Available online at www.sciencedirect.

com

Computers and Chemical Engineering 32 (2008) 2155–2167

Free-radical polymerization at higher temperatures:

Systems impacts of secondary reactions
Masoud Soroush a,∗ , Michael C. Grady b , George A. Kalfas b
a Department of Chemical and Biological Engineering, Drexel University, Philadelphia, PA 19104, United States
b Marshall Laboratory, DuPont, Philadelphia, PA 19146, United States

Received 4 June 2007; accepted 26 October 2007

Available online 24 November 2007

Abstract
This paper presents an evaluation of systems challenges in free-radical solution homo-polymerization as temperature increases. ‘Classical’ chain-
polymerization kinetic models (accounting for initiation, propagation, chain transfer to solvent and monomer, and termination reactions) are unable
to describe the dynamics of these processes. This inability is a consequence of the significant contributions of secondary (other) reactions, such as
depropagation, self-initiation, ␤-scission, and inter/intra-molecular chain-transfer reactions, to the process dynamics. These secondary reactions
have less significant influence on the polymerization at low temperatures. An application of high-temperature polymerization is in the production
of lower average molecular weight, higher solid content resins. The quality of these resins is not characterized completely by measurements of solid
content and average molecular weights alone. Advanced chromatographic and spectroscopic measurements of the concentrations of functional
groups, terminal double bonds, chain branches and solvent groups, provide a fuller characterization of the polymer quality. Mathematical models
describing these processes are of higher order and are more complex than their low-temperature counterparts. These processes are of lower degree of
controllability, as a higher number of properties should be controlled (to achieve quality control) using the same number of manipulated inputs. For
these same reasons, these processes have less degree of observability than their low-temperature counterparts, unless more advanced measurements
are available.
© 2007 Elsevier Ltd. All rights reserved.

Keywords: High-temperature free-radical polymerization; Reaction kinetics; Microstructure properties; Complex nonlinear dynamics; Chromatographic properties;
Spectroscopic properties

1. Introduction Air Act Amendments, 1990) on volatile organic content (VOC)

of coatings require the reduction of allowable VOC’s from
The macromolecular structure and the material properties 480 g/L of paint in 1990 to below 300 g/L by 2010. These and
of polymers are mainly determined at the synthesis stage. similar regulations (Schultz, 1992; VOC’s Directive, 1996) are
Therefore, there is a great incentive for efficient operations of driving change in the basic nature of resin in coatings. Low
polymerization reactors. The crucial role, that good control can molecular weight, highly functionalized polymer and oligomer
play in achieving an efficient operation, is well-recognized in the solutions at 60–80 weight percent solids (Adamsons, Blackman,
polymer industry. Control of these processes has been challeng- Gregorovich, Lin, & Matheson, 1998) have replaced high molec-
ing for several reasons, among them, their complex nonlinear ular weight, non-functional1 polymer solutions at 30–40 weight
behavior. For example, steady-state multiplicity, parametric sen- percent solids as key components in coatings formulas. Molec-
sitivity and limit cycles are usually observed in continuous-tank ular weight control is achieved in a variety of methods, and
polymerization reactors (Schmidt, Clinch, & Ray, 1981). unfortunately each method can detract from the end use prop-
North American regulations (Reish, 1993; Superintendent of erties of the coating. One method is to increase the amount of
Documents Title 1, 1990; Superintendent of Documents Clean

1 A polymer solution is non-functional, if the dead polymer chains are unable

∗ Corresponding author. Tel.: +1 215 895 1710; fax: +1 215 895 5837. to participate in further polymerization reactions such as cross-linking; high
E-mail address: masoud.soroush@drexel.edu (M. Soroush). molecular weight lacquer resins are such polymers.

0098-1354/$ – see front matter © 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2007.10.023
2156 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167

Nomenclature
μi ith moment of the chain length distribution of dead
[A] concentration of species A (mol/L). A = polymer chains
S,M,I,TDB,SCB,D,P,Q,PTDB ,R ρ̄ constant average density of reactor content
[B̂] estimated value of the variable B. B =
I,S,TDB,SCB,D
D dead polymer chain
free-radical initiator relative to monomer on formula (Prane,
Dn dead polymer chain with n monomer units
1986). This reduces the molecular weight of the resin, but adds
fj nonlinear function describing the time derivative
cost since initiators are often the most expensive component of a
of the state variable j = V ,I,M,S,TDB,SCB,D
formula. It also increases the safety risk of handling concentrated
Fi volumetric flow rate of species i (L/s), i = S,M,I
initiator solutions (Balchan, Klein, & Klein, 1995) and makes
Hi nonlinear function describing the dependence of
dissolution of solid initiators more difficult. Another method is
the slow measurement i on the state variables. i =
to use chain transfer agents like mercaptans, reversible addition
Mn , Mw
fragmentation agents (Chiefari et al., 1998), cobalt porphyrin
I initiator
complexes (Wayland et al., 1998), nitroxide mediated polymer-
ki reaction rate constant of reaction i
ization catalysts (Georges, Veregin, Kazmaier, & Hamer, 1993)
kt kt = ktc + ktd
or atom transfer radical catalysts (Wang & Matyjaszewski, 1995)
Lij ij th entry of the estimator gain (matrix)
but these invariably leave a moiety in the solution or attached to
M monomer
the polymer chain that can cause difficulty later in the coat-
Mn number-average molecular weight
ing application. For example, mercaptans have an offensive
Mw weight-average molecular weight
smell and the sulfur is known to reduce long term durability
Mwi molecular weight of species i, i = S,M,I
(Mesrobian & Tobolsky, 1947). Cobalt glyoximes impart a dark,
P live polymer chain of any type excluding PTDB1 reddish color to the resin limiting their usefulness in clear-coat
Pn live polymer chain with a secondary radical and
formulations, and the cobalt species can also catalyze post-
n monomer units
reactions of the resin solution. Nitroxide and other controlling
P0 live polymer chain with a secondary radical and
agents can slow polymerization rates and limit complete conver-
no monomer units
sion of monomer to polymer. A third method to lower molecular
P1 live polymer chain with a secondary radical and
weight is to polymerize at temperatures greater than ca. 120 ◦
one monomer units
C (Hakim, Verhoeven, McManus, Dube, & Penlidis, 2000) to
P2 live polymer chain with a secondary radical and
yield low molecular resin at reasonable initiator levels without
two monomer units
the use of chain transfer agents. This method also has drawbacks,
PTDB
1 live polymer chain with a double bond and of one
including higher rates of competing secondary reactions that can
monomer unit length
detract from final coating properties, if the reactions are not well
Qn live polymer chain with a tertiary radical and n
understood. Some secondary reactions impact the polymeriza-
monomer units
tion rate and polymer molecular weight and structure. Some,
R all types of live polymer chains
such as the rearrangement of hydroxyl functionality from sec-
rk rate of production of k, k = I,M,S,TDB,SCB,D
ondary to primary, change the reactivity of the functionalized
S solvent
polymer, affecting film curing rates and coating properties. Sec-
SH number of terminal solvent groups per one hun-
ondary reactions, such as spontaneous initiation, scission, and
dred repeat units
double bond formation reactions, contribute positively to the
SCB short chain branch
formation of the desired product. Because of these features,
T reactor temperature, (◦ C)
high-temperature polymerization is often the most attractive
TDB terminal double bond
alternative: it is easy to implement and adds no additional moi-
V reactor volume (L)
eties to the polymer chain or into the polymer solution. To make
VM cumulative volumetric amount of monomer added
best of the secondary reactions in the production of the resins,
at a given time (L)
a good quantitative understanding of the reactions is needed to
xM monomer conversion
steer the reactions in optimal directions during the course of the
δi ith moment of the chain length distribution of live
polymerization.
polymer chains with a tertiary radical
Recent studies of high-temperature polymerization kinetics
 impurities
have focused on (i) the identification of dominant poly-
λi ith moment of the chain length distribution of live
merization reactions through spectroscopic methods (i.e.,
polymer chains with a secondary radical
polymer characterization studies) (Ahmad, Heatley, & Lovell,
λ̃i ith moment of the chain length distribution of live
1998; Chiefari et al., 1999; McCord, Shaw, & Hutchinson,
polymer chains with three or more monomer units
1997; Peck, Hutchinson, Grady, 2002), (ii) calculation
and a secondary radical
of reaction rate orders (Fernandez-Garcia, Fernandez-Sanz,
& Madruga, 2000), (iii) individual estimation of reaction
 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167 2157

