chapter 5 optimization examiner correction
TRANSCRIPT
Chapter FiveChapter FiveChapter FiveChapter Five
OPTIMAL OPERATING STRATEGIES
OF EMULSION
TERPOLYMERISATION PRODUCTS
ABSTRACT .................................................................................................5.1
5.1 INTRODUCTION ...................................................................................5.1
5.2 DYNAMIC OPTIMISATION REVIEW ....................................................5.2 5.2.1 Formulation of Dynamic Optimisation ......................................................5.5
5.2.2 gPROMS Optimisation............................................................................5.11
5.3 OPTIMAL CONTROL OBJECTIVES ..................................................5.13 5.3.1 Particle Size Optimisation .......................................................................5.14
5.3.2 Molecular Weight Distribution Optimisation...........................................5.17
5.3.3 Composition Optimisation.......................................................................5.18
5.4 OPTIMISATION PROCEDURE ...........................................................5.20
5.5 EXPERIMENTAL VALIDATION OF OPTIMISATION PROFILES ......5.21 5.5.1 Optimising Particle Size..........................................................................5.22
5.5.2 Optimising Molecular Weight .................................................................5.26
5.5.3 Optimising Composition .........................................................................5.31
5.6 CONCLUSION.....................................................................................5.32
BIBLIOGRAPHY .......................................................................................5.34
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 1
Optimal Operating Strategies for Emulsion Terpolymerisation Products
ABSTRACT
In this chapter, the dynamic optimisation of emulsion terpolymerisation is presented.
A literature review for the emulsion polymerisation optimisations is firstly addressed.
The theory of process optimisation especially the dynamic one is then explained. The
different problem formulations and solution techniques in gPROMS are discussed.
Off-line dynamic optimisations application to the emulsion terpolymerisation process
of styrene, MMA and MA are investigated for the PSD, MWD and polymer
composition control. The obtained optimal profiles are then validated experimentally.
5.1 INTRODUCTION
Optimisation plays a crucial role in the design of industrial processes, due to the
desire to save valuable resources, meet ambitious production goals, and achieve the
best possible profit margins. The polymer manufacturing and characterisation has
attracted the focus of numerous industrial companies and research institutes. Polymer
manufacturer’s face increasing pressures for production cost reductions and stringent
quality requirements. Many applications of polymer latexes, which are mainly
produced through emulsion polymerisation, require specific physical and mechanical
properties, such as in paints, adhesives and paper coating. These properties are
mainly controlled by their chemical composition, particle size distribution (PSD) and
molecular weight distribution (MWD). Chemical composition of terpolymers for
example determines the glass transition temperature (Tg). The particle size and size
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 2
distribution affect the dispersion rheology which is critical in achieving the desired
surface finish in coating applications (Anderson and Rapra Technology Limited
(GB), 2003). Also content adhesives require bimodal or multimodal particle size
distributions to maintain a manageable viscosity (Anderson and Rapra Technology
Limited (GB), 2003). MWD has a significant effect on the important end-use
properties of the polymer film such as for adhesion, elasticity, strength, stress-strain
relationships, toughness and solvent resistance(Vicente et al., 2002). Advanced
process optimisation techniques allow the achievement of these requirements as well
as productivity/quality improvements to the chemical process. Therefore, there are
strong incentives to develop strategies to optimise and control the PSD, MWD and
polymer composition.
5.2 DYNAMIC OPTIMISATION REVIEW
Dynamic optimisation is the process of optimising some aspects of process design
and performance taking into account both steady-state and dynamic considerations.
Dynamic optimisation allows operability considerations to be taken into account at
the design stage. This method is applicable to continuous, batch- and semibatch
processes. Dynamic optimisation, in contrast with simulation, allows in formulating
the problem to be solved directly, and then useing the start-of-the-art numerical and
optimisation solvers to determine the optimal values or time profiles of the
optimisation variables.
Several studies have been reported on the optimisation and control of polymerisation
reactors, and these have been recently reviewed by (Cheremisinoff, 1989,
Chakravarthy et al., 1997, Mitra et al., 1998, Kiparissides, 2006). Most of these
studies are on the optimisation and control of lumped properties (Arzamendi et al.,
1992, Kozub and Macgregor, 1992, Crowley et al., 2000, Canu et al., 1994, Saldivar
and Ray, 1997, Sayer et al., 2001b, Vicente et al., 2002, Schoonbrood et al., 1996,
Vicente et al., 2001) and deal with homo- and copolymerisation rather than
terpolymerisation. Table 5.1 which was reported by (Kiparissides, 2006), depicts the
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 3
most recent publications on monitoring, estimation, optimisation and control of
particulate polymerisation processes.
The major issues regarding the control of emulsion polymerisation reactors was
summarized by (Dimitratos et al., 1994). A hybrid modelling strategy for batch - to -
batch optimisation of PSD for emulsion polymerisation was presented by (Crowley
et al., 2001) and (Doyle et al., 2003). An open loop optimisation study for the
control of the PSD in a semibatch emulsion copolymerisation of vinyl acetate and
butyl acrylate was described by (Immanuel and Doyle, 2002). Semino and Ray
(Semino and Ray, 1995b, Semino and Ray, 1995a) studied the controllability of PSD
in emulsion polymerisation reactors using lumped manipulated variables. They found
that there are appropriate strategies for the control of the PSD through the
manipulation of surfactant, initiator and inhibitor feed concentrations. Models for
emulsion homo- and co-polymerisation were developed by (Zeaiter et al., 2002,
Alhamad et al., 2005) respectively to optimise and control the PSD.
Approaches for the calculation of the optimal monomer addition policies for polymer
composition control in emulsion terpolymerisation are limited (Arzamendi et al.,
1992, Schoonbrood et al., 1996). (Congalidis et al., 1989) presented a feed forward-
feedback control system to regulate the polymer production rate, copolymer
composition, molecular weight and reactor temperature in a simulated solution
copolymerisation of methyl-methacrylate/ vinyl acetate. (Saldivar and Ray, 1997)
presented the control of copolymer composition and averaged molecular weight for
semi-continuous emulsion polymerisation. (Clarke-Pringle and MacGregor, 1998)
developed a batch to batch adjustment strategy for the control of molecular weight
distribution to re-optimise the inputs for the next batch based on end-point
measurement of the MWD.
An iterative dynamic programming (IDP) procedure was applied by (Sayer et al.,
2001b) to optimise the MWD and polymer composition during isothermal
semicontinuous emulsion polymerisation. Bojkov and Luus (Bojkov and Luus,
1995, Bojkov and Luus, 1994a, Bojkov and Luus, 1994b) used intervals of varying
lengths in IDP, and observed that accurate switching times can be obtained, as is
required for time optimal control problems.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 4
The optimisation of polymerisation processes may often present conflicting criteria
when multi-objective optimisation for polymerisation reactors is designed. This is the
main motivation of several studies dealing with the use of multi-objective
optimisation for polymerisation reactors (Gupta and Gupta, 1999, Garg and Gupta,
1999, Sareen and Gupta, 1995, Mitra et al., 1998, Mitra et al., 2004, Bhaskar et al.,
2001, Bhaskar et al., 2000, Yee et al., 2003). Bhaskar et al.(Bhaskar et al., 2001,
Bhaskar et al., 2000) pointed out that the multi-objective optimisation of real-life
systems is quite complex and each new application may require the development of
several adaptations of optimisation algorithms to obtain meaningful solutions,
irrespective of which mathematical procedure is used for the purpose.
In this work, the optimisation of semibatch emulsion terpolymerisation reactors was
investigated. The optimisation is performed by IDP procedure through gPROMS
with variable time intervals and is based on our detailed mechanistic model (Srour et
al., 2005) to develop an advanced control strategy for the optimal operation of the
reactor. The major objective of our optimisations is to obtain on a profile of optimal
strategy that would calculate set-point trajectories for the manipulated variables in
order to ensure the production of a polymer with a desired PSD, MWD or terpolymer
composition in the minimum reaction time. The optimisation procedure was applied
to styrene, methyl methacrylate and methyl acrylate emulsion terpolymerisation. Six
variables were used as manipulated variables: styrene monomer feed rate, MMA
monomer feed rate, MA monomer feed rate, surfactant feed rate, initiator feed rate,
and the temperature of the reactor.
