, 2007). 56% of the PMI in a typical
pharmaceutical process originates from solvents, while 32% origi-
nates from water use (note that unlike other sustainability metrics,
the PMI considers process water as well) (Jimenez-Gonzalez et al.,
2011). The poor water-solubility of many APIs means that at several
points during API manufacture, organic solvents are required (usu-
ally for extraction or crystallization). The ACS GCI Pharmaceutical
Roundtable recently produced an industry-wide solvent selec-
tion guide, freely available on-line (www.acs.org/greenchemistry).
While this guide considers safety, health and environmental
aspects of solvent selection, process design aspects need to be
addressed with complementary methodologies. In this respect, sys-
tematic computer-aided solvent selection methodologies (e.g. Gani
et al., 2008) can be very useful, in order to find alternative solvents
in an efficient way with minimum experimental work, or to come
up with green solvents for new production processes, for example
using computer-aided molecular design (CAMD) (e.g. Weis & Visco,
2010). Hsieh et al. (2009) recently presented a successful indus-
trial case study demonstrating the use of solvent selection methods
in the context of conceptual design of an API production process,
where a ternary solvent system was considered for a reaction,
extraction, distillation, and crystallization sequence. Chakraborty
and Linninger (2002) introduced combinatorial process syn-
thesis, which in essence allows an optimal trade-off between
process economics and environmental impact during flowsheet
synthesis.
Despite the large quantity of solvents used in the pharmaceuti-
cal industry, solvent recovery and recycling practices are not very
extended, due to the potential risk of cross-contamination and the
high solvent purity required (resulting in a high recovery cost). Cur-
rently, much of the solvent used for pharmaceutical production is
discarded as waste and incinerated. For example, within GSK, on
average less than 50% of solvent used was recycled and reused
by 2007 (Constable et al., 2007). Moving batch processes toward
continuous production opens up many possibilities for process
integration (Gerogiorgis & Barton, 2009), which should be sim-
pler to implement than in batch mode. Ahmad and Barton (1999)
presented a systematic design approach to the generation of batch
process flowsheets with integrated solvent recovery and recycling.
Chakraborty and Linninger (2002) demonstrate the simultaneous
optimization of structural decisions while finding the best oper-
ating parameters for an optimal sequencing of a solvent recovery
system.
4.3. Biocatalytic processes
The application of more efficient (or more selective) catalysts
is another way to obtain a greener pharmaceutical production
process. In this respect, biocatalysis has the potential to deliver
‘greener’ chemical syntheses (Pollard & Woodley, 2007; Woodley,
2008), since biocatalysts are more selective compared to tradi-
tional catalysts, and will usually assist in the synthesis of optically
pure molecules, which are essential for the manufacture of many
pharmaceutical compounds (Savile et al., 2010). Moreover, bio-
catalysts, due to their biological origin, normally work well in
water, although the recovery steps may still demand extrac-
tion with organic solvents (Henderson, Jiménez-González, Preston,
Constable, & Woodley, 2008). Furthermore, biocatalysis overcomes
the risk of using toxic metal based catalysts, which convention-
ally require significant effort to be removed to ensure a clean
product (Reginato, Sadler, & Wilkes, 2011). The mild conditions
(ambient temperature and pressure) also reduce energy costs
and improve selectivity via the minimization of side reactions
(Henderson et al., 2008; Straathof, Panke, & Schmid, 2002). How-
ever, integrating one or several biocatalytic steps into an existing
organic synthesis is often far from straightforward, due to e.g.
26 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
the occurrence of product inhibition which leads to low prod-
uct concentration and/or poor use of catalyst unless processes are
designed to overcome these problems (Tufvesson, Lima-Ramos,
Nordblad, & Woodley, 2011b). PSE methods can provide a sys-
tematic route to implementation of biocatalysis (Mandenius &
Brundin, 2008; Rohner & Meyer, 1995) and a framework for such
a scheme has recently been developed for transaminase–based
reactions for the synthesis of chiral amines – a particularly impor-
tant pharmaceutical target (Tufvesson, Lima-Ramos, Jensen, et al.,
2011). Increasingly PSE methods are also being applied to assist
in the identification of process bottlenecks via early stage eco-
nomic analysis. The identification of process bottlenecks enables
suitable development in a timely manner. Finally the development
and implementation of biocatalytic processes can benefit substan-
tially from computer-aided process performance evaluation, based
for example on mechanistic understanding and modeling of the
specific reaction of interest. For example Blayer, Woodley, Lilly, and
Dawson (1996) and Chen and Woodley (2002) have used a model to
develop windows of operation to predict the ideal operating space
for the aldolase-catalyzed synthesis of N-acetyl-d-neuraminic acid
from N-acetyl-d-mannosamine (see also Zimmermann et al., 2007).
An interesting review of the role of modeling in biocatalysis has
been recently published (Vasić-Rački, Findrik, & Vrsalović Presečki,
2011).
5. Future developments
5.1. Development of generic solutions
While pharmaceutical production offers a large space for
improvement in terms of technology, capital and operating costs
and sustainability, for many years innovative solutions have been
difficult to implement. The pressure to develop processes rapidly
(due to the limited patent life) used to outweigh the added benefits,
meaning that the extra development time required was difficult to
justify. Today, however, that situation is changing, and as the pres-
sure builds on the manufacturing cost, and the FDA opens up new
opportunities via the PAT initiative, a new situation is establishing
which facilitates – and in fact, demands – innovation. In our opin-
ion, PSE must play a central role as a driver of innovative solutions
and rational and efficient design, for example introducing rigorous
quantitative models, able to evaluate process performance from an
objective standpoint. Hence, the application and development of
PSE tools should be continuously encouraged.