rate constants through pulsed-laser initiated polymeriza- Table 1

tion (PLP) experiments (Asua et al., 2004; Beuermann, Classical reaction network for solution free-radical homo-polymerization at low
temperatures
& Buback, 2002; Buback, Egorov, & Feldermann, 2004;
kd
Nikitin, Castignolles, Charleux, & Vairon, 2003;), and (iv) I2 →2P0 Initiation
simultaneous estimation of a set of reaction rate con- P0 + M→P1
kii

stants based on advanced (e.g., spectroscopic measurement

kp
of the chain branch concentration) and conventional (e.g., PTDB
1 + M→P2 + (TDB) Propagation
kp
gravimetric measurement of polymeric solids content and chro- Pn + M→Pn+1 Propagation
matographic measurement of molecular weight distribution) ktm
Pn + M→Dn + PTDB Chain transfer to monomer
1
measurements (Busch & Muller, 2004; Peck & Hutchinson, kts
Pn + S→Dn + P0 Chain transfer to solvent
2004).
ktd
This paper presents an evaluation of systems challenges Pn + Pm →Dn + Dm + (TDB) Termination
ktc
in free-radical solution homo-polymerization as temperature Pn + Pm →Dn+m Termination
increases. ‘Classical’ chain-polymerization kinetic models
(accounting for initiation, propagation, chain transfer to
solvent and monomer, and termination reactions) are unable
to describe the dynamics of these processes. This inability is 2.1. Reaction kinetics
a consequence of the significant contributions of secondary
(other) reactions, such as depropagation, self-initiation, and A typical low-temperature (classical) free-radical solution
inter/intra-molecular chain-transfer reactions, to the process homo-polymerization kinetic mechanism is given in Table 1.
dynamics. These secondary reactions have less significant The mechanism in Table 2 is a simplified version of that in
influence on the polymerization at low temperatures. An Table 1. It accounts for polymer chain type, but not for the chain
application of high-temperature polymerization is in the length. It leads to a simpler model that can describe the total
production of lower average molecular weight, higher solid molar concentration of each polymer type but not the molar con-
content resins. The quality of these resins is not characterized centration of chains of the same type and length. Therefore, the
completely by measurements of solid content and average simpler model can provide very little information on polymer
molecular weights alone. Advanced chromatographic and chain length distributions.
spectroscopic measurements of the concentrations of functional
groups, terminal double bonds, chain branches and solvent 2.1.1. Detailed reaction rate equations
groups, provide a fuller characterization of the polymer In the derivation of the reaction rate equations, the following
quality. Mathematical models describing these processes usual assumptions were made:
are of higher order and are more complex than their low-
temperature counterparts. These processes are of lower degree
of controllability, as a higher number of properties should be • long chain hypothesis (LCH),
controlled (to achieve quality control) using the same number • elementary and irreversible reaction steps,
of manipulated inputs. For these same reasons, these processes • insignificance of the gel and glass effects, and
have less degree of observability than their low-temperature • chain length independence of the reaction rates.
counterparts, unless more advanced measurements are
available. On the basis of the reaction mechanism in Table 1, the rates of
The organization of this paper is as follows. Systems aspects production of different species, structural characters, and the first
of low-temperature free-radical solution homo-polymerization three leading moments of live and dead polymer chain length
are described in Section 2. Section 3 deals with high-
temperature free-radical solution polymerization. Section 4
presents an application of a reduced-order high-temperature Table 2
polymerization reactor model; it describes a multi-rate Simplified classical reaction network for solution homo-polymerization at low
state estimation case study of a high-temperature polymer- temperatures
ization reactor. Finally, concluding remarks are given in kd
I2 →2P Initiation
Section 5. kii
P + M→P
2. Free-radical polymerization at low temperatures PTDB + M→P + (TDB)
kp
Propagation
kp
P + M→P Propagation
Low-temperature free-radical polymerization reactors have ktm
rich dynamic behavior. For example, the polymerization in a P + M→D + PTDB Chain transfer to monomer
kts
continuous stirred tank reactor may exhibit multiple steady P + S→D + P Chain transfer to solvent
states, parametric sensitivity, and limit cycles; the models rep- ktd
2P→2D + (TDB) Termination
resenting these processes are highly nonlinear (Schmidt et al., ktc
2P→D Termination
1981).
2158 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167