The strategy of implementing optimal trajectories obtained from off-line optimisation
problem to corrected process has drawbacks. In the presence of disturbances, such as
changes in the reactor temperature or contamination with inhibitors/impurities, the
optimal trajectories may produce sub-optimal results. In our work, the experimental
validation was carried out in open loop mode with precise reaction procedure and
tuning to minimise the effect of disturbances. Consequently, this work represents an
introduction to model based control and is the first step in the advanced control
hierarchy adopted.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 5
Table 5.1 Selective publications on monitoring, estimation, optimisation and control
of polymer quality (Kiparissides, 2006).
Author Test case Comments
Congalidis et al. So: MMA/VAc Control: MWD/CCD
Ellis et al. So: MMA Estimation/control: MWD
Embirucu et al. E/S/C Modelling/estimation/optimisation/control: MW/CCD
Saldivar and Ray E: MMA/Vac Control: MWD/CCD
Yabuki and MacGregor E: SBR Control: MWD
Crowley and Choi So: MMA Optimisation/control: MWD
Crowley and Choi So: MMA Optimisation/control: MWD
Pringle and MacGregor So: St Optimisation/MWD
Crowley and Choi So: St/MMA Optimisation/control: MWD/CCD
Yoo et al. So: St Control: MWD
Sayer et al. E: MMA/BuA Control: MWD/CCD
Santos et al. E: MMA/VAc Modelling/estimation: CCD
Stavropoulos et al. E: St/2-EHA Modelling: CCD
Sayer et al. E: MMA/BuA Optimisation: MWD/CCD
Vicente et al. E: St/BuA Control: MWD/CCD
Chang and Hung So: MMA Optimisation: MWD
Park et al. So: MMA/MA Modelling/estimation: MWD/CCD
Valappil and Georgakis E: St Control: MWD/PSD
Kiparissides et al. S: MMA Optimisation/control: MWD
Catalgil-Giz et al. S: MMA/St Monitoring: MWD/CCD
Chatzidoukas et al. C: Olefin Optimisation: MWD/CCD
Dimitratos et al. E:acryl/vinyl/St Control: MWD/CCD/PSD
Semino and Ray E Control: PSD
Ohmura et al. E: VAc Control: PS
Fevotte et al. E: MMA/VAc Modelling/estimation: PSD
Crowley et al. E: St Control: PSD
Christofides E Control: PSD
Cerrillo and MacGregor E: St Control: PSD
Immanuel et al. E: Vac/BuA Estimation: PSD
Immanu el and Doyle E: VAc/BuA Optimisation/control: PSD
Doyle et al. E: St Control: PSD
Process C: catalytic; E: emulsion; S: suspension; So: solution.
5.2.1 Formulation of Dynamic Optimisation
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 6
Three main issues need to be addressed in the development of an optimisation/
control policy for a polymer process. Firstly, an understanding of the process
dynamics via mechanistic or data-based modelling and experimentation. Secondly,
the ability to measure and characterize a range of polymer quality variables via on-
line sensors or nonlinear estimation algorithms. Thirdly, the development of
nonlinear model-based predictive controllers with emphasis on achieving superior
performance and constraint handling. Thus, the following steps were used to
optimise our process:
• Development a reliable process model. The key to building a reliable model
is to mathematically describe the chemical and physical phenomena involved
in a process and to derive the necessary material, energy and momentum
balances. These mathematical descriptions involve nonlinear algebraic and
differential equations of a dynamic polymerisation model are strongly related.
• Identify the process variables which can be manipulated and controlled.
• Develop the objective functions. This model involves an equation that
represents the profit made from the sale of products and costs associated with
their production, such as raw materials, operating costs, fixed costs, taxes,
etc. The structure and complexity of the equations for the economic model
and process or plant constraints are very important, since most mathematical
programming procedures take advantage of the mathematical form of these
models. Linear programming is used when all of the equations are linear,
while geometric programming is used when all of the equations are
polynomials.
• Adjust a suitable optimisation procedure. This procedure locates the values of
the independent variables of the process to produce the maximum profit or
minimum cost as measured by the economic model. The procedure also
includes the constraints in materials, process equipment and manpower.
The mathematical model that was used for optimisation purposes is based on the
detailed dynamic model (Srour et al., 2007). Based on this comprehensive model, a
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 7
multi-layer model-based framework was developed and implemented within an
industrial distributed control system environment, to optimise the PSD, MWD and
terpolymer composition. The expected form of the optimisation algorithm is one that
would calculate set-point trajectories for the manipulated variables, so as to ensure
the production of a polymer with a defined PSD, MWD and terpolymer composition
in the minimum reaction time. The feed rates of different processes, the length of
each interval and temperature profile are the manipulated variables. The off-line
optimal trajectory obtained in this work will be used as a set point within an on-line
multivariable model predictive control scheme to control the PSD, MWD and
polymer composition.
The particle size evolution equations are a set of three coupled partial 2-D integro-
differential equations in radius and time. Also the molecular weight equations are a
set of partial-integro-differential equations in chain length and time. To overcome
this problem, the discretisation technique is used. The particle size evolution
equations and the molecular weight equations are explained and formulated in
chapter three.
The continuous dynamic system is a process model which contains a combination of
differential and algebraic equations of the form:
( ) 0)),(),(),(),( =vtutytxtxf & (5.1)
where x(t) and y(t) are differential and algebraic variables of the model, respectively.
)(tx& is the time derivative of variable x(t), and u(t) is a vector of time-variant control
variables and v is the vector of time-invariant parameters which to be determined by
optimisation. This model description is also subject to the following initial
conditions:
( ) 0)),0(),0(),0(),0( =vuyxxI & (5.2)
The optimisation function (gOPT code in gPROMS) seeks to determine the time
horizon of the process, tf, the time invariant parameters and the dynamic variation of
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 8
control variables over the whole time horizon, so that the final value of the objective
function ,z, being one of either differential or algebraic variables of the system, to be
maximized or minimised. The mathematical description of the proposed dynamic
optimisation problem (DOP) can be defined as:
)]([],0[),(,,
ftttuvt
tzOptimizeff ∈
(5.3)
This is subject to Eq.s. (5.1) and (5.2) as model description and initial conditions.
When the time variations of the controls (u) and the values of any time invariant
parameters [tk-1, tk] are fixed, the modelling equations together with the initial
conditions completely determine the transient response of the system by performing
a dynamic simulation. The solutions of dynamic optimisation will determine:
• The time horizon (tf).
• The values of the time invariant parameters (v).
• The time variation of the control variables, u(t), over the entire time horizon
],0[ ftt ∈
It is important to note that the maximisation problem is equivalent to the
minimisation problem if the sign of the objective function is reversed; that is,
)]([)]([],0[),(,,],0[),(,,
ftttuvt
ftttuvt
tzMintzMaxffff
−=∈∈ (5.4)
The vector of manipulated inputs u1, u2… um-1 are chosen to have the greatest
influence on the objective function. These variables must be successively adjusted,
during optimisation, to obtain the desired maximum or minimum. Each set of
adjustments to these variables is termed iteration, and in general a number of
iterations are required before an optimum is obtained. In the iterative algorithm a
first estimate of the decision variables must be supplied as a starting point. In the
context of semibatch emulsion terpolymerisation processes as systems studied here,
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 9
control variable can be PSPI, MWPI, Mn, process conversion and average radius of
the polymer particles, throughout the process, while time invariant parameters can be
feed rates of monomers, initiator, and surfactants ,and reactor temperature.
The optimisation problem often has more than one objective to be maximised or
minimised. The problem then is formulated as a multi-objective optimisation case.
Such problem is dealt with by reformulating the multi-objective problem as a single-
objective case by forming a weighted combination of the different objectives (i.e.
weighted-sum strategy) or else by placing some objectives in the form of constraints
which is called ε-constraint method (Choi and Butala, 1991). The optimal control
problem takes then the following form for the weighted-sum strategy:
...332211)(
+++ JJJMaxix
λλλ i=1,2,…,n (5.5)
Where J1, J2, J3,... are the objectives and λ1, λ2, λ3, ... are the corresponding weights.
Note that, because J is a vector, if any of the components of J are competing, there is
no unique solution to this problem. Instead, the concept of non-inferiority (also
called Pareto optimality) must be used to characterise the objectives. A non-inferior
solution is one in which an improvement in one objective requires a degradation of
another.