It is also important to realize that a major strength of PSE meth-
ods and tools lies in the structured approach to solving problems.
The pharmaceutical industry – multipurpose in nature – indeed
needs generic approaches in order to solve in a systematic way mul-
tiple problems sharing common features. The scientific literature
frequently reports specific case studies or solutions to a particular
problem, while in our opinion, the focus should change to methods
and tools able to address a complete class of problems. Experi-
ence and confidence will thereby be built within pharmaceutical
companies to gradually implement innovating solutions. An addi-
tional benefit of adopting generic approaches is that it should be
easier to get approval from regulatory bodies such as the FDA. For
example, with the introduction of the PAT guidance, it has become
quite common to involve regulatory bodies as early as the develop-
ment phase, whenever new methods or tools need to be tested or
implemented. Recent PSE examples of such approaches are generic
modeling frameworks for prediction of the crystal size distribution
obtained in batch cooling crystallization processes (Abdul Samad et
al., 2011), and generic approaches for designing a pilot-plant able
to produce kg amounts of a family of related components (Singh et
al., 2011), for example during screening.
5.2. PSE methods and the product life cycle
A key question when considering the use of PSE methods and
tools during the development of a production process for an API is
when to start using these methods. Clearly, one will not harvest all
potential benefits if PSE is only used in a late process design stage.
Therefore, in our opinion, PSE methods and tools should be used
throughout the product life cycle, to facilitate continuous learning
and collection of process/product knowledge.
Likewise, we could start assuming that a mechanistic process
model is constructed early on in process development, based on
lab-scale data that were collected to characterize the reaction
kinetics. This model could then be used to increase available process
knowledge during the development phase while leading the path
for experimental design, thus fully integrating modeling and lab-
oratory research. Similarly, model transfer across different scales
would be greatly facilitated. Last but not least, small-scale con-
tinuous production – designed, monitored and controlled using a
variety of PSE methods and tools – can, in theory, lead to more
rapid process development by scaling-out the production as a valid
alternative to scaling-up.
6. Conclusions
• PSE methods and tools are essential for successful implementa-
tion of Process Analytical Technology (PAT) based processes.
• PSE methods and tools play a key role in the development of
continuous manufacturing processes as well as development of
greener pharmaceutical production processes, two major trends
in pharmaceutical manufacturing.
• Achieving the full benefit from PSE methods and tools is only
possible by applying them during the entire life cycle of a phar-
maceutical product and its production process.
• There are difficulties in transferring PSE methods from academia
to industry (see Table 1).
• The regulatory bodies should be involved early on when adopting
a new PSE tool in pharmaceutical production processes, in order
to minimize potential problems with process validation.
• The PSE methods and tools developed thus far are often focused
on reaction-separation sequences. Similar developments are
needed for downstream processing and formulation (e.g. gran-
ulation, drying, and tablet press).
• Pharmaceutical companies require generic and systematic
approaches to problem solving.
Acknowledgements
The authors would like to thank Prof. Rafiqul Gani (Technical
University of Denmark) for numerous fruitful discussions. The PhD
project of Albert E. Cervera-Padrell was financed by the Technical
University of Denmark and H. Lundbeck A/S.
References
Aamir, E., Nagy, Z. K., & Rielly, C. D. (2010). Optimal seed recipe design for crystal
size distribution control for batch cooling crystallisation processes. Chemical
Engineering Science, 65, 3602–3614.
Abdul Samad, N. A. F., Singh, R., Sin, G., Gernaey, K. V., & Gani, R. (2011). A generic
multi-dimensional model-based system for batch cooling crystallization pro-
cesses. Computers & Chemical Engineering, 35, 828–843.
Adam, S., Suzzi, D., Radeke, C., & Khinast, J. G. (2011). An integrated Quality by Design
(QbD) approach towards design space definition of a blending unit operation by
discrete element method (DEM) simulation. European Journal of Pharmaceutical
Sciences, 42, 106–115.
Ahmad, B. S., & Barton, P. I. (1999). Process-wide integration of solvent mixtures.
Computers & Chemical Engineering, 23, 1365–1380.
Akkisetty, P. K., Lee, U., Reklaitis, G. V., & Venkatasubramanian, V. (2010). Popula-
tion balance model-based hybrid neural network for a pharmaceutical milling
process. Journal of Pharmaceutical Innovation, 5, 161–168.
 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
Alonso, D., Velazquez, A., & Vega, J. M. (2009). A method to generate computation-
ally efficient reduced order models. Computer Methods in Applied Mechanics and
Engineering, 198, 2683–2691.
am Ende, M. T., Moses, S. K., Carella, A. J., Gadkari, R. A., Graul, T. W., Otano, A. L.,
et al. (2007). Improving the content uniformity of a low-dose tablet formula-
tion through roller compaction optimization. Pharmaceutical Development and
Technology, 12, 391–404.
Asprey, S. P., & Macchietto, S. (2000). Statistical tools for optimal dynamic model
building. Computers & Chemical Engineering, 24, 1261–1267.
Atkinson, A. C., & Donev, A. N. (1992). Optimum experimental designs. Oxford Univer-
sity Press.
Barthe, P., Guermeur, C., Lobet, O., Moreno, M., Woehl, P., Roberge, D. M., et al.