distributions are then obtained: Number- and weight-average molecular weights and
monomer conversion are then obtained from:
rM = −(kp + ktm )[M]λ0
μ1 μ2 [M]
rS = −kts [S]λ0 Mn = MwM , Mw = MwM , xm = 1 − (3)
μ0 μ1 [M]0
rI = −kd [I]
rTDB = kp [M][PTDB
1 ] + ktd λ20
2.1.2. Simplified reaction rate equations
rPTDB = ktm [M]λ0 − kp [M][PTDB
1 ] The reaction rate equations can be simplified through con-
1
rμ0 = (ktm [M] + kts [S])λ0 + (0.5ktc + ktd )λ20 sidering the macromolecule kinetics as small-molecule kinetics
(Villermaux & Blavier, 1984). The underlying assumption
rμ1 = (ktm [M] + kts [S])λ1 + (ktc + ktd )λ0 λ1 (1)
behind the simplification of the reaction network is in the clas-
rμ2 = (ktm [m] + kts [S] + (ktd + ktc )λ0 )λ2 + ktc λ21 sification of the reacting species. This approximation becomes
rλ0 = 2fkd [I] − (ktc + ktd )λ20 more accurate as the polymer chain length and the concentra-
rλ1 = kp [M]λ0 + (ktm [M] + kts [S])λ0 − (ktm [M] tion of polymer decrease. Using the resulting reaction network
given in Table 2, the rates of production of different species and
+ kts [S])λ1 − (ktc + ktd )λ0 λ1
structural characters are given by:
rλ2 = kp [M]λ0 + (ktm [M] + kts [S])λ0 + 2kp λ1 [M]
− (ktm [M] + kts [S])λ2 − (ktc + ktd )λ0 λ2 rM = −(kp + ktm )[M][P]
rS = −kts [S][P]
where the moments of the chain length distribution of the live
rI = −kd [I]
secondary propagating radicals and the dead chains are defined
as: rTDB = kp [M][PTDB ] + ktd [P]2 (4)
∞ ∞
rPTDB = ktm [M][P] − kp [M][PTDB ]
 
λk = k
n Pn , μk = n k Dn rD = (ktm [M] + kts [S])[P] + (0.5ktc + ktd )[P]2
n=0 n=1 rP = 2fkd [I] − (ktc + ktd )[P]2
Making the quasi-steady-state-assumption (QSSA) for the
Making the QSSA for the live polymer chains, the last rate
live polymer chains, the last three rate equations in Eq. (1) lead
equation reduces to:
to:
   0.5
2fkd [I] 0.5 2fkd [I]
λ0 = [P] = ,
ktd + ktc
, ktd + ktc
λ0 ((kp + ktm )[M] + kts [S]) Batch process model: Assuming perfect mixing and constant
λ1 = ,
ktm [M] + kts [S] + kt λ0 reaction solution volume, species and structural character bal-
λ0 ((kp + ktm )[M] + kts [S]) + 2kp λ1 [M] ances for a batch reactor lead to:
λ2 = .
kfm [M] + kts [S] + kt λ0
d[M]
= rM , [M](0) = [M]0
Batch process model: Assuming perfect mixing and constant dt
reaction solution volume, species and structural character bal- d[S]
= rS , [S](0) = [S]0
ances for a batch reactor result in: dt
d[I]
d[M] = rI , [I](0) = [I]0
= rM , [M](0) = [M]0 dt
d[TDB] (5)
dt
d[S] = rTDB , [TDB](0) = 0
= rS , [S](0) = [S]0 dt
dt d[D]
d[I] = rD , [D](0) = 0
= rI , [I](0) = [I]0 dt
dt d[PTDB ]
d[TDB] = rPTDB , [PTDB ](0) = 0
= rTDB , [TDB](0) = 0 dt
dt
d[μ0 ] (2)
Number-average molecular weight is obtained approximately
= rμ0 , [μ0 ](0) = 0
dt from:
d[μ1 ]
= rμ1 , [μ1 ](0) = 0 Mm ([M]0 − [M])
dt Mn = (6)
d[μ2 ] D
= rμ2 , [μ2 ](0) = 0
dt As stated before, a model based on the simplified reaction
d[PTDB ] mechanism provides little information on the polymer chain
1
= rPTDB , [PTDB
1 ](0) =0
dt 1 length distribution.
 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167 2159