In the ε-constraint method, only one of the objectives (mainly the primary one) is
expressed in the cost function while the other objectives take inequality constraints
form.
),....,,( 211)(
nix
xxxJMax
Subject to (5.6)
ini xxxJ ε<),....,,( 21 i=1, 2,…, n
This approach is able to identify a number of non-inferior solutions on a non-convex
boundary that are not obtainable using the weighted sum technique. A problem with
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 10
this method is, however, a suitable selection of λ to ensure a feasible solution. A
further disadvantage of this approach is that the use of hard constraints is rarely
adequate for expressing true design objectives. There are several types of constraints
which are: Algebraic equality constraints which express the relationships connecting
the decision variables and thus reduces the degree of freedom in the system;
Algebraic inequality constraints which specify the practical operating limits of
certain variables within the process; Differential equality constraints which are
commonly used in chemical processes, where the rate of formation or consumption
of specie is a function of the state variables. Constrained optimisation is very much
more difficult than unconstrained optimisation and a great deal of effort has been
expended to reformulate constrained problems so that constraints are avoided.
For this, our system was subjected to some constraints including the constraints on
the time horizon and decision variables,
maxmin
fff tft ≤≤ (5.7)
],0[),()()( maxmin ftttututu ∈≤≤ (5.8)
maxmin vvv ≤≤ (5.9)
The constraints on the decision variables are stated explicitly since modern
optimization algorithms can handle them very efficiently. This is not generally the
case of other types of constraints. Then we have the end-point constraint variables,
which usually represent certain conditions that the process system must satisfy at the
end of the optimization horizon. For convenience, end-point constraints are divided
into inequality and equality constraints. Although, the latter are a special case of the
former, differentiating them simplifies the definition of some optimization problems.
maxmin )( effef wtww ≤≤ (5.10)
tgt
eefef wtw =)( (5.11)
Due to the transient nature of the emulsion polymerisation process and the inherent
nonlinearities in the system, non-linear constrained dynamic optimisation
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 11
formulation was used in this work. The optimisation was performed by solving the
comprehensive dynamic models on gPROMS software package (PSE, UK) in the
gOPT dynamic optimisation code. The dynamic formulation of chemical processes in
gPROMS software package is well explained in (PSE, 2004).
The gOPT code utilises the control vector parameterisation algorithm, employing
piecewise control variables. The main attraction of this method is due to its
suitability for solving large-scale dynamic optimisation problems. Five or more equal
horizons were identified for the six manipulating variables. These manipulating
variables remain constant during each horizon and move discretely from one horizon
to the next.
5.2.2 gPROMS Optimisation
The dynamic optimisation problem is set-up in gPROMS using a dynamic model and
a separate optimisation file. This file includes information on the time horizon,
objective function, form of the control variables and any constraints that need to be
imposed on the process.
The solution of dynamic optimisation problems is based on the classical calculus of
variations, the maximum principle of (Pontryagin et al., 1962), and the dynamic
programming of (Bellman and Kalaba, 1965). The most commonly used member of
the triad is the maximum principle as most computational techniques are concerned
with satisfying the maximum principle necessary conditions for an optimum. The
solution of the optimality conditions problem is generally obtained using the quasi-
linearisation approach (Miele, 1975) and through the use of multiple shooting
algorithms such as those proposed by (Bulirsch, 1971, Dixon and Bartholomew-
Biggs, 1981). Other techniques employ a discretisation approach whereby the
optimal control problem is converted to a non-linear problem (NLP) through the
discretisation of all variables. This can be done using the finite difference and
orthogonal collocations methods (Biegler, 1984).
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 12
The type of control variable to be implemented during the dynamic optimisation in
gPROMS requires specification. The ideal choice generally depends on the
capabilities of the control system from an engineering perspective rather than the
mathematical features of the problem. There are four common control variable
profiles in gPROMS which are: piecewise constant controls where the control
variable remains constant over a period of time before jumping discretely to a
different value for the next interval, piecewise linear controls which follow a linear
path over one time period before jumping discretely to a different linear variation for
the next interval, piecewise linear continuous controls which are the same as
piecewise linear controls, with the added constraint that their values be continuous
over the interval boundaries and polynomial controls which vary smoothly over time.
Dynamic optimisation in gPROMS facilitates piecewise constant and piecewise
linear controls. However, the remaining types of control are easily implemented by
adding additional equations to the model (PSE, 2004). For instance, piecewise
constant controls may often be preferable to other types as they are much easier to
implement and still perform the job efficiently.
The algorithm used in gPROMS, named gOPT code, employs the control vector
parameterisation (CVP) approach coupled with a multi-step backward-difference
method for the integration of the set of differential algebraic equations (Vassiliadis et
al., 1994b, Vassiliadis et al., 1994a, Vassiliadis et al., 1993). When piecewise-
constant approximation over equally spaced time intervals is made for the inputs, the
method is referred to as Control Vector Parameterization (CVP) in the literature
(Srinivasan et al., 2003). The quality of the solution via the CVP approach is strongly
dependent on the parameterization of the control profile. For applying CVP method,
five or more equal horizons were identified for the six manipulating variables. These
manipulating variables remain constant during each horizon and move discretely
from one horizon to the next. The time horizon, over which the process is optimised,
is partitioned into a pre-defined number of stages. The duration of each stage is
divided into a number of control intervals and the continuity of the differential
variables is enforced at the boundaries through simple junction conditions.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 13
The solution of the optimisation problem comprises the time horizon, the value of
time invariant parameters and the variation of the control variables over the time
horizon that minimise (or maximise) the objective function, while satisfying the
constraints. gPROMS outputs this data on screen as well as into four files:
� The optimisation report file which contains a summary report on the
optimisation run.
� The detailed results file contains the optimal time profiles of all variables in
the problem.
� The schedule file presents the solution in the form of a gPROMS schedule.
� The gOPT file contains the same information as the schedule file and is
generated at the end of iteration.
The detailed operating procedures of the dynamic optimisation processes within
gPROMS are presented in gPROMS user guide (PSE, 2004).
5.3 OPTIMAL CONTROL OBJECTIVES
The production of emulsion polymers by batch and semibatch processes is of
significant industrial importance. In many cases the amount of material that must be
produced does not justify the use of a continuous process so batch operation is
preferred. However, due to the non-steady nature of batch processes and the different
rates of reaction of the monomers in a recipe, these processes will produce polymers
whose properties change with the progress of the reaction. The instantaneous
terpolymer produced will exhibit a drift in terpolymer composition, particle size and
molecular weight. The end result is that the final terpolymer, which is a blend of all
the terpolymers produced instantaneously, will be heterogeneous. An obvious
remedy to this problem is the use of semibatch operation in which flow rates for
some of the reactants are introduced in order to nullify the drift in composition,
particle size and molecular weight. When the emulsion polymerisation operates,
semibatch process has six manipulating variables while batch has only one which is
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 14
the temperature. Thus, semibatch has more degree of freedom than that of batch.
Moreover, terpolymer composition control can not be attained in batch process
especially when the process involves monomers with highly different reactivity
ratios. Therefore, semibatch operations are used to optimise and control the end-use
properties of the terpolymer.
Based on the dynamic model, three main different optimal control objectives for
semibatch emulsion polymerisation of styrene, MMA and MA were formulated in
this work and validated experimentally by implementation within a distributed
control system (DCS) environment. These optimal control objectives were set up to
determine the control policy for the control variables (manipulating variables) that
would produce our desired PSD, MWD and polymer composition in a specified
semibatch operation time and monomers amount.
5.3.1 Particle Size Optimisation
The particle size in an emulsion has a significant effect on the conditions of the
polymerisation reaction and the properties of the final product. The particle size of
the latex influences the emulsion viscosity. The viscosity of an emulsion will
increase as the particle size is reduced. Mixtures of particles on the other hand tend
to yield lower viscosities (Fitch et al., 1987, Fitch, 1980). Monodisperse (narrow)
size distribution is required for a glossy finish to latex paints while a polydisperse
(broad) particle size distribution results in a mat finish (Bartsch et al., 1999). The
particle size and size distribution affect the dispersion rheology which is critical in
achieving a smooth surface in coating applications (Anderson and Rapra Technology
Limited (GB), 2003). Adhesives require a broad distribution to maximise their
strength since the different size particles pack closer together, minimising the void
area and thus improving adhesive performance (Fitch, 1980). Also high solids
adhesives applications require bimodal or multimodal particle size distributions to
maintain a manageable viscosity (Anderson and Rapra Technology Limited (GB),
2003). Therefore, two objectives are considered in this study to optimise the
productivity of the reactor with a defined PSD. One objective is to produce
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 15
monodisperse PSD and the other is to produce a very broad distribution of PSD
(~bimodal).