(2008). Continuous multi-injection reactor for multipurpose production – Part
I. Chemical Engineering Technology, 31, 1146–1154.
Bersimis, S., Psarakis, S., & Panaretos, J. (2006). Multivariate statistical process con-
trol charts: An overview. Quality and Reliability Engineering International, 23,
517–543.
Bhatia, T., & Biegler, L. T. (1996). Dynamic optimization in the design and scheduling
of multiproduct batch plants. Industrial & Engineering Chemistry Research, 35,
2234–2246.
Bilgili, E., & Scarlett, B. (2005). Population balance modeling of non-linear effects in
milling processes. Powder Technology, 153, 59–71.
Blayer, S., Woodley, J. M., Lilly, M. D., & Dawson, M. J. (1996). Characterization of the
chemoenzymatic synthesis of N-Acetyl-d-neuraminic acid (Neu5Ac). Biotech-
nology Progress, 12, 758–763.
Boukouvala, F., Muzzio, F. J., & Ierapetritou, M. G. (2010). Design space of pharma-
ceutical processes using data-driven-based methods. Journal of Pharmaceutical
Innovation, 5, 119–137.
Boukouvala, F., Muzzio, F. J., & Ierapetritou, M. G. (2011). Feasibility analysis
of black-box processes using an adaptive sampling kriging based method.
Computer-Aided Chemical Engineering, 29, 432–436.
Boukouvala, F., Ramachandran, R., Vanarase, A., Muzzio, F. J., & Ierapetritou, M.
G. (2011). Computer aided design and analysis of continuous pharmaceutical
manufacturing processes. Computer-Aided Chemical Engineering, 29, 216–220.
Boukouvala, F., Dubey, A., Vanarase, A., Ramachandran, R., Muzzio, F. J., & Ierapetri-
tou, M. (2012). Computational approaches for studying the granular dynamics of
continuous blending processes, 2 – Population balance and data-based method.
Macromolecular Materials and Engineering, 297, 9–19.
Camacho, J., Pico, J., & Ferrer, A. (2008). Bilinear modelling of batch processes. Part
II: A comparison of PLS soft-sensors. Journal of Chemometrics, 22, 533–547.
Capece, M., Bilgili, E., & Dave, R. (2011). Identification of the breakage rate and distri-
bution parameters in a non-linear population balance model for batch milling.
Powder Technology, 208, 195–204.
Carvalho, A., Gani, R., & Matos, H. (2008). Design of sustainable chemical processes:
Systematic retrofit analysis generation and evaluation of alternatives. Process
Safety and Environmental Protection, 86, 328–346.
Carvalho, A., Matos, H. A., & Gani, R. (2009). Design of batch operations: Systematic
methodology for generation and analysis of sustainable alternatives. Computers
& Chemical Engineering, 33, 2075–2090.
Cervera, A. E., Gani, R., Kiil, S., Skovby, T., & Gernaey, K. V. (2011). A systematic
methodology for the design of continuous active pharmaceutical ingredient
production processes. Computer-Aided Chemical Engineering, 29, 271–275.
Cervera-Padrell, A. E., Nielsen, J. P., Pedersen, M. J., Müller, K., Mortensen, A. R.,
Skovby, T., et al. (2012). Monitoring and control of a continuous Grignard
alkylation reaction for the synthesis of an active pharmaceutical ingredient
intermediate using in-line NIR spectroscopy. Organic Process Research and Devel-
opment, doi:10.1021/op2002563, in press.
Chakraborty, A., & Linninger, A. A. (2002). Plant-wide waste management. 1. Syn-
thesis and multiobjective design. Industrial & Engineering Chemistry Research, 41,
4591–4604.
Chen, J., & Trout, B. L. (2010). Computer-aided solvent selection for improving the
morphology of needle-like crystals: A case study of 2,6-dihydroxybenzoic acid.
Crystal Growth & Design, 10, 4379–4388.
Chen, B., & Woodley, J. M. (2002). Wavelet shrinkage data processing for neu-
ral networks in bioprocess modeling. Computers & Chemical Engineering, 26,
1611–1620.
Constable, D. J. C., Curzons, A. D., & Cunningham, V. L. (2002). Metrics to ‘green’
chemistry-which are the best? Green Chemistry, 4, 521–527.
Constable, D. J. C., Jiménez-González, C., & Henderson, R. K. (2007). Perspective on
solvent use in the pharmaceutical industry. Organic Process Research & Develop-
ment, 11, 133–137.
Curzons, A. D., Constable, D. J. C., Mortimer, D. N., & Cunningham, V. L. (2001). So you
think your process is green, how do you know? – Using principles of sustain-
ability to determine what is green – A corporate perspective. Green Chemistry,
3, 1–6.
Dassau, E., Zadok, I., & Lewin, D. R. (2006). Combining six-sigma with integrated
design and control for yield enhancement in bioprocessing. Industrial & Engi-
neering Chemistry Research, 45, 8299–8309.
Diedrichs, A., & Gmehling, J. (2011). Solubility calculation of active pharmaceuti-
cal ingredients in alkanes, alcohols, water and their mixtures using various
activity coefficient models. Industrial & Engineering Chemistry Research, 50,
1757–1769.
Dong, D., & McAvoy, T. J. (1996). Batch tracking using non-linear principal component
analysis. AIChE Journal, 42, 2199–2208.
Downs, J. J., & Vogel, E. F. (1993). A plant-wide industrial process control problem.
Computers & Chemical Engineering, 17, 245–255.