2.2. Product quality Some, such as the rearrangement of hydroxyl functionality from
secondary to primary, change the reactivity of the functionalized
The quality of a low-solids high-molecular-weight resin polymer, affecting film curing rates and coating properties. Sec-
(polymer solution) strongly depends on the final chain length ondary reactions, such as spontaneous initiation, scission, and
distribution of the polymer and the level of polymer content double bond formation reactions, contribute positively to the
(solids) in the resin. Controlling polymer number- and weight- formation of the desired product.
average molecular weights as well as the solids level usually
ensures the production of a resin with acceptable properties. 3.1. Reaction kinetics
Furthermore, in the past when low-solids high-molecular-weight
resins were used in auto paints, a painter could adjust the paint Studies in spontaneous polymerization of methyl methacry-
viscosity that strongly affects the quality of the paint finish by late (MMA) have cited both self-initiation (Brand, Stickler,
changing the level of solvent in the paint can. & Meyerhoff, 1980; Lingnau, Stickler, & Meyerhoff, 1980;
Lingnau, Stickler, & Meyerhoff, 1983; Lingnau & Meyerhoff,
2.3. Control problems 1984a, 1984b; Stickler & Meyerhoff, 1978; Stickler &
Meyerhoff, 1981) and decomposition of impurities, i.e. oxy-
Because the macromolecular structure and the material prop- gen or peroxides in general (Clouet, Chaumont, & Corpart,
erties of polymers are mainly determined at the synthesis stage, 1993; Lehrle & Shortland, 1988), as the underlying mecha-
there is a great incentive for optimal operations of polymeriza- nism for the generation of initiating radicals. In a recent study
tion reactors. Control of the reactors is challenging because of on thermally-initiated MMA polymerization, it is reported that
the complex nonlinear behavior of the reactors, the lack of fre- MMA could also undergo reaction with air to form macro-
quent on-line measurements that are with little time delays and molecular peroxides (Nising, Meyer, Carloff, & Wicker, 2005).
are directly related to product quality, the lack of controllability Both peroxide decomposition (McManus, Penlidis, & Dube,
adequate to produce polymer products with an exact molecular 2002) and self-initiation (Cao, Zhu, Zhang, & Yuan, 2004) have
weight distribution, and the lack of observability sufficient to been assumed to be responsible for the generation of initiat-
reconstruct molecular weight distribution from a few measure- ing radicals in nBA polymerization (Rantow et al., 2006). In
ments. Furthermore, low-solids, high-molecular weight acrylics this paper, mechanisms of thermal self-initiation proposed for
were more frequently made with batch processes in which all methyl methacrylate (Lingnau et al., 1983) and the decomposi-
of the monomer, solvent and initiator were added to the reactor tion of peroxide (Cerinski & Jelencic, 2002) are adopted. The
at the start of the batch. Heat was applied and the polymer- initiation step dictates the amount of live radicals (initially)
ization would begin once the increasing temperature drove the present, and will eventually dictate the amount of dead polymer
decomposition of the initiator. The exothermic reaction drove the chains formed. The initiating radicals formed by the peroxide
temperature higher until the reaction mass would boil or reflux. decomposition is represented by live radical chains of length
The energy of the reaction was removed by a total condenser zero (P0 ), while in the self-initiation reaction, the initiating rad-
in the form of condensed reflux. Batch to batch reproducibility icals are assumed to be the monomer itself, thus represented as
was influenced by heat-up profiles and raw material quality and live (secondary) radical chains of length one (P1 ).
purity; both difficult to control. Later on, as semi-batch processes Depropagation in methacrylate polymerization has been
were introduced to insure uniform functional group distribu- known for some time and studied extensively; propagating
tion, the control problem again focused on proper heat input and methacrylate radical chains unzip with increasing frequency at
exotherm removal through condensation. Raw material quality higher temperature. This reaction is thermodynamically con-
again remained an important unknown. trolled and the ratio of the forward to reverse rate constants is
given by the equilibrium constant (Grady et al., 2002). In acry-
3. Free-radical polymerization at higher temperatures late polymerization, intramolecular chain-transfer to polymer or
backbiting has been postulated to dominate at lower monomer
Molecular weight distribution and monomer conversion in content (ca. below 50 wt.%) (Peck & Hutchinson, 2004). It can
high-temperature (above 120 ◦ C) polymerization depart signif- also occur even at lower temperatures (Clouet et al., 1993; Lehrle
icantly from those predicted by classic free-radical kinetics. & Shortland, 1988). The backbiting event allows for the creation
Significant difference between measurements of the molecular of short chain-branches, upon propagation of the tertiary (mid-
weight from industrial polymerizations and theoretical predic- chain) radical near the end of the chain. This tertiary radical can
tions, made by using a classic kinetic model that includes alternatively undergo ␤-scission, to create a dead polymer chain
initiation, propagation and termination, reported in Grady, with a terminal double bond (TDB) and a live polymer chain.
Simonsick, and Hutchinson (2002) clearly shows that classi- As only averages are considered in this study, only the scission
cal kinetic models are unable to represent high-temperature reactions which create long dead chains instead of dead trimer
polymerizations. Proper design and safe industrial operation of chains with TDBs are included in the overall kinetic scheme. The
high-temperature free-radical polymerization processes require propagation, chain transfer to polymer, branching and scission
a fundamental quantitative understanding of the kinetics of the reactions are presented in unison, since these reactions simul-
secondary reactions. Some secondary reactions impact the poly- taneously affect the average molecular weights, the number of
merization rate and polymer molecular weight and structure. TDBs and the number of branching points. Dead polymer chains
2160 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167

are created through the scission reaction, and are governed by postulated polymerization reaction network is presented in
the amount of tertiary radicals (Qn ) present. The scission reac- Table 3.
tion leads to the creation of a dead polymer chain (Dn ) with a On the basis of the detailed reaction mechanism given in
TDB in one step. At the same time the tertiary radicals can also Table 3, production rate equations for dimers, impurity radical,
form chain branches through further propagation. SCB denotes xylol radical, monomer, solvent and micro-structural properties
a short chain-branch point formed through such a reaction. such as short chain-branching points (SCBs) and terminal double
Chain transfer to monomer allows for the creation of a dead bonds (TDBs) are derived. The rate equations are:
chain with a TDB and a live chain. The rate of chain transfer
to monomer is influenced by the amount of monomer and live rM = −2kti [M]2 − kp [M]([P0 ] + [P1 ] + [P2 ] + [PTDB
1 ] + λ̃0 )
chains present at a given time. The concentration of monomer − kpq [M]δ0 − ktm [M]λ0 + kdp ([P2 ] + λ0 )
is much larger compared to the concentration of live chains. rS = −kts [S]λ0
The significance of chain transfer to monomer, however, is
rI = −kd [I]
often discounted due to the low reaction rate constant relative
to the scission reaction. A recent estimation of the rate con- r = −kd []
stant for chain transfer to monomer reactions in high-pressure rSCB = kpq [M]δ0
high-temperature polymerization of nBA indicated that trans- rTDB = kβ δ0 + kp [M][PTDB ] + ktd
sec−sec
(λ0 )2
1
fer to monomer might have a more significant contribution
towards the overall kinetic scheme (Busch & Muller, 2004). + ktd
sec−tert
λ0 δ0 + ktd
tert−tert
(δ0 )2
In this study, since temperature effects on the number of termi- rP0 = 2kd [I] + 2kd [] + kts [S]λ0 − (kp + kii )[M][P0 ]
nal double bonds are of interest, β-scission and chain transfer rP1 =2kti [M]2 +kp [M][P0 ]−kp [M][P1 ]−kii [P0 ][M]+kdp [P2 ]
to monomer reactions are considered in the model develop- rP2 = kβ δ0 + kp [M][PTDB ] + kdp ([P3 ] − [P2 ])
1
ment. Along with chain transfer to monomer, nBA has also been
reported to undergo chain transfer to solvent (Maeder & Gilbert, rPTDB = ktm [M]λ0 − kp [M][PTDB
1 ] + kdp [P2 ]
1
1998; Peck et al., 2002; Raghuram & Nandi, 1969). Upon chain- (7)
transfer to monomer, a live secondary radical of length one or the
monomer itself (PTDB1 ) is created. When this radical propagates
rλ̃0 = kp [M][P2 ] − kdp [P3 ] + kpq [M]δ0 − (kbb + ktm [M] + kts [S])λ0 − 2(ktc
sec−sec
+ ktd
sec−sec
)(λ0 )2 −
sec−tert
(ktc + ktd
sec−tert
)λ0 δ0
rδ0 = kbb λ0 − (kpq [M] + kβ )λ0 − 2(ktc
tert−tert
+ ktd
tert−tert
)(δ0 )2 − (ktc
sec−tert
+ ktd
sec−tert
)λ0 δ0
rμ0 = (ktm [M] + kts [S])λ0 + kβ δ0 + (ktc
sec−sec
+ 2ktd
sec−sec
)(λ0 )2 + (2ktc
sec−tert
+ 4ktd
sec−tert
)λ0 δ0 +
tert−tert
(ktc + 2ktd
tert−tert
)(δ0 )2
rλ̃1 = kp [M]([P2 ] + λ0 )−kdp λ1 +kpq [M]δ1 − (kbb + ktm [M] + kts [S])λ1 − 2(ktc
sec−sec
+ ktd
sec−sec
)λ0 λ1
−(ktc
sec−tert
+ ktd
sec−tert
)λ1 δ0
rδ1 = kbb λ1 − (kpq [M] + kβ )δ1 − 2(ktc
tert−tert
+ ktd
tert−tert
)δ1 δ0 − (ktc
sec−tert
+ ktd
sec−tert
)λ0 δ1 (8)
rμ1 = (ktm [M] + kts [S])λ1 + kβ δ1 + 2ktc sec−sec
(λ0 λ1 ) + 2ktc
sec−tert
(λ0 δ1 + λ1 δ0 )+
tert−tert
2ktc (δ0 δ1 ) + 2ktd
sec−sec
λ0 λ1 + 2ktd
sec−tert
(λ1 δ0 + λ0 δ1 ) + 2ktdtert−tert
(δ0 δ1 )
rλ̃2 = kp [M]([P2 ] + λ0 + 2λ1 ) − kdp λ2 + kpq [M]δ2 − (kbb + ktm [M] + kts [S])λ2
−2(ktc
sec−sec
+ ktd
sec−sec
)λ0 λ2 − (ktc
sec−tert
+ ktd
sec−tert
)λ2 δ0
rδ2 = kbb λ2 − (kpq [M] + kβ )δ2 − 2(ktc
tert−tert
+ ktd
tert−tert
)δ2 δ0 − (ktc
sec−tert
+ ktd
sec−tert
)λ0 δ2
rμ2 = (ktm [M] + kts [S])λ2 + kβ δ2 + 2ktc
sec−sec
[λ0 λ2 + (λ1 )2 ] + 2ktc
sec−tert
(λ0 δ2 + 2λ1 δ1 + λ2 δ0 )+
tert−tert
2ktc [δ0 δ2 + (δ1 )2 ] + 2ktd
sec−sec
λ0 λ2 + 2ktd
sec−tert
(λ2 δ0 + λ0 δ2 ) + 2ktd
tert−tert
(δ0 δ2 )
further, a TDB is created for the resulting chain. The reactive where the moments of the chain length distribution of the live
solvent species (i.e., in this study, xylol radical) is represented secondary propagating radicals, live tertiary propagating radi-
as a live chain of length zero (P0 ), and is assumed to propagate cals, and the dead chains are defined as:
at the same rate as a secondary radical.
While termination for acrylates usually take place almost ∞
 ∞
 ∞