The framework of population balances is ideally suited to the description of the
complex dynamics of particulate polymerisation processes due to the highly coupled
kinetics, thermodynamics, heat and mass transfer phenomena, taking place in a
heterogeneous process. The numerical solution of the dynamic PBE for a particulate
system, especially for a reactive one, is a notably difficult problem due to both
numerical complexities and model uncertainties regarding the particle nucleation,
growth, aggregation and breakage mechanisms that are often poorly understood.
Usually, the numerical solution of the PBE requires the discretisation of the particle
volume domain into a number of discrete elements that results in a system of stiff,
nonlinear differential or algebraic/differential equations that is solved numerically.
For this reason, the PBE are discretised according to the radius of the particle using
backward finite difference method which has been described in section 3.2.
For achieving the two control objectives of PSD, the particle size polydispersity
index (PSPI), which indicates the spread of the size distribution, is selected as the
objective function. For obtaining monodisperse, PSPI objective function is
minimised while it is maximised to obtain on bimodal PSD. Also this formulation
can control the exact breadth of PSD by setting the PSPI objective function an
appropriate value with specific average radius as a constraint. The particle size
polydispersity index (PSPI) is estimated according to Eq. (5.12). The number
average radius and the mean squared radius are determined by the Eqns. (5.13 &
5.14).
2
2
><
><=
r
rPSPI (5.12)
∑
∑
=
=
⋅
=G
i
G
i
uns
in
inir
r
1
1
)(
)()( (5.13)
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 16
∑
∑
=
=
⋅
=G
i
G
i
uns
in
inir
r
1
1
2
2
)(
)()( (5.14)
The objective functions to be optimised were defined as:
( )[ ] [ ]ft0, t ; r ; tr,PSPIJmax ∈∀ℜ∈∀= (5.15)
( )[ ] [ ]ft0, t ; r ; tr,PSPIJmin ∈∀ℜ∈∀= (5.16)
The final PSD shape was included in this optimisation in the form of end point
inequality constraints formulated in terms of the final molar concentration density of
particles.
maxmin ),( ntrnn final ≤≤ (5.17)
where nmin and nmax denote the lower and upper limits respectively, and were
specified to match the required distribution.
The system decision variables (manipulating variables) were bounded and explained
in details in section 5.6. In PSPI optimisation, five equal control intervals are
implemented and the time horizon and control horizon interval are subject to the
following bounds:
sts f 250007500 ≤≤ (5.18)
stk 50001500 ≤≤ 5,..,2,1=k (5.19)
where k represent the number of the control horizon interval.
The average particle diameter and the monomer conversion are subjected to the
following end-point constraints:
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 17
Conversion: 175.0 ≤≤ w (5.20)
Average particle diameter: nmR avg 10075 ≤≤ (5.21)
5.3.2 Molecular Weight Distribution Optimisation
The end-use properties of the produced polymers are directly linked with the
molecular characteristics of the polymer chains. For instance, the proper mixing of
small-size and large-size polymer chains (bimodal MWD) may lead to simultaneous
improvement of both the flow characteristics of the polymer latex and the
mechanical properties of the final polymer film (Chia-Fen et al., 2005). Therefore,
there is a strong incentive to develop strategies to control the complete polymer chain
microstructure (MWD), and not only the molecular weight averages (Mn) (Sayer et
al., 2001a). In this study, we set two different optimised objectives for MWD, one to
produce monodisperse MWD and the other broad MWD. The Molecular weight
polydispersity index (MWPI), which gives an indication of the width of the
distribution is selected as the objective function to control the MWD. For obtaining
monodisperse, MWPI objective function is minimised while it is maximised to
obtain broad MWD. MWPI is defined in Eq. (3.12) which does not indicate that the
distribution is at high or low molecular weight. It mainly indicates the breadth of the
distribution. For this reason, Mn is used as a constraint in the optimisation process in
order to obtain the MWD at specific molecular range. The objective functions to be
optimised were defined as:
( )[ ] [ ]ft0, t ; M ;tM,MWPIJmax ∈∀ℜ∈∀= (5.22)
( )[ ] [ ]ft0, t ; M ;tM,MWPIJmin ∈∀ℜ∈∀= (5.23)
Five equal control intervals are implemented and the time horizon and control
horizon interval are subject to the following bounds:
sts f 250004500 ≤≤ (5.24)
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 18
st 50001500 1 ≤≤ (5.25)
stk 5000750 ≤≤ 5,4,3,2=k (5.26)
Subjected to process conversion end-point constraint: 18.0 ≤≤ w (5.27)
5.3.3 Composition Optimisation
One of the main objectives in the control of an emulsion polymerisation process is to
maintain the polymer composition at a predetermined level because this variable has
an important role in determining end-use properties of the product, such as glass
transition temperature, particle morphology, and mechanical and chemical resistance.
Intelligent control of the polymer composition is very important when several
monomers with different reactivity ratios are involved in the reaction at a
nonazeotropic monomer composition, which is often the case for industrial systems.
If no control action is taken significants polymer composition drifts occur in batch
process, and this leads to heterogeneity in the polymer properties. Therefore, it is
important to operate the polymerisation reactor in a way such that this desired
polymer composition is obtained. The overall polymer composition Fcomp,i is
determined by integrating the instantaneous polymer composition produced per unit
time. In a terpolymerisation process, constant composition is obtained if the
following two ratios are maintained at the desired values (m & n) throughout the
reaction:
mF
F
comp
comp=
3,
1, (5.28)
nF
F
comp
comp=
3,
2, (5.29)
From the dynamic models, it is noted that the overall polymer composition is a
function of the ratio of the monomers and their reactivity ratios. The reactivity ratios
are mainly affected by the operating temperature. Thus it is sufficient to manipulate
the ratios of the monomers in the reactor in order to keep the composition at a
desired value. Thus, the values of ‘m’ and ‘n’ must be controlled simultaneously in
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 19
order to guarantee the production of a polymer with the desired composition. This
can be accomplished by manipulating the flow rates of the monomers with ‘m’ and
‘n’ as constraints in a dynamic optimisation scheme to maximise the conversion.
High monomer conversion with desired polymer composition is expected from such
an optimisation. The objective function to be optimised was defined as:
( )[ ] [ ]ft0, t twJmax ∈∀= (5.30)
And the composition of the terpolymer is controlled via the constrained values of ‘m’
and ‘n’ in the following equations.
[ ]maxmin m,m m ∈ (5.31)
[ ]maxmin n,n n ∈ (5.32)
In the case of optimising the terpolymer composition to produce constant
composition at ratios of 5/3/2 for Sty/MMA/MA, the m and n were bounded as
follows:
[ ]2.32,2.83 m ∈ (5.33)
[ ]1.38,1.63 n ∈ (5.34)
The system decision variables (manipulating variables) were bounded and explained
in details in section 5.6. In Composition optimisation, twenty equal control intervals
(not five as for PSPI and MWPI) are implemented because composition control
needs more control action especially when their rates of polymerisation are widely
different, i.e. have different reactivity ratios. The time horizon and control horizon
interval are subject to the following bounds:
sts f 400008000 ≤≤ (5.35)
stk 2000200 ≤≤ 5,..,2,1=k (5.36)
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 20
5.4 OPTIMISATION PROCEDURE
The objective functions were first defined as mentioned before. For operational and
technical reasons such as the pumps potential limit, the feed rates of monomers,
surfactant and initiator, and temperature were specified within the following upper
and lower bounds:
[ ]0.2g/s 0.0g/s, F lm, ∈ (5.37)
[ ]0.05g/s 0.0g/s, Fs ∈ (5.38)
[ ]0.05g/s 0.0g/s, FI ∈ (5.39)
[ ]K 355 K, 338 T ∈ (5.40)
dt
tdT )( [ ]4,4-∈ (5.41)
The optimisations are for a semibatch operations and so an additional end equality
constraints control the total moles of monomers in the recipe (Nm,T) was added.