27
Doyle, F. J., Harrison, C. A., & Crowley, T. J. (2003). Hybrid model-based approach to
batch-to-batch control of particle size distribution in emulsion polymerization.
Computers & Chemical Engineering, 27, 1153–1163.
Dubey, A., Sarkar, A., Ierapetritou, M., Wassgren, C. R., & Muzzio, F. J. (2011). Compu-
tational approaches for studying the granular dynamics of continuous blending
processes, 1 – DEM based methods. Macromolecular Materials and Engineering,
296, 290–307.
Eriksson, L., Johansson, E., Kettaneh-Wold, N., Wikström, C., & Wold S. (2008). Design
of experiments. Principles and applications (3rd ed.) ISBN: 13-978-91-973730-4-3.
U.S. Food and Drug Administration (FDA) (2004). PAT guidance.
Federsel, H. (2003). Searching for scalable processes: Addressing the challenges in
times of increasing complexity. Current Opinion in Drug Discovery & Development,
6, 838–847.
Franceschini, G., & Macchietto, S. (2008). Model-based design of experiments
for parameter precision: State of the art. Chemical Engineering Science, 63,
4846–4872.
Fujiwara, M., Nagy, Z. K., Chew, J. W., & Braatz, R. D. (2005). First-principles and direct
design approaches for the control of pharmaceutical crystallization. Journal of
Process Control, 15, 493–504.
Gani, R., Gómez, P. A., Folić, M., Jiménez-González, C., & Constable, D. J. C. (2008).
Solvents in organic synthesis: Replacement and multi-step reaction systems.
Computers & Chemical Engineering, 32, 2420–2444.
García Muñoz, S., Dolph, S., & Ward, H. W., II. (2010). Handling uncertainty in the
establishment of a design space for the manufacture of a pharmaceutical prod-
uct. Computers & Chemical Engineering, 34, 1098–1107.
Gernaey, K. V., & Gani, R. (2010). A model-based systems approach to pharma-
ceutical product-process design and analysis. Chemical Engineering Science, 65,
5757–5769.
Gernaey, K. V., Eliasson Lantz, A., Tufvesson, P., Woodley, J. M., & Sin, G. (2010).
Application of mechanistic models to fermentation and biocatalysis for next
generation processes. Trends in Biotechnology, 28, 346–354.
Gerogiorgis, D. I., & Barton, P. I. (2009). Steady-state optimization of a continuous
pharmaceutical process. Computer-Aided Chemical Engineering, 27, 927–932.
Grossmann, I. (2005). Enterprise-wide optimization: A new frontier in process sys-
tems engineering. AIChE Journal, 51, 1846–1857.
Hartman, R. L., & Jensen, K. F. (2009). Microchemical systems for continuous-flow
synthesis. Lab on a Chip, 9, 2495–2507.
Hartman, R. L., Sahoo, H. R., Yen, B. C., & Jensen, K. F. (2009). Distillation in micro-
chemical systems using capillary forces and segmented flow. Lab on a Chip, 9,
1843–1849.
Hartman, R. L., Naber, J. R., Buchwald, S. L., & Jensen, K. F. (2010). Multistep micro-
chemical synthesis enabled by microfluidic distillation. Angewandte Chemie
International Edition, 49, 899–903.
Hartman, R. L., Naber, J. R., Zaborenko, N., Buchwald, S. L., & Jensen, K. F. (2010). Over-
coming the challenges of solid bridging and constriction during Pd-catalyzed
C N bond formation in microreactors. Organic Process Research & Development,
14, 1347–1357.
Hartman, R. L., McMullen, J. P., & Jensen, K. F. (2011). Deciding whether to go with the
flow: Evaluating the merits of flow reactors for synthesis. Angewandte Chemie
International Edition, 50, 7502–7519.
Henderson, R. K., Jiménez-González, C., Preston, C., Constable, D. J. C., & Woodley, J. M.
(2008). EHS & LCA assessment for 7-ACA synthesis: A case study for comparing
biocatalytic & chemical synthesis. Industrial Biotechnology, 4, 180–192.
Hermanto, M. W., Braatz, R. D., & Chiu, M.-S. (2011). Integrated batch-to-batch and
nonlinear model predictive control for polymorphic transformation in pharma-
ceutical crystallization. AIChE Journal, 57, 1008–1019.
Hessel, V. (2009). Novel process windows – Gate to maximizing process intensifica-
tion via flow chemistry. Chemical Engineering Technology, 32, 1655–1681.
Hsieh, D., Marchut, A. J., Wei, C., Zheng, B., Wang, S. S. Y., & Kiang, S. (2009).
Model-based solvent selection during conceptual process design of a new drug
manufacturing process. Organic Process Research & Development, 13, 690–697.
ICH. (2009). International conference on harmonisation of technical requirements for
registration of pharmaceuticals for human use ICH Harmonised Tripartite Guide-
line, Pharmaceutical Development. Step 4, August 2009
Jaksland, C. A., Gani, R., & Lien, K. M. (1995). Separation process design and synthesis
based on thermodynamic insights. Chemical Engineering Science, 50, 211–530.
Jensen, K. F. (2001). Microreaction engineering – Is small better? Chemical Engineer-
ing Science, 56, 293–303.
Jimenez-Gonzalez, C., Ponder, C. S., Broxterman, Q. B., & Manley, J. B. (2011). Using
the right green yardstick: Why process mass intensity is used in the pharma-
ceutical industry to drive more sustainable processes. Organic Process Research
& Development, 15, 912–917.