exclusively by combination, at high temperatures, the extent rλ̃k = nk Pn , δk = n k Qn , μk = nk Dn ,
of termination by disproportionation can increase (Odian., n=3 n=3 n=3
1991). It is assumed that termination occurs by both modes, λk = λ̃k + [P0 ] + 1 [P1 ] + 2 [P2 ]
k k
and the estimates reported in Peck and Hutchinson (2004)
on the relative extent of each termination mode are used in Batch process model: Assuming perfect mixing and constant
this study to describe the various termination reactions. The reaction solution volume, species and structural character bal-
 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167 2161

ances for a batch reactor lead to: Table 3

Typical reaction network for solution free-radical homo-polymerization at high
d[M]
= rM , [M](0) = [M]0 temperatures
dt kd
d[S] 2 →2P0 Initiation
= rS , [S](0) = [S]0 kti
dt 2M→2P1
d[I] kd
I2 →2P0
= rd , [I](0) = [I]0 kii
dt P0 + M→P1
d[]
= r , [](0) = []0 kp
dt PTDB
1 + M  P2 + (TDB) Propagation
d[TDB] kdp
= rTDB , [TDB](0) = 0 kp
dt Pn + M  Pn+1 Propagation
d[SCB] kdp
= rSCB , [SCB](0) = 0 ktm
Pn + M→Dn + PTDB Chain transfer to monomer
dt 1
kts
d[P0 ] Pn + S→Dn + P0 Chain transfer to solvent
= rP0 , [P0 ](0) = 0 (9)
dt kbb
d[P1 ] Pn →Qn Backbitting n≥3
= rP1 , [P1 ](0) = 0 kpq
dt Qn + M→Pn+1 + (SCB) Propagation n≥3
d[P2 ] kβ
= rP2 , [P2 ](0) = 0 Qn →Dn−2 + P2 + (TDB) ␤-scission n≥3
dt sec−sec
ktc
d[PTDB ] Pn + Pm → Dn+m Termination
1
= rPTDB , [PTDB
1 ](0) = 0 sec−sec
ktd
dt 1
Pn + Pm → Dn + Dm + (TDB) Termination
dλ̃k sec−tert
ktc
= rλ̃k , λ̃k (0) = 0, k = 0, 1, 2 Pn + Qm → Dn+m Termination m≥3
dt sec−tert
ktd
dδk Pn + Qm → Dn + Dm + (TDB) Termination n, m ≥ 3
= rδk , δk (0) = 0, k = 0, 1, 2 tert−tert
ktc
dt Qn + Qm → Dn+m Termination m≥3
dμk tert−tert
= rμk , μk (0) = 0, k = 0, 1, 2 ktd
Qn + Qm → Dn + Dm + (TDB) Termination m≥3
dt
Average number of terminal double bonds per 100 repeat unit
(TDBH), average number of short branch points per 100 repeat steady-state assumption:
units (SCBH), and average number of terminal solvent groups  0.5
per 100 repeat unit (TSGH) are calculated from: 2kti [M]2 + 2kd [I] + 2kd 
[R] ≈
[TDB] [SCB] ktd + ktc
TDBH = 100 , SCBH = 100 ,
μ1 μ1 1
[P] ≈ [R]
[S]0 − [S] 1+(ktm )/(kp ) + (kbb )/(kpq [M]+kβ +kt [R]) (12)
TSGH = 100 (10) kbb
μ1 [Q] ≈ [P]
kpq [M] + kβ + kt [R]
ktm
[PTDB ] ≈ [P]
3.1.1. Simplified reaction rate equations kp
On the basis of the reaction mechanism in Table 4 (mod-
eling macromolecular kinetics as small-molecular kinetics Batch process model: Assuming perfect mixing and con-
(Villermaux & Blavier, 1984)), the rates of production of dif- stant reaction solution volume, species and structural character
ferent species and structural characters are given by: balances for a batch reactor lead to:

rM = −2kd [M]2 − kp [M]([P] + [PTDB ]) − kpq [M][Q] − ktm [M][P] + kdp ([P] + [P2 ])
rS = −kts [S][P]
rI = −kd [I]
r = −kd []
(11)
rTDB = kβ [Q] + kp [M][PTDB ] + ktd
sec−sec
[P]2 + ktd
sec−tert
[P][Q] + ktd
tert−tert
[Q]2
rSCB = kpq [M][Q]
rD = kβ [Q] + (ktm [M] + kts [S])[P] + (ktc
sec−sec
+ 2ktd
sec−sec
)[P]2
+(ktc
sec−tert
+ 2ktd
sec−tert
)[P][Q] + (ktc
tert−tert
+ 2ktd
tert−tert
)[Q]2
where the concentrations of the free radicals are given by the
following algebraic equations, obtained by applying the quasi-
2162 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167