[ ]mol mol,4 1.5 N t ∈ (5.42)
An initial time length of 1500 seconds is specified for the first stage of the intervals
which is for seeding stage. The overall horizon is bounded between 6000.0 and
32000 seconds. Note that the feed rates starts after 1500s of the starting point of the
operation even though the feed rates in the feed profiles starts from the starting point
of the reaction. This is because the dynamic model included a 1500s time delay for
all the feed rates. This first interval of the operation is for producing polymers
particles and for this it is called seeding interval.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 21
The process equations are solved with a differential algebraic equation solver
(DAES) under the initial guessed conditions. This produces the value of the objective
function, which the optimisation routine then iteratively uses to find the optimal
parameters in the control parameterisation. The sequential method is of the feasible
path type, in which all process equations are feasible during the calculation of the
objective value for the iteration. The piecewise-constant control is set for the flow
rates and temperature in which they remain constant for the first period of time, and
become another value in the second interval of time. This will continue for all the
intervals.
The CVP_SS, a standard mathematical solver for optimisation in gPROMS, is
implemented to solve the model based optimisation problem in this work. CVP SS
can solve optimisation problems with both discrete and continuous decision
variables. For dynamic optimisation problems, CVP_SS is based on a control vector
parameterisation (CVP) approach which assumes that the time-varying control
variables are piecewise constant (or piecewise linear) functions of time over a
specified number of control intervals. The precise values of the controls over each
interval, as well as the duration of the latter, are generally determined by the
optimisation algorithm. The solver implements standard DASOLV and SRQPD
codes. The DASOLV code is to find the solution of the underlying differential
equation (DAE) problem and the computation of its sensitivities. While SRQPD
solver employs a sequential quadratic programming (SQP) method for the solution of
the nonlinear programming (NLP) problem. Both solvers have their own algorithmic
parameters described in the manual of gPROMS (PSE, 2004).
5.5 EXPERIMENTAL VALIDATION OF OPTIMISATION
PROFILES
Semibatch emulsion terpolymerisation reactions of styrene, MMA and MA were
carried out according to the optimal profiles obtained from the optimisation. The
experiential set up is discussed in details in chapter three. The optimal control
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 22
strategy involved the computation of the monomers feed including styrene and
MMA, surfactant feed, initiator feed and temperature trajectory which would drive
the process to give a terpolymer with a specified properties. The obtained optimal
duration to produce the desired terpolymer was varied between 1.5 and 3 hours
depending on the objectives functions and the constraints.
Three cases were carried out in this work as mentioned above. A standard PID
controller was used for internal and external temperature control, so the temperature
of the reactor was controlled within ±1.5oC to the setting point. A circulator provided
a high (constant) water flow rate through the external reactor jacket, with the water
inside the circulator. The circulator is supplied with a coil for heating and a
refrigeration system for cooling. All the manipulated variables profiles are put into
the Control Builder to be operated automatically.
The experimental results (sections 5.5.1-5.5.3) have good agreement with model
predictions. There is slight difference (≤±5%) between the experimental results and
the model simulations in which the experimental results are still accepted for model
validation. This difference is due to some normal experimental error and reactor
temperature control. The temperature of the model simulation is maintained exactly
the same as its optimal trajectory through the whole reaction, while the temperature
of the reactor that reflecting the experimental results under a PID controller, means
the temperature would oscillate over and under the set point temperature. This
explains that the slight difference is mainly with the PSD and MWD results.
5.5.1 Optimising Particle Size
Particle size is mainly affected by the amount of surfactant, monomers and initiator
involved in the operation. Surfactants are the sources of micelles which change to
new particles through micellar nucleation. Initiators are the sources of oligomers
which generate new particles by homogeneous nucleation or micellar nucleation
when the reactor contains micelles. Temperature has an indirect effect on the
particle size. The strength of its effect depend on the amount of un-reacted
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 23
monomers presenting in the reaction since temperature increases the rate of
polymerisation which leads to an increase in the growth rate of the particles, hence
narrowing the particle size distribution. Two optimisation operations were carried out
in this section. The first optimisation was carried out to produce bimodal PSD by
maximising the PSPI function (figure 5.1). The second optimisation was for
minimizing the breadth of PSD by minimizing the PSPI function (figure 5.2).
In figure 5.1, the optimisation was aimed to produce bimodal particle size
distribution by maximising the PSPI function. The optimal profiles in figure 5.1
reveal that the monomer feed of Styrene, MMA and MA are at a fixed ratio, since a
constraint was put for their flow rates to be at 5/3/2 ratio respectively. The surfactant
feed started at 0.0014 g/s and decreased to reach about 0.000086g/s at 5513s. This
low feed rate is for stabilising the system since the adsorption of surfactant molecules
increases as the surface area of polymer particle increases. After that the surfactant
flow rate increases significantly to 0.005g/s in order to generate micelles which cause
secondary nucleation to occur through micellar nucleation and hence a new small
stream of polymer particles is generated. This is accompanied with a sharp decrease
in the average size of particles as shown in figure 5.1d.
The amount of oligomeric radicals in the reactor was enough to generate the
secondary nucleation by undertaking entry to the micelles and for this reasons the
flow rate of initiator was zero for the whole reaction. The secondary nucleation
occurred after 177s of increasing the surfactant feed rate which is the time of
generating micelles. The concentration of surfactants was lower than critical micelle
concentration (cmc) and so the surfactant flow rate for a period of 177s was needed
to make the surfactant concentration equal to cmc and then the feed rate generated
micelles which led to micellar nucleation. Thus, bimodal particle size distribution is
obtained as seen in figure 5.1c.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 24
0.000
0.002
0.004
0.006
0 2000 4000 6000 8000 10000Time (s)
Fs
(g/s
) &
Fi
(g/s
)
Fs Fi
0.00
0.01
0.02
0.03
0 2000 4000 6000 8000 10000Time (s)
Flo
wra
te (
g/s
)
335
340
345
350
T (
K)
FC FA FB T
(a) (b)
0.0
0.2
0.4
0.6
0.8
1.0
0 20 40 60 80 100 120 140 160Diameter (nm))
Vo
lum
e F
rac
tio
n
0.0
20.0
40.0
60.0
80.0
100.0
0 2000 4000 6000 8000 10000Time (s)
Dia
me
ter
(nm
)
(c) (d)
0.0
0.2
0.4
0.6
0.8
1.0
0 2000 4000 6000 8000 10000Time (s)
Co
nv
ers
ion
(e)
Figure 5.1 Validation of optimisation for maximising PSPI: (a) & (b) Variable
optimal profiles, (c) PSD, (d) Particle Diameter, (e) Conversion (● experiment; ▬
simulation).
The polymerisation conversion, depicted in figure 5.1e was also used for the
validation of the optimisation process, since a constraint was put to make the
conversion above 80% and also to ensure that the model is able to predict most of
the key polymer properties at different conditions, especially under optimised
conditions. The first section of figure 5.1e [first 1500 s] shows batch reaction with
seeding. The overall conversion dropped at the instant of feeding the monomer. This
is because most of the extant monomers had converted to polymers, and at the
instant of monomer addition, there is more monomer in the reactor than it previously
contained. Thus the overall conversion drops at that instant. Subsequently, the
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 25
monomers start to propagate and convert to polymers. The decrease in the overall
conversion varies with the feed rate, and is in proportion to the amount added. Thus,
we record a larger drop for a higher flow rate, and a smaller drop for a lower flow
rate.
0.0E+00
2.0E-07
4.0E-07
6.0E-07
8.0E-07
1.0E-06
1.2E-06
1.4E-06
0 4000 8000 12000Time (s)
Fs (
g/s
)
0
0.001
0.002
0.003
0.004
0.005
0.006
Fi
(g/s
)
Fs Fi
0.000
0.008
0.016
0.024
0.032
0 4000 8000 12000Time (s)
Flo
wra
te (
g/s
)
346
348
350
352
T (
K)
FA FB FC T
(a) (b)
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0 20 40 60 80 100 120 140 160Diameter (nm)
Vo
lum
e F
rac
tio
n
0
20
40
60
80
100
120
0 2000 4000 6000 8000 10000 12000Time (s)
Dia
me
ter
(nm
)
(c) (d)
0
0.2
0.4
0.6
0.8
1
0 2000 4000 6000 8000 10000 12000Time (s)
Co
nv
ers
ion
(e)
Figure 5.2 Results for minimising PSPI, (a) & (b) Variable optimal profiles, (c) PSD,
(d) Particle Diameter, (e) Conversion (● experiment; ▬ simulation).