Jiménez-González, C., Curzons, A. D., Constable, D. J. C., & Cunningham, V. L. (2005).
Expanding GSK’s Solvent Selection Guide – Application of life cycle assessment
to enhance solvent selections. Clean Technologies and Environmental Policy, 7,
42–50.
Jiménez-González, C., Poechlauer, P., Broxterman, Q. B., Yang, B. S., am Ende, D., Baird,
J., et al. (2011). Key green engineering research areas for sustainable manufac-
turing: A perspective from pharmaceutical and fine chemicals manufacturers.
Organic Process Research & Development, 15, 900–911.
Kadlec, P., Gabrys, B., & Strandt, S. (2009). Data-driven soft sensors in the process
industry. Computers & Chemical Engineering, 33, 795–814.
Kadlec, P., Grbic, R., & Gabrys, B. (2011). Review of adaptation mechanisms for data-
driven soft sensors. Computers & Chemical Engineering, 35, 1–24.
Kahrs, O., & Marquardt, W. (2008). Incremental identification of hybrid process
models. Computers & Chemical Engineering, 32, 694–705.
28 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
Kang, L., Chung, B. G., Langer, R., & Khademhosseini, A. (2008). Microfluidics for
drug discovery and development: From target selection to product lifecycle
management. Drug Discovery Today, 13, 1–13.
Kashid, M. N., Agar, D. W., & Turek, S. (2007). CFD modelling of mass transfer with and
without chemical reaction in the liquid–liquid slug flowmicroreactor. Chemical
Engineering Science, 62, 5102–5109.
Klatt, K., & Marquardt, W. (2009). Perspectives for process systems engineering –
Personal views from academia and industry. Computers & Chemical Engineering,
33, 536–550.
Koch, K., van Weerdenburg, B. J. A., Verkade, J. M. M., Nieuwland, P. J., Rutjes, F. P. J. T.,
& van Hest, J. C. M. (2009). Optimizing the deprotection of the amine protecting
p-methoxyphenyl group in an automated microreactor platform. Organic Process
Research & Development, 13, 1003–1006.
Kockmann, N., & Roberge, D. M. (2009). Harsh reaction conditions in continuous-flow
microreactors for pharmaceutical production. Chemical Engineering Technology,
32, 1682–1694.
Kockmann, N., Gottsponer, M., & Roberge, D. M. (2011). Scale-up concept of single-
channel microreactors from process development to industrial production.
Chemical Engineering Journal, 167, 718–726.
Kourti, T. (2006). Process analytical technology beyond real-time analyzers: The role
of multivariate analysis. Critical Reviews in Analytical Chemistry, 36, 257–278.
Kralj, J. G., Sahoo, H. R., & Jensen, K. F. (2007). Integrated continuous microfluidic
liquid–liquid extraction. Lab on a Chip, 7, 256–263.
Krasnyk, M., Mangold, M., & Kienle, A. (2010). Reduction procedure for parametrized
fluid dynamics problems based on proper orthogonal decomposition and cali-
bration. Chemical Engineering Science, 65, 6238–6246.
Krasnyk, M., Mangold, M., Ganesan, S., & Tobiska, L. (2012). Numerical reduction of a
crystallizer model with internal and external coordinates by proper orthogonal
decomposition. Chemical Engineering Science, 70, 77–86.
Kremer, D. M., & Hancock, B. C. (2006). Process simulation in the pharmaceutical
industry: A review of some basic physical models. Journal of Pharmaceutical
Sciences, 95, 517–529.
Kruger, U., Zhang, J., & Xie, L. (2008). Developments and applications of nonlinear
principal component analysis – A review. Lecture Notes in Computational Science
and Engineering, 58, 1–43.
Lang, Y., Malacina, A., Biegler, L. T., Munteanu, S., Madsen, J. I., & Zitney, S. E. (2009).
Reduced order model based on principal component analysis for process simu-
lation and optimization. Energy & Fuels, 23, 1695–1706.
LaPorte, T. L., & Wang, C. (2007). Continuous processes for the production of pharma-
ceutical intermediates and active pharmaceutical ingredients. Current Opinion
in Drug Discovery & Development, 10, 738–745.
Leardi, R., & Nørgaard, L. (2004). Sequential application of backward interval par-
tial least squares and genetic algorithms for the selection of relevant spectral
regions. Journal of Chemometrics, 18, 486–497.
Lencastre Fernandes, R., Nierychlo, M., Lundin, L., Pedersen, A. E., Puentes Tellez,
P. E., Dutta, A., et al. (2011). Experimental methods and modeling techniques
for description of cell population heterogeneity. Biotechnology Advances, 29,
575–599.
Linninger, A. A., & Malcolm, A. (2006). Pollution prevention for batch pharmaceutical
and specialty chemical processes. In E. Korovessi, & A. A. Linninger (Eds.), Batch
processes. Boca Raton, FL, USA: CRC Press, Taylor & Francis Group.
MacGregor, J. F., & Bruwer, M. J. (2008). A framework for the development of design
and control spaces. Journal of Pharmaceutical Innovation, 3, 15–22.
Mandenius, C., & Brundin, A. (2008). Bioprocess optimization using design-of-
experiments methodology. Biotechnology Progress, 24, 1191–1203.
Marchetti, A., Srinivasan, B., Bonvin, D., Elgue, S., Prat, L., & Cabassud, M. (2005).
Measurement-based run-to-run optimization of a batch reaction-distillation
system. Computer-Aided Chemical Engineering, 20, 1417–1422.