Table 4 tion of a resin with acceptable properties. Furthermore, viscosity

Simplified reaction network for solution homo-polymerization at high adjustment of the paint by an auto painter improves brushability
temperatures
and sprayability of the paint but alone cannot ensure satisfactory
kd
2 →2P Initiation quality of the paint finish.
kti
2M→2P
kd 3.3. Control problems
I2 →2P
kii
P + M→P
Control of significantly-high-temperature polymerization
kp
reactors is more challenging than their low-temperature coun-
PTDB + M  P + (TDB) Propagation
kdp terparts. First, the dynamics of high-temperature polymerization
kp
reactors are more complex and nonlinear and faster. Initiators
P + MP Propagation
kdp decompose at higher rates at higher temperatures and as such
ktm
P + M→D + PTDB Chain transfer to monomer drive faster kinetics. This adds to the difficulty of control. Sec-
P + S→D + P
kts
Chain transfer to solvent ond, the quality of a high-solids low-molecular-weight resin
kbb
(polymer solution) is more complex to characterize and measure.
P→Q Backbitting Third, with the same number of manipulated inputs one has to
kpq
Q + M→P + (SCB) Propagation control more distributions to ensure satisfactory product quality;
kβ high-temperature polymerization reactors are less controllable.
Q→D + P + (TDB) ␤-scission
sec−sec
Fourth, from the same number of output measurements one has
ktc
2P → D Termination to estimate more distributions to ensure satisfactory product and
sec−sec
ktd process monitoring; high-temperature polymerization reactors
2P → 2D + (TDB) Termination
sec−tert
ktc
are less observable.
P+Q → D Termination
sec−tert
ktd
P + Q → 2D + (TDB) Termination 4. Multi-rate nonlinear state estimation in a
tert−tert
ktc high-temperature polymerization reactor
2Q → D Termination
tert−tert
ktd
2Q → 2D + (TDB) Termination The method-of-moments model presented in Rantow,
Soroush, Grady, and Kalfas (2006) is able to successfully pre-
dict monomer conversion and number-average molecular weight
d[M] (Fig. 1), but it over-predicts the polymer polydispersity index
= rM , [M](0) = [M]0
dt and under-predicts the concentration of terminal double bonds
d[S] (Fig. 2). In view of these predictive deficiencies, the multi-rate
= rS , [S](0) = [S]0
dt state estimator design method described in Zambare, Soroush,
d[I] and Ogunnaike (2003) is used; the poor model prediction is
= rI , [I](0) = [I]0
dt improved through adding an on-line corrective term representing
d[TDB] (13) the difference between on-line measurements and the model-
= rTDB , [TDB](0) = 0 predicted values of these measured variables (i.e., using the
dt
d[D] feedback feature of the state estimator). The performance of the
= rD , [D](0) = 0 state estimator in predicting number- and weight-average molec-
dt
d[PTDB ] ular weights (chromatographic properties) and polymer chain
= rPTDB , [PTDB ](0) = 0 microstructures (spectroscopic properties) is then evaluated.
dt
The estimator is designed on the basis of a simplified reduced-
order polymerization reactor model for high-temperature
3.2. Product quality polymerization of n-butyl acrylate; the model is based on the
mechanism in Table 4. The polymerization reaction network
The quality of a high-solids low-molecular-weight resin under study contains 15 irreversible reaction pathways and 10
(polymer solution) is more complex to characterize and con- rate coefficients. The simplified reactor model is of order 7 (has
trol than that of a low-solids high-molecular-weight resin. It 7 state variables), compared to the 20 state variables of the model
depends on the polymer chain length distribution (including the based on the mechanism presented in Table 3.
degree of short and long chain branching), the functional group
distribution, the level of solids (or polymer content) in the resin 4.1. Off-line and on-line measurements
solutions, as well as other properties including the amount of
terminal bond unsaturation in the case of polyacrylates made at The measurements of the reactor temperature and the feed
higher temperature. All of these properties, set at the polymer- flow rates of monomer, initiator solution and solvent are the fre-
ization stage, influence the properties of the final paint finish. quent measurements, and the measurements of molecular weight
As such, controlling the number- and weight-average molecular averages and microstructural quantities are the infrequent mea-
weights and the solids level can no longer ensure the produc- surements. The multi-rate state estimator estimates from these
 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167 2163

Fig. 2. Predictions made by the method-of-moments model reported in Rantow

et al. (2006) (solid lines = model prediction; triangles = measurements). Details
of the experimental conditions are given in the experimental procedure section
(also depicted in Fig. 3).

mixture inside the reactor, and the feed flow rates are measured
at a sampling rate of one minute with almost no time delay; these
are frequent measured inputs.

4.2. Polymerization reactor model

The reduced-order reactor model is then developed by writ-

ing mass and structural-character balances for the semi-batch
Fig. 1. Predictions made by the method-of-moments model reported in Rantow
et al. (2006) (solid lines = model prediction; triangles = measurements). Details
of the experimental conditions are described in the experimental procedure
section (also depicted in Fig. 3).