In figure 5.2, narrow particle size distribution was optimised by minimizing the
PSPI. As expected, the feed flow rates of initiator and surfactants were very low in
order to prevent secondary nucleation. The total monomer feed rate was at 0.0001g/s,
which is the lower limit of its constraint, for the first 3000s of the reaction. This
explain the sharp decrease in conversion at 3000s (figure 5.2e) and not at 1500s as
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 26
shown in figure 5.2e. The growth rate in this interval is very low which is observed
from the slow change in the average size of the particles (figure 5.2d). After that, the
flow rate increased significantly to increase the growth rate of particles and hence
homogenizing the particles size distribution as shown in figure 5.2c. This increase in
the growth rate led to increase the average size of the particles.
5.5.2 Optimising Molecular Weight
In this section, the complete polymer chain microstructure distribution (MWD) and
its molecular weight averages were optimised. The MWD is mainly affected by the
polymerisation temperature and monomer feed rate. The higher the monomer feed
rates, the higher the amount of monomers in the reactor which leads to a higher
propagation rate and thus a higher molecular weight is produced. As the temperature
of polymerisation increases, the coefficients of transfer and termination increase
which leads to a decrease in the degree of polymerisation and thus produces broad
MWD with a low average molecular weight.
The feed rate of initiator and surfactant has a slight effect on the MWD in which
they are indirect proportional to the molecular weights. The reason is that adding
surfactant would cause micellar nucleation, resulting in decreased molecular weight.
Initiator allows for more oligomeric radicals to be formed which increases their rate
of entry and thus increases the termination coefficient with low molecular weights
produced. The agreement between the experimental results and simulations were
good, as can be observed in figures 5.3, 5.4 & 5.5, on optimizing the MWD and Mn.
The conversion and average radius were validated to these processes to prove the
validity of the model to act as a soft sensor under different conditions.
Three optimisation operations were carried out in this section. The first optimisation
was carried out to maximise the breadth of MWD by maximising the MWPI function
(figure 5.3). The second optimisation was for minimising the breadth of MWD by
minimising the MWPI function (figure 5.4). The third optimisation was for
minimising Mn (figure 5.5).
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 27
In the first optimisation (figure 5.3), the monomers feed rates and reaction
temperature were the only manipulated variables used to produce broad MWD with
high Mn. In figure 5.3b, the temperature of the reaction increases 3oC at the middle of
the reaction intervals in order to increase the conversion since a conversion
constraint (>80%) is set up. But the temperature decreased at the end of the reaction
interval to increase MWPI. Figures 5.3c shows that MWPI decreases sharply during
the first 1500sec and then increases sharply. This is because the first 1500s of the
course of the reaction is a batch process to generate polymer particles (seeding part)
and so the amount of monomers decreases with reaction time and consequently the
rate of polymerisation decreases which led to a sharp decrease in MWPI (figure
5.3b). After that the monomer started to feed and thus MWPI increased sharply
depending on the monomer flow rate.
In the second optimisation (figure 5.4) where MWPI was minimised, monomers feed
rates, initiator feed rate and temperature were the manipulating variables. Figure 5.4a
shows that the initiator has a significant effect in producing polymers with low
molecular weights. This explains the reason for setting initiator feed rate equal to
zero in the first optimisation where MWPI was maximised. This is due to the fact
that as the amount of initiator increases, the rate of termination increases and so low
molecular weights are produced. Thus, the probability of chain growth decreases
resulting in low MWPI. Figure 5.4b shows that the feed rates of styrene and MMA
were zero while the MA monomer feed rate was too high (0.04-0.06g/s).
Styrene and MMA have high polymerisation rate coefficients with low transfer rate
which lead to high molecular weights with high MWPI. This is why the MWPI of all
the polymers produced from styrene and MMA is higher than 2.4. On the other hand,
MA has a high transfer rate coefficient which produces highly branched polymers
with low MWPI and can be reached to 1.01 (Banerjee and Konar, 1986). It can be
also seen from figure 5.4b that the temperature of the operation was high, since
temperature increases the transfer rate coefficients more than polymerisation rate
coefficients. Figure 5.4c shows a simultaneous sharp decrease and then increase to
the MWPI within the first 1800s of the operation. This can be investigated in the
same manner as that for figure 5.4c.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 28
0
0.01
0.02
0.03
0.04
0.05
0 1000 2000 3000 4000 5000
Time (s)
Flo
wra
te (
g/s
)
FA FB FC
340
341
342
343
344
345
0 1000 2000 3000 4000 5000
Time (s)
T (
K)
(a) (b)
0
1
2
3
4
0 1000 2000 3000 4000 5000
Time (s)
MW
PI
0.0
0.3
0.6
0.9
1.2
4 6 8log(M)
dw
/dlo
g(M
)
●
●
●
●
●
●
●
●
●
●
(c) (d)
0
0.2
0.4
0.6
0.8
1
0 1000 2000 3000 4000 5000
Time (s)
Co
nv
ers
ion
0
20
40
60
80
100
0 1000 2000 3000 4000 5000
Time (s)
Dia
me
ter
(nm
)
(e) (f)
Figure 5.3 Validation of optimisation for maximising MWPI, (a) & (b) Variable
optimal profiles, (c) MWPI, (d) MWD, (e) Conversion, (f) Particle Diameter (●
experiment; ▬ simulation).
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 29
0.E+00
5.E-05
1.E-04
0 1000 2000 3000 4000 5000Time (s)
Fs
(g
/s)
0.000
0.002
0.004
0.006
0.008
0.010
Fi
(g/s
)
Fs Fi
0.00
0.02
0.04
0.06
0.08
0 1000 2000 3000 4000 5000
Time (s)
Flo
wra
te (
g/s
)
346
348
350
352
T (
K)
FA FB FC T
(a) (b)
0.0
2.0
4.0
0 1000 2000 3000 4000Time (s)
MW
PI (n
m)
0.0
0.3
0.6
0.9
1.2
3.00 4.00 5.00 6.00 7.00
log(M)
dw
/dlo
g(M
)
●
●●
●
●
●
●
●
●
●
●
●
(c) (d)
0.0
0.2
0.4
0.6
0.8
1.0
0 1000 2000 3000 4000Time (s)
Co
nvers
ion
0.0
20.0
40.0
60.0
80.0
0 1000 2000 3000 4000Time (s)
Dia
mete
r (n
m)
(e) (f)
Figure 5.4 Optimisation results for minimizing MWPI, (a) & (b) Variable optimal
profiles, (c) MWPI, (d) MWD, (e) Conversion (f) Particle Diameter (● experiment;
▬ simulation).
In the third optimisation (figure 5.5) where only Mn was minimised with a
conversion constraint (>80%), the optimal profiles, as shown in figures 5.5a and
5.5b, show approximately the same feed rates as that in the second optimisation. The
slight difference is because MWPI was not set as an objective. So the initiator and
MA feed rates were slightly lower than those in the second optimisation, while the
feed rates of styrene and MMA were not zero but very low. Therefore, it can be
concluded that monomer feed rates, initiator feed rate and reaction temperature are
the main manipulation variables for controlling the MWD of the terpolymer. Figure
5.5c shows two small peaks at 1500s and 2500s in the MWPI. These two peaks are
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 30
due to the stepwise change in the reactor temperature and monomer flow rate. This
also led to the sharp change in the conversion as shown in figure 5.5e.
The MWPI in Figure 5.4 was slightly greater than the MWPI in figure 5.5. It might
be expected that the MWPI in figure 5.4 should be less than that in figure 5.5,
because the objective for figure 5.4 was to minimise MWPI while that for figure 5.5
was to minimise Mn. This slight difference is due to the type of constraints in each
optimisation process. In figure 5.4, the conversion constraint was greater than 85%
while for figure 5.5, it was set greater than 80%. Thus, the optimisation processes of
both figures followed different paths to reach their objectives.