Marquardt, W. (2005). Model-based experimental analysis of kinetic phenomena
in multi-phase reactive systems. Chemical Engineering Research and Design, 83,
561–573.
Marquardt, W., Morbach, J., Wiesner, A., & Yang, A. (2010). OntoCAPE: A re-usable
ontology for chemical process engineering. Springer., ISBN 978-3-642-04654-4.
Matsuda, H., Kaburagi, K., Kurihara, K., Tochigi, K., & Tomono, K. (2010). Prediction of
solubilities of pharmaceutical compounds in water + co-solvent systems using
an activity coefficient model. Fluid Phase Equilibria, 290, 153–157.
McMullen, J. P., & Jensen, K. F. (2010a). An automated microfluidic system for online
optimization in chemical synthesis. Organic Process Research & Development, 14,
1169–1176.
McMullen, J. P., & Jensen, K. F. (2010b). Integrated microreactors for reaction automa-
tion: New approaches to reaction development. Annual Review of Analytical
Chemistry, 3, 19–42.
McMullen, J. P., & Jensen, K. F. (2011). Rapid determination of reaction kinetics with
an automated microfluidic system. Organic Process Research & Development, 15,
398–407.
Mendorf, M., Nachtrodt, H., Mescher, A., Ghaini, A., & Agar, D. W. (2010). Design and
control techniques for the numbering-up of capillary microreactors with uni-
form multiphase flow distribution. Industrial & Engineering Chemistry Research,
49, 10908–10916.
Michaut, F., Busignies, V., Fouquereau, C., Huet de Barochez, B., Leclerc, B., &
Tchoreloff, P. (2010). Evaluation of a rotary tablet press simulator as a tool for the
characterization of compaction properties of pharmaceutical products. Journal
of Pharmaceutical Sciences, 99, 2874–2885.
Mills, P. L., Quiram, D. J., & Ryley, J. F. (2007). Microreactor technology and process
miniaturization for catalytic reactions – A perspective on recent developments
and emerging technologies. Chemical Engineering Science, 62, 6992–7010.
Morbach, J., Yang, A., & Marquardt, W. (2007). OntoCAPE – A large-scale ontology for
chemical process engineering. Engineering Applications of Artificial Intelligence,
20, 147–161.
Mortier, S. T. F. C., De Beer, T., Gernaey, K. V., Remon, J. P., Vervaet, C., & Nopens, I.
(2011). Mechanistic modelling of fluidized bed drying processes of wet porous
granules: A review. European Journal of Pharmaceutics and Biopharmaceutics, 79,
205–225.
Nagy, Z. K., Fujiwara, M., & Braatz, R. D. (2008). Modelling and control of combined
cooling and antisolvent crystallization processes. Journal of Process Control, 18,
856–864.
Nomikos, P., & McGregor, J. F. (1995). Multi-way partial least squares in monitoring
batch processes. Chemometrics and Intelligent Laboratory Systems, 30, 97–108.
Nørgaard, L., Saudland, A., Wagner, J., Nielsen, J. P., Munck, L., & Engelsen, S. B. (2000).
Interval Partial Least-Squares Regression (iPLS): A comparative chemometric
study with an example from near-infrared spectroscopy. Applied Spectroscopy,
54, 413–419.
O’Connell, J. P., Gani, R., Mathias, P. M., Maurer, G., Olson, J. D., & Crafts, P. A.
(2009). Thermodynamic property modeling for chemical process and product
engineering: Some perspectives. Industrial & Engineering Chemistry Research, 48,
4619–4637.
Paakkunainen, M., Matero, S., Ketolainen, J., Lahtela-Kakkonen, M., Poso, A., &
Reinikainen, S.-P. (2009). Uncertainty in dissolution test of drug release. Chemo-
metrics and Intelligent Laboratory Systems, 97, 82–90.
Plumb, K. (2005). Continuous processing in the pharmaceutical industry – Changing
the mindset. Chemical Engineering Research and Design, 83, 730–738.
Pollard, D. J., & Woodley, J. M. (2007). Biocatalysis for pharmaceutical intermediates:
The future is now. Trends in Biotechnology, 25, 66–73.
Pollet, P., Cope, E. D., Kassner, M. K., Charney, R., Terett, S. H., Richman, K. W., et
al. (2009). Production of (S)-1-benzyl-3-diazo-2-oxopropylcarbamic acid tert-
butyl ester, a diazoketone pharmaceutical intermediate, employing a small scale
continuous reactor. Industrial & Engineering Chemistry Research, 48, 7032–7036.
Poon, J. M. H., Ramachandran, R., Sanders, C. F. W., Glaser, T., Immanuel, C. D., Doyle,
F. J., III, et al. (2009). Experimental validation studies on a multi-dimensional and
multi-scale population balance model of batch granulation. Chemical Engineering
Science, 64, 775–786.
Pordal, H. S., Matice, C. J., & Fry, T. J. (2002). The role of computational fluid dynamics
in the pharmaceutical industry. Pharmaceutical Technology, 26(2), 72–77.
Prpich, A., am Ende, M. T., Katzschner, T., Lubczyk, V., Weyhers, H., & Bernhard, G.
(2010). Drug product modeling predictions for scale-up of tablet film coating – A
quality by design approach. Computers & Chemical Engineering, 34, 1092–1097.
Radeke, C. A., Glasser, B. J., & Khinast, J. G. (2010). Large-scale powder mixer simula-
tions using massively parallel GPU architectures. Chemical Engineering Science,
65, 6435–6442.