measurements. A sample is taken almost every 40 min. The

drawn samples are diluted (50/50, v/v) in a cold inhibitor solution
(1000 pap 4-methods phenol in xylene). Polymer solids content
in the samples is determined by gravimetry, while molecular
weight distributions are determined using an HP 1090 high-
performance liquid chromatography (HAPLY). Polymer-chain
microstructure properties are measured using proton nuclear
magnetic spectroscopy (1 HNMR) and carbon NMR (13 C NMR).
More details on these analytical procedures can be found in
Rantow et al. (2006). Time delay associated with each of the
infrequent measurements is 30 min. The concentrations of ter-
minal double bonds and short chain branches are inferred from Fig. 3. Profiles of the reactor temperature and the masses of the monomer (n-
the NMR measurements. The reactor temperature, the volume butyl acrylate), solvent (xylene), and initiator (t-butyl peroxy acetate) added to
of monomer added to the reactor, the volume of the reacting the reactor, corresponding to Figs. 1 and 2.
2164 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167

reactor. An energy balance is not needed for the process model, Table 5
because the reactor temperature is an input to the model (is a Local observability
measured input). It should be noted that the density of reacting State variable Measurement
solution is assumed to be constant. This assumption is deemed [S] [TDB] [SCB] Mn Mw
appropriate as polymerization is carried out at the lower end of
the concentration range (less than 40 wt.%), the operating tem- [I] Yes Yes Yes Yes Yes
[M] Yes Yes Yes Yes Yes
perature range is narrow, and that the solvent (xylene) and the [S] Yes No No Yes Yes
monomer (n-butyl acrylate) have similar densities. The model [TDB] No Yes No No No
parameter (rate coefficient) values are taken from (Rantow et [SCB] No No Yes No No
al., 2006). The resulting ordinary differential equations (ODEs) [D] No No No Yes Yes
describing the dynamics of the reactor are:
⎡ ⎤ ⎡ ⎤
[İ] fI ([I], T, V, FI , FM , FS ) measurements obtained from batch experiments with the same
⎢ [Ṁ] ⎥ ⎢ ⎥ polymerization system (Rantow, 2006).
⎢ ⎥ ⎢ fM ([I], [M], T, V, FI , FM , FS ) ⎥
⎢ ⎥ ⎢ ⎥
⎢ [Ṡ] ⎥ ⎢ fS ([I], [M], [S], T, V, FI , FM , FS ) ⎥
⎢ ⎥=⎢ ⎥
⎢ [TḊB] ⎥ ⎢ f ⎥ 4.3. Reduced-order multi-rate state estimator design
⎢ ⎥ ⎢ TDB ([I], [M], [TDB], T, V, FI , FM , FS ) ⎥
⎢ ⎥ ⎢ ⎥
⎣ [SĊB] ⎦ ⎣ fSCB ([I], [M], [SCB], T, V, FI , FM , FS ) ⎦ On the basis of the developed reduced-order model, a
[Ḋ] fD ([I], [M], [S], [D], T, V, FI , FM , FS ) reduced-order multi-rate state estimator is designed using the
(14) method described in Tatiraju, Soroush, and Ogunnaike (1999)
and Zambare et al. (2003). The estimator is in the form:
where

1 ρI
fI ([I], T, V, FI , FM , FS ) = rI ([I], T ) − [I](FI + FM + FS ) − FI
V MwI
1
fM ([I], [M], T, V, FI , FM , FS ) = rM ([M], [P], [Q], [PTDB ], T ) −
V 
ρM
[M](FI + FM + FS ) − FM
MwM

1 ρS
fS ([I], [M], [S], T, V, FI , FM , FS ) = rS ([S], [P], T ) − [S](FI + FM + FS ) − FS
V MwS
F I + FM + FS
fTDB ([I], [M], [TDB], T, V, FI , FM , FS ) = rTDB ([Q], [M], [PTDB ], T ) − [TDB]
V
F I + FM + FS
fSCB ([I], [M], [SCB], T, V, FI , FM , FS ) = rSCB ([Q], [M], T ) − [SCB]
V
FI + F M + F S
fD ([I], [M], [S], [D], T, V, FI , FM , FS ) = rD ([Q], [M], [S], [P], T ) − [D]
V
The state variables, [I], [M], [S], [TDB], [SCB and [D] are the
concentration of the initiator (t-butyl phenoxy acetate), concen- [˙Î] = fI ([Î], T, V, FI , FM , FS )
tration of the monomer (n-butyl acrylate), concentration of the ˙ = f ([Î], [M̂], T, V, F , F , F )
solvent (xylene), concentration of terminal double bonds, con- [M̂] M I M S
centration of short chain branches, and concentration of ‘dead’ ˙
[Ŝ] = fS ([Î], [M̂], [Ŝ], T, V, FI , FM , FS )
polymer chains, respectively. The measured inputs V, T, FI , FM , ˙ =f
and FS are the volume of the reacting mixture, the reactor tem- [TD̂B] TDB ([Î], [M̂], [T D̂B], T, V, FI , FM , FS )
perature, and the volumetric flow rates of the initiator solution, + L41 {[TDB]∗ − [TD̂B]} (15)
the monomer and the solvent, respectively. ˙ = f ([Î], [M̂], [SĈB], T, V, F , F , F )
[SĈB] SCB I M S
Note that weight-average molecular weight is calculated by ∗
multiplying number-average molecular weight by polydispersity + L 52 {[SCB] − [S ĈB]}
index, PDI, given by the empirical correlation (Rantow, 2006): ˙ = f ([Î], [M̂], [Ŝ], [D̂], T, V, F , F , F )
[D̂] D I M S
 
ρ M VM + L63 {Mn∗ − HMn (VM , V, [M̂], [D̂])}
PDI = c1 + c2 + c3 T
ρ̄V
where each term with an asterisk represents the predicted present
where the correlation constants c1 = 2.66031, c2 = 0.03337 value of the variable with the corresponding infrequent measure-
and c3 = −0.00929. The second term on the right hand side of ment. This present value is obtained by the least-squared-error
the correlation is c2 times the weight fraction of the monomer fed fit of a line to the three most recent measurements of the infre-
to the reactor. The correlation constants are obtained by fitting quent measured output. The second term in the right hand
the correlation to PDI, temperature and initial monomer-content side of the estimator equation is simply a corrective feedback
 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167 2165

term; as long as there are discrepancies between the predicted

present value of a variable (variable with asterisk) and the cor-
responding estimate (variable with hat), the corrective action
is made.
The estimator calculates continuous estimates of the con-
centrations of the initiator, the solvent, terminal double bonds,
short polymer chain branches, and dead polymer chains
from:

• The frequent measured inputs: the reactor temperature, the

reacting mixture volume, the flow rates of the monomer, the
solvent and the initiator solution feed streams, and the volume
of monomer fee to the reactor; and
• The infrequent measured outputs: the concentrations
of terminal double bonds and short polymer chain
branches, and the number-average molecular weight of the
polymer.