0
0.02
0.04
0.06
0.08
0.1
0 1000 2000 3000 4000
Time (s)
Flo
wra
te (
g/s
)
335
340
345
350
355
T (
K)
Fi FA FB FC T
0
100000
200000
300000
400000
500000
0 1000 2000 3000 4000
Time (s)
Mn
(a) (b)
0.0
1.0
2.0
3.0
4.0
0 1000 2000 3000 4000
Time (s)
MW
PI
0.0
0.3
0.6
0.9
1.2
4 5 6 7 8
log(M)
Vo
lum
e F
rac
tio
n
●
●
●
●
●●●
●
●
●●●
(c) (d)
0
0.2
0.4
0.6
0.8
1
0 1000 2000 3000 4000
Time (s)
Co
nv
ers
ion
0
10
20
30
40
50
60
70
80
0 1000 2000 3000 4000Time (s)
Dia
me
ter
(nm
)
(e) (f)
Figure 5.5 Results for minimizing Mn, (a) optimal profiles, (b) Mn, (c) MWPI, (d)
MWD, (e) Conversion, (f) Particle Diameter (● experiment; ▬ simulation).
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 31
5.5.3 Optimising Composition
In this case, the terpolymer composition is optimised to produce constant
composition at ratios of 5/3/2 for Sty/MMA/MA respectively with a maximum
conversion (>80%). The composition of the terpolymer was determined using the
Proton Nuclear Magnetic Resonance (1H-NMR) spectroscopy as mentioned in
chapter 4.
Figure 5.6 presents the profiles of the monomer flow rates and the operation
temperature trajectories during the experiment. It also shows the evolution of
cumulative terpolymer composition, conversion and average radius. The terpolymer
composition was controlled by keeping the different monomers concentration ratios
at the required values. In this system, Styrene is the most reactive monomer and MA
is the lowest one. Therefore, the flow feed rate of styrene was the highest flow rate
while MA feed rate was the lowest one as shown in figure 5.6a.
The first 1500s interval of the operation was a batch process for seeding and for this
the terpolymer composition was not controlled in this interval. But the initial amount
of the monomers added to the operation was controlled to get the same terpolymer
composition at the end of the batch interval of the operation. Figure 5.6c shows a
simultaneous sharp decrease and then increase to the conversion within the first
1800s of the operation. This can be investigated in the same manner as that for figure
5.1e.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 32
0
0.01
0.02
0.03
0.04
0.05
0 2000 4000 6000 8000Time(s)
Flo
wra
te (
g/s
)340
342
344
346
348
350
352
T(K
)
FA FB FC T
0
0.2
0.4
0.6
0.8
0 2000 4000 6000 8000
Time(s)
Mo
lar
Co
mp
os
itio
n
Sty
MA
MMA
(a) (b)
0
0.2
0.4
0.6
0.8
1
0 2000 4000 6000 8000Time (s)
Co
nv
ers
ion
0
20
40
60
80
100
120
0 2000 4000 6000 8000Time (s)
Dia
me
ter
(c) (d)
Figure 5.6 Results for optimising composition, (a) Variable optimal Profiles, (b)
Mass Composition, (c) Conversion, (d) Particle Diameter (● experiment; ▬
simulation).
5.6 CONCLUSION
In this chapter, a multiobjective dynamic optimisation method for the calculation of
the Optimal control policies for polymer composition, PSD, MWD and Mn in a
semibatch emulsion terpolymerisation is presented. The optimisation procedure is
applied to styrene, Methyl methacrylate and methyl acrylate emulsion
terpolymerisation. The PSPI, MWPI, Mn and composition ratios were used as
objective functions. Six variables were used as manipulated variables, styrene
monomer feed rate, MMA monomer feed rate, MA monomer feed rate surfactant
feed rate, initiator feed rate, and the temperature of the reaction. The mathematical
dynamic model which has been presented in chapter three is used to represent
emulsion polymerisation. The optimisation problem was formulated as a multi-
objective optimisation by reformulating the multi-objective problem as a single-
objective case by placing some objectives in the form of constraints. The gOPT tool
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 33
of gPROMS was used to perform the dynamic optimisation calculations in which the
control vector parameterisation approach was adopted. The optimisation is carried
out using the SRQPD code which implements a reduced quadratic programming
algorithm. Equality and inequality safety constraints were imposed to the system
operation.
Results show that the optimisation procedure was able to minimise the reaction time
and, simultaneously, obtain a polymer with a desired quality (composition, size
distribution or molecular weight) while taking the safety aspect into account. The
obtained control policies were validated on an experimental stirred tank
polymerisation system and an excellent agreement between them was found. On
maximising PSPI, a bimodal distribution was produced being broader. Temperature
showed to have a strong effect on the molecular weight due to the effect of the
transfer rate coefficient. On maximising the molecular weight, the initiator
concentration was zero at all times showing that any initiator addition could cause
more radicals to be produced, and therefore more transfer. Surfactant feed rate was
also zero at all times when temperature was used as a manipulated variable
concluding that adding surfactant may produce more micelles forming new particles,
and the monomers would then tend to get to the new particles rather than propagating
to produce larger molecular weights.
The results showed that the intelligent control of the polymer composition for several
monomers with different reactivity ratios needs continuous control action and for
this, its time horizon partitioned in twenty short time stages. This work is a first step
in the advanced control hierarchy to the MWD, PSD and terpolymer composition of
the terpolymer product through the manipulation of the reaction temperature and the
monomers, surfactant and initiator feed rates.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 34
BIBLIOGRAPHY
Alhamad, B., Romagnoli, J. A. & Gomes, V. G. (2005) Advanced modelling and
optimal operating strategy in emulsion copolymerization: Application to
styrene/MMA system. Chemical Engineering Science, 60, 2795.
Anderson, C. D. & Rapra Technology Limited (GB) (2003) Emulsion polymerisation
and applications of latex, Shrewsbury, UK, Rapra Technology Limited.
Arzamendi, G., Delacal, J. C. & Asua, J. M. (1992) Optimal Monomer Addition
Policies For Composition Control Of Emulsion Terpolymers. Angewandte
Makromolekulare Chemie, 194, 47-64.
Banerjee, M. & Konar, R. S. (1986) Mechanism of the emulsion polymerization of
methyl acrylate: 2. Kinetics and growth of the polymers. Polymer, 27, 147-
57.
Bartsch, S., Kulicke, W. M., Fresen, I. & Moritz, H. U. (1999) Seeded emulsion
polymerization of styrene. Determination of particle size by flow field-flow
fractionation coupled with multi-angle laser light scattering. Acta Polymerica,
50, 373-380.
Bellman, R. E. & Kalaba, R. (1965) Dynamic Programming and Modern Control
Theory.
Bhaskar, V., Gupta, S. K. & Ray, A. K. (2000) Multiobjective optimisation of an
industrial wiped-film pet reactor. Aiche Journal, 46, 1046-1058.
Bhaskar, V., Gupta, S. K. & Ray, A. K. (2001) Multiobjective optimisation of an
industrial wiped film poly(ethylene terephthalate) reactor: some further
insights. Computers & Chemical Engineering, 25, 391-407.
Biegler, L. T. (1984) Solution of dynamic optimisation problems by successive
quadratic programming and orthogonal collocation. Computers & Chemical
Engineering, 8, 243-7.
Bojkov, B. & Luus, R. (1994a) Application of Iterative Dynamic-Programming to
Time-Optimal Control. Chemical Engineering Research & Design, 72, 72-80.
Bojkov, B. & Luus, R. (1994b) Time-Optimal Control by Iterative Dynamic-
Programming. Industrial & Engineering Chemistry Research, 33, 1486-1492.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 35
Bojkov, B. & Luus, R. (1995) Time-Optimal Control of High-Dimensional Systems
by Iterative Dynamic-Programming. Canadian Journal of Chemical
Engineering, 73, 380-390.
Bulirsch, R. (1971) Die Mehrzielmethods Zur Numerischen Losung Von
Nichtilnearen Randwerproblemen Und Aufgaben Der Optimalen Stewerung',.
Deutsche Forschungs-und Versuchsanstalt fur Luft-und Raumfahrt,
Oberphaffenhofen, Federal Republic of Germany, Carl-Cranz Gesellschaft.
Canu, P., Canegallo, S., Morbidelli, M. & Storti, G. (1994) Composition control in
emulsion copolymerization. I. Optimal monomer feed policies. Journal of
Applied Polymer Science, 54, 1899.
Chakravarthy, S. S. S., Saraf, D. N. & Gupta, S. K. (1997) Use of genetic algorithms
in the optimisation of free radical polymerizations exhibiting the
Trommsdorff effect. Journal of Applied Polymer Science, 63, 529-548.