Ramachandran, R., Immanuel, C. D., Stepanek, F., Litster, J. D., & Doyle, F. J., III. (2009).
A mechanistic model for breakage in population balances of granulation: The-
oretical kernel development and experimental validation. Chemical Engineering
Research and Design, 87, 598–614.
Reginato, G., Sadler, P., & Wilkes, R. D. (2011). Scaling up metal scavenging opera-
tions for pharmaceutical pilot plant manufactures. Organic Process Research &
Development, 15, 1396–1405.
Reich, G. (2005). Near-infrared spectroscopy and imaging: Basic principles and phar-
maceutical applications. Advanced Drug Delivery Reviews, 57, 1109–1143.
Reklaitis, G. V. R. (2007). Perspectives on Process Systems Engineering R&D in
support of pharmaceutical product/process development and manufacturing.
Computer-Aided Chemical Engineering, 24, 35–38.
Remy, B., Khinast, J. G., & Glasser, B. J. (2009). Discrete element simulation of free
flowing grains in a four-bladed mixer. AIChE Journal, 55, 2035–2048.
Remy, B., Glasser, B. J., & Khinast, J. G. (2010). The effect of mixer properties and fill
level on granular flow in a bladed mixer. AIChE Journal, 56, 336–353.
Roberge, D. M., Ducry, L., Bieler, N., Cretton, P., & Zimmermann, B. (2005). Microre-
actor technology: A revolution for the fine chemical and pharmaceutical
industries? Chemical Engineering & Technology, 28, 318–323.
Rohner, M., & Meyer, H.-P. (1995). Applications of modelling for bioprocess design
and control in industrial production. Bioprocess and Biosystems Engineering, 13,
69–78.
Rosas, J. G., Blanco, M., González, J. M., & Alcalá, M. (2011a). Quality by design
approach of a pharmaceutical gel manufacturing process. Part 1: Determination
of the design space. Journal of Pharmaceutical Sciences, 100, 4432–4441.
Rosas, J. G., Blanco, M., González, J. M., & Alcalá, M. (2011b). Quality by design
approach of a pharmaceutical gel manufacturing process. Part 2: Near infrared
monitoring of composition and physical parameters. Journal of Pharmaceutical
Sciences, 100, 4442–4451.
Sahoo, H. R., Kralj, J. G., & Jensen, K. F. (2007). Multistep continuous-flow microchem-
ical synthesis involving multiple reactions and separations. Angewandte Chemie,
119, 5806–5810.
Sales-Cruz, M., & Gani, R. (2003). S. P. Asprey, & S. Macchietto (Eds.). A modelling
tool for different stages of the process life. Computer-Aided Chemical Engineering,
16, 209–249.
Saling, P., Kicherer, A., Dittrich-Krämer, B., Wittlinger, R., Zombik, W., Schmidt, I., et
al. (2002). Eco-efficiency analysis by BASF: The method. The International Journal
of Life Cycle Assessment, 7, 203–218.
Savile, C. K., Janey, J. M., Mundorff, E. C., Moore, J. C., Tam, S., Jarvis, W. R., et al.
(2010). Biocatalytic asymmetric synthesis of chiral amines from ketones applied
to sitagliptin manufacture. Science, 329, 305–309.
Schaber, S. D., Gerogiorgis, D. I., Ramachandran, R., Evans, J. M. B., Barton, P. I.,
& Trout, B. L. (2011). Economic analysis of integrated continuous and batch
 K.V. Gernaey et al. / Computers and Chemical Engineering 42 (2012) 15–29
pharmaceutical manufacturing: A case study. Industrial & Engineering Chemistry
Research, 50, 10083–10092.
Seborg, D. E., Mellichamp, D. A., Edgar, T. F., & Doyle, F. J., III. (2010). Process dynamics
and control (3rd ed.). Wiley and Sons.
Sharma, C., Malhotra, D., & Rathore, A. S. (2011). Review of computational fluid
dynamics applications in biotechnology processes. Biotechnology Progress, 27,
1497–1510.
Sheldon, R. A. (1992). Organic synthesis – Past, present and future. Chemistry and
Industry, 23, 903–906.
Sheldon, R. A. (2007). The E factor: Fifteen years on. Green Chemistry, 9, 1273–1283.
Shonnard, D. R., Kicherer, A., & Saling, P. (2003). Industrial applications using BASF
eco-efficiency analysis: Perspectives on green engineering principles. Environ-
mental Science & Technology, 37, 5340–5348.
Sin, G., Ödman, P., Petersen, N., Eliasson Lantz, A., & Gernaey, K. V. (2008). Matrix
notation for efficient development of first-principles models within PAT applica-
tions: Integrated modeling of antibiotic production with Streptomyces coelicolor.
Biotechnology and Bioengineering, 101, 153–171.
Sin, G., Gernaey, K. V., & Eliasson Lantz, A. (2009). Good modelling practice (GMoP)
for PAT applications: Propagation of input uncertainty and sensitivity analysis.
Biotechnology Progress, 25, 1043–1053.
Singh, R., Gernaey, K. V., & Gani, R. (2009). Model-based computer aided framework
for design of process monitoring and analysis systems. Computers & Chemical
Engineering, 33, 22–42.
Singh, R., Gernaey, K. V., & Gani, R. (2010a). An ontological knowledge based system
for selection of process monitoring and analysis tools. Computers & Chemical
Engineering, 34, 1137–1154.
Singh, R., Gernaey, K. V., & Gani, R. (2010b). ICAS-PAT: A software for design,
analysis and validation of PAT systems. Computers & Chemical Engineering, 34,
1108–1136.