An analysis of the local observability of the system

(Table 5) (Tatiraju et al., 1999; Zambare et al., 2003) led
to setting all of the estimator gain entries except for the
entries L41 , L52 and L63 to zero. The non-zero estimator
gain entries L41 = 1.5 × 10−1 , L52 = 1.5 × 10−1 , and L63 =
−7.0 × 10−6 , which ensure that all eigenvalues of the Jacobian
of the estimator error dynamics lie in the left half of the complex Fig. 5. Estimated vs. measured values of the concentrations of terminal dou-
plane. The non-zero term in each row in effect indicates which ble bonds and short chain branches (solid lines = estimates obtained using the
slow measurement is used to correct the state variable estimate estimator; diamonds = measurements).
corresponding to the row.
4.4. State estimator performance

The performance of the estimator in providing estimates of

the number-average and weight-average molecular weights is
shown in Fig. 4. The estimator initial conditions were set to those
of the actual reactor. In the estimator implementation, the entries
L41 , L52 and L63 are set to zero until the first set of infrequent
measurements are available. To calculate the present value of
each infrequently measured output, a linear polynomial is fitted
to the most recent three data points (measurements), except for
when two sets of infrequent measurements are available. In the
latter case, a line is fitted to two measurements.
The estimates of the polymer microstructure quantities calcu-
lated by the multi-rate state estimator are shown in Fig. 5, which
demonstrate that the state estimator provides accurate continu-
ous estimates of the variables that have slow measurements. The
selection of the non-zero estimator gain entries on the basis of
the local observability analysis is shown to be good enough to
ensure satisfactory estimates.

5. Concluding remarks

An evaluation of systems challenges in free-radical solution

homo-polymerization as temperature increases was presented.
The inaccuracy of classical chain-polymerization kinetic mod-
Fig. 4. Estimated vs. measured values of number- and weight-average
els to describe the dynamics of these processes was discussed.
molecular weights (solid lines = estimates obtained using the estimator; dia- The more complex nature of the dependence of quality of
monds = measurements). high-solids low-molecular weight resins on operating conditions
2166 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167

was explained; the quality of these resins cannot be character- Cerinski, B., & Jelencic, J. (2002). Modeling of high-pressure ethylene poly-
ized by measurements of solid content and average molecular merization. I. Kinetic parameters of oxygen initiation. Journal of Applied
weights alone. Advanced chromatographic and spectroscopic Polymer Science, 83, 2043–2051.
Chiefari J., Chong Y. K., Ercole F., Kristina J., Le T.P., Mayadunne R. T. A.,
measurements of the concentrations of functional groups, termi- et al. (1998). Living free-radical polymerization by reversible addition-
nal double bonds, chain branches and solvent groups, provide fragmentation chain transfer: The RAFT process. Macromolecules, 31,
a fuller characterization of the polymer quality. Mathematical 5559.
models describing these processes are of higher order and are Chiefari, J., Jeffery, J., Mayadunne, R. T. A., Moad, G., Rizzardo, E., & Thang, S.
more complex than their low-temperature counterparts. These H. (1999). Chain transfer to polymer: A convenient route to macromonomers.
Macromolecules, 32, 7700–7702.
processes are of lower degree of controllability, as a higher Clouet, G., Chaumont, P., & Corpart, P. (1993). Studies on bulk-polymerization
number of properties should be controlled (to achieve quality of methylmethacrylate. 1. Thermal polymerization. Journal of Polymer Sci-
control) using the same number of manipulated inputs. For these ence Part A-Polymer Chemistry, 31, 2815–2824.
same reasons, these processes have less degree of observability Fernandez-Garcia, M., Fernandez-Sanz, M., & Madruga, E. L. (2000). A kinetic
than their low-temperature counterparts, unless more advanced study of butyl acrylate free radical polymerization in benzene solution.
Macromolecular Chemistry and Physics, 201, 1840–1845.
measurements are available. As an application of reduced-order Georges, M. K., Veregin, R. P. N., Kazmaier, P. M., & Hamer, G. K. (1993).
models, a case study of multi-rate nonlinear state estimation in Narrow molecular weight resins by a free-radical polymerization process.
an nBA polymerization reactor was presented. Real-time results Macromolecules, 26, 2987.
from the on-line implementation of the estimator (soft sensor) Grady, M. C., Simonsick, W. J., & Hutchinson, R. A. (2002). Studies of higher
confirmed the reliability of the estimator to provide accurate temperature polymerization of n-butyl methacrylate and n-butyl acrylate.
Macromolecular Symposia, 182, 149–168.
estimates of spectroscopic and chromatographic polymer prop- Hakim, M., Verhoeven, V., McManus, N. T., Dube, M. A., & Penlidis, A.
erties. The feedback feature of ‘closed-loop’ estimator provided (2000). High-temperature solution polymerization of butyl acrylate/methyl
the reactor model with very needed robustness to process-model methacrylate: Reactivity ratio estimation. Journal of Applied Polymer Sci-
mismatch. ence, 77, 602.
Lehrle, R. S., & Shortland, A. (1988). A study of the purification of methyl
mecthacrylate suggests that the ‘thermal’ polymerisation of this monomer is
Acknowledgements initiated by adventitious peroxides. European Polymer Journal, 24, 425–429.
Lingnau, J., & Meyerhoff, G. (1984a). The spontaneous polymerization of
This study was supported in part by DuPont Marshall Labo- methyl methacrylate. 7. External heavy atom effect on the initiation. Makro-
ratory and the National Science Foundation (NSF) through the molekulare Chemie, 185, 587–600.
Lingnau, J., & Meyerhoff, G. (1984b). Spontaneous polymerization of methyl
grant CBET–0651706. Any opinions, findings, conclusions or methacrylate. 8. Polymerization kinetics of acrylates containing chlorine
recommendations expressed in this material are those of the atoms. Macromolecules, 17, 941–945.
authors and do not necessarily reflect the views of DuPont or Lingnau, J., Stickler, M., & Meyerhoff, G. (1980). The spontaneous polymer-
the NSF. The authors would like to thank Felix S. Rantow for ization of methyl methacrylate. IV. Formation of cyclic dimers and linear
his invaluable help in the simulation study. trimers. European Polymer Journal, 16, 785–791.
Lingnau, J., Stickler, M., & Meyerhoff, G. (1983). The spontaneous polymer-
ization of methyl methacrylate. 6. Polymerization in solution: Participation
References of transfer agents in the initiation reaction. Polymer, 24, 1473–1478.
Maeder, S., & Gilbert, R. G. (1998). Measurement of transfer constant for butyl
Adamsons, K., Blackman, G., Gregorovich, B., Lin, L., & Matheson, R. (1998). acrylate free radical polymerization. Macromolecules, 31, 4410–4418.
Progress in Organic Coatings, 34, 64. McCord, E. F., Shaw, W. H., & Hutchinson, R. A., Jr. (1997). Short-chain branch-
Ahmad, N. M., Heatley, F., & Lovell, P. A. (1998). Chain transfer to polymer ing structures in ethylene copolymers prepared by