Cheremisinoff, N. P. (1989) Handbook of polymer science and technology, New
York, M. Dekker.
Chia-Fen, L., Hung-Hsin, T., Lee-Yin, W., Chia-Fu, C. & Wen-Yen, C. (2005)
Synthesis and properties of silica/polystyrene/polyaniline conductive
composite particles. Journal of Polymer Science Part A-Polymer Chemistry,
43, 342-54.
Choi, K. Y. & Butala, D. N. (1991) An experimental study of multiobjective
dynamic optimisation of a semibatch copolymerization process. Polymer
Engineering and Science, 31, 353-64.
Clarke-Pringle, T. L. & MacGregor, J. F. (1998) Optimisation of molecular-weight
distribution using batch-to-batch adjustments. Industrial & Engineering
Chemistry Research, 37, 3660-3669.
Congalidis, J. P., Richards, J. R. & Ray, W. H. (1989) Feedforward and feedback
control of a solution copolymerization reactor. AIChE Journal, 35, 891-907.
Crowley, T. J., Harrison, C. A. & Doyle III, F. J. (2001) Batch-to-batch optimisation
of PSD in emulsion polymerization using a hybrid model. 2001 American
Control Conference, Jun 25-27 2001. Arlington, VA, Institute of Electrical
and Electronics Engineers Inc.
Crowley, T. J., Meadows, E. S., Kostoulas, E. & Doyle, F. J., III (2000) Control of
particle size distribution described by a population balance model of
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 36
semibatch emulsion polymerization. Journal of Process Control, 10, 419-
432.
Dimitratos, J., Elicabe, G. & Gerogakis, C. (1994) Control of emulsion
polymerization reactors. AIChE Journal, 40, 1993-2021.
Dixon, L. C. & Bartholomew-Biggs, M. C. (1981) Adjoint-control Transformations
for Solving Practical Optimal Control Problems. Optim. Control Appl.
Methods, 2, 365-381.
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.
Fitch, R. M. (1980) Polymer colloids II, New York, Plenum Press.
Fitch, R. M., El-Aasser, M. S. & North Atlantic Treaty Organization. Scientific
Affairs Division. (1987) Future directions in polymer colloids, Dordrecht ;
Boston, Published in cooperation with NATO Scientific Affairs Division by
M. Nijhoff.
Garg, S. & Gupta, S. K. (1999) Multiobjective optimisation of a free radical bulk
polymerization reactor using genetic algorithm. Macromolecular Theory and
Simulations, 8, 46-53.
Gupta, R. R. & Gupta, S. K. (1999) Multiobjective optimisation of an industrial
nylon-6 semibatch reactor system using genetic algorithm. Journal of Applied
Polymer Science, 73, 729-739.
Immanuel, C. D. & Doyle, F. J., III (2002) Open-loop control of particle size
distribution in semibatch emulsion copolymerization using a genetic
algorithm. Chemical Engineering Science, 57, 4415.
Kiparissides, C. (2006) Challenges in particulate polymerization reactor modeling
and optimisation: A population balance perspective. Journal of Process
Control, 16, 205-224.
Kozub, D. & Macgregor, J. F. (1992) Feedback control of polymer quality in
semibatch copolymerization reactors. Chemical Engineering Science, 47,
929-942.
Miele, A. (1975) Recent Advances in Gradient Algorithms for Optimal Control
Problems. JOTA, 17, 361-430.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 37
Mitra, K., Deb, K. & Gupta, S. K. (1998) Multiobjective dynamic optimisation of an
industrial nylon 6 semibatch reactor using genetic algorithm. Journal of
Applied Polymer Science, 69, 69-87.
Mitra, K., Majumdar, S. & Raha, S. (2004) Multiobjective dynamic optimisation of a
semibatch epoxy polymerization process. Computers & Chemical
Engineering, 28, 2583-2594.
Pontryagin, L. S., Boltyanskii, V. G., Gamkrelidze, R. V. & Mishchenko, E. F.
(1962) The Mathematical Theory of Optimal Processes. Wiley-Interscience,
NY.
PSE (2004) gPROMS Advanced User Guide, Process System Enterprise Ltd ,
London, UK.
Saldivar, E. & Ray, W. H. (1997) Control of semicontinuous emulsion
copolymerization reactors. Journal of Engineering and Applied Science, 43,
2021.
Sareen, R. & Gupta, S. K. (1995) Multiobjective Optimisation of an Industrial
Semibatch Nylon-6 Reactor. Journal of Applied Polymer Science, 58, 2357-
2371.
Sayer, C., Arzamendi, G., Asua, J. M., Elima, E. L. & Pinto, J. C. (2001a) Dynamic
optimisation of semicontinuous emulsion copolymerization reactions:
composition and molecular weight distribution. Computers & Chemical
Engineering, 25, 839-849.
Sayer, C., Arzamendi, G., Asua, J. M., Lima, E. L. & Pinto, J. C. (2001b) Dynamic
optimisation of semicontinuous emulsion copolymerization reactions:
Composition and molecular weight distribution. Computers and Chemical
Engineering, 25, 839.
Schoonbrood, H. A. S., vanEijnatten, R. & German, A. L. (1996) Emulsion co- and
terpolymerization of styrene, methyl methacrylate and methyl acrylate .2.
Control of emulsion terpolymer microstructure with optimal addition profiles.
Journal of Polymer Science Part a-Polymer Chemistry, 34, 949-955.
Semino, D. & Ray, W. H. (1995a) Control of systems described by population
balance equations - I. Controllability analysis. Chemical Engineering Science,
50, 1805-24.
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 38
Semino, D. & Ray, W. H. (1995b) Control of systems described by population
balance equations - II. Emulsion polymerization with constrained control
action. Chemical Engineering Science, 50, 1825-39.
Srinivasan, B., Palanki, S. & Bonvin, D. (2003) Dynamic optimisation of batch
processes - I. Characterization of the nominal solution. Computers &
Chemical Engineering, 27, 1-26.
Srour, M., Gomes, V. G. & Romagnoli, J. A. (2005) An Advanced Model for
Optimal Operation of Emulsion Terpolymerisation Processes: Application to
Styrene/MMA/MA. AIChE Annual Meeting. Cincinnati, US.
Srour, M. H., Gomes, V. G. & Romagnolib, J. A. (2007) Online inferential product
attribute estimation for optimal operation of emulsion terpolymerisation:
Application to styrene/MMA/MA. Chemical Engineering Science, 62, 4420-
4438.
Vassiliadis, V. S., Pantelides, C. C. & Sargent, R. W. H. (1993) Optimisation of
discrete charge batch reactors. Computers & Chemical Engineering, 18,
S415-S419.
Vassiliadis, V. S., Sargent, R. W. H. & Pantelides, C. C. (1994a) Solution of a Class
of Multistage Dynamic Optimisation Problems. 1. Problems without Path
Constraints. Industrial & Engineering Chemistry Research, 33, 2111-22.
Vassiliadis, V. S., Sargent, R. W. H. & Pantelides, C. C. (1994b) Solution of a Class
of Multistage Dynamic Optimisation Problems. 2. Problems with Path
Constraints. Industrial & Engineering Chemistry Research, 33, 2123-33.
Vicente, M., Leiza, J. R. & Asua, J. M. (2001) Simultaneous control of copolymer
composition and MWD in emulsion copolymerization. AIChE Journal, 47,
1594.
Vicente, M., Sayer, C., Leiza, J. R., Arzamendi, G., Lima, E. L., Pinto, J. C. & Asua,
J. M. (2002) Dynamic optimisation of non-linear emulsion copolymerization
systems open-loop control of composition and molecular weight distribution.
Chemical Engineering Journal, 85, 339.
Yee, A. K. Y., Ray, A. K. & Rangaiah, G. P. (2003) Multiobjective optimisation of
an industrial styrene reactor. Computers & Chemical Engineering, 27, 111-
130.
Zeaiter, J., Romagnoli, J. A., Barton, G. W., Gomes, V. G., Hawkett, B. S. & Gilbert,
R. G. (2002) Operation of semibatch emulsion polymerisation reactors:
Chapter 5 Optimal Operating Strategies for Emulsion Terpolymerisation Products
5. 39
Modelling, validation and effect of operating conditions. Chemical
Engineering Science, 57, 2955-2969.