Singh, R., Rozada-Sanchez, R., Wrate, T., Muller, F., Gernaey, K. V., Gani, R., et al.
(2011). A retrofit strategy to achieve fast, flexible, future (F3) pharmaceutical
production processes. Computer-Aided Chemical Engineering, 29, 291–295.
Stephanopoulos, G., & Reklaitis, G. V. (2011). Process systems engineering: From
Solvay to modern bio- and nanotechnology: A history of development, successes
and prospects for the future. Chemical Engineering Science, 66, 4272–4306.
Straathof, A. J. J., Panke, S., & Schmid, A. (2002). The production of fine chemicals by
biotransformations. Current Opinion in Biotechnology, 13, 548–556.
Taskinen, J., & Yliruusi, J. (2003). Prediction of physicochemical properties based on
neural network modelling. Advanced Drug Delivery Reviews, 55, 1163–1183.
Tucker, J. L. (2006). Green chemistry, a pharmaceutical perspective. Organic Process
Research & Development, 10, 315–319.
Tufvesson, P., Lima-Ramos, J., Jensen, J. S., Al-Haque, N., Neto, W., & Woodley, J. M.
(2011). Process considerations for the asymmetric synthesis of chiral amines
using transaminases. Biotechnology and Bioengineering, 108, 1479–1493.
Tufvesson, P., Lima-Ramos, J., Nordblad, M., & Woodley, J. M. (2011). Guidelines and
cost analysis for catalyst production in biocatalytic processs. Organic Process
Research & Development, 15, 266–274.
Vasić-Rački, D., Findrik, Z., & Vrsalović Presečki, A. (2011). Modelling as a tool of
enzyme reaction engineering for enzyme reactor development. Applied Micro-
biology and Biotechnology, 91, 845–856.
29
Venkatasubramanian, V., Rengaswamy, R., Yin, K., & Kavuri, S. N. (2003). A review of
process fault detection and diagnosis. Part I: Quantitative model-based methods.
Computers & Chemical Engineering, 27, 293–311.
Venkatasubramanian, V., Rengaswamy, R., & Kavuri, S. N. (2003). A review of process
fault detection and diagnosis. Part II: Qualitative models and search strategies.
Computers & Chemical Engineering, 27, 313–326.
Venkatasubramanian, V., Rengaswamy, R., Kavuri, S. N., & Yin, K. (2003). A review of
process fault detection and diagnosis. Part III: Process history based methods.
Computers & Chemical Engineering, 27, 327–346.
Venkatasubramanian, V., Zhao, C., Joglekar, G., Jain, A., Hailemariam, L., Suresh, P.,
et al. (2006). Ontological informatics infrastructure for pharmaceutical prod-
uct development and manufacturing. Computers & Chemical Engineering, 30,
1482–1496.
Vervaet, C., & Remon, J. P. (2005). Continuous granulation in the pharmaceutical
industry. Chemical Engineering Science, 60, 3949–3957.
Villez, K., Srinivasan, B., Rengaswamy, R., Narasimhan, S., & Venkatasubramanian,
V. (2011). Kalman-based strategies for Fault Detection and Identification (FDI):
Extensions and critical evaluation for a buffer tank system. Computers & Chemical
Engineering, 35, 806–816.
Wassgren, C., & Curtis, J. S. (2006). The application of computational modeling to
pharmaceutical materials science. MRS Bulletin, 31, 900–904.
Webb, D., & Jamison, T. F. (2010). Continuous flow multi-step organic synthesis.
Chemical Science, 1, 675–680.
Wei, H., Zhou, W., & Garside, J. (2001). Computational fluid dynamics modeling of
the precipitation process in a semibatch crystallizer. Industrial & Engineering
Chemistry Research, 40, 5255–5261.
Weis, D. C., & Visco, D. P. (2010). Computer-aided molecular design using the
signature molecular descriptor: Application to solvent selection. Computers &
Chemical Engineering, 34, 1018–1029.
Wold, S., Sjöström, M., & Eriksson, L. (2001). PLS-regression: A basic tool of chemo-
metrics. Chemometrics and Intelligent Laboratory Systems, 58, 109–130.
Woo, X. Y., Tan, R. B. H., & Braatz, R. D. (2009). Modeling and computational fluid
dynamics-population balance equation-micromixing simulation of impinging
jet crystallizers. Crystal Growth & Design, 9, 156–164.
Woodley, J. M. (2008). New opportunities for biocatalysis: Making pharmaceutical
processes greener. Trends in Biotechnology, 26, 321–327.
Wu, Y. X., Wang, X., & Ching, C. B. (2002). Computational fluid dynamics simulation
of the adsorption separation of three components in high performance liquid
chromatography. Chromatographia, 55, 439–445.
Zamprogna, E., Barolo, M., & Seborg, D. E. (2004). Estimating product composition
profiles in batch distillation via partial least squares regression. Control Engi-
neering Practice, 12, 917–929.
Zhang, J., Martin, E. B., & Morris, A. J. (1997). Process monitoring using non-linear
statistical techniques. Chemical Engineering Journal, 67, 181–189.
Zimmermann, V., Hennemann, H., Daußmann, T., & Kragl, U. (2007). Modelling
the reaction course of N -acetylneuraminic acid synthesis from N -acetyl-d -
glucosamine – New strategies for the optimisation of neuraminic acid synthesis.
Applied Microbiology and Biotechnology, 76, 597–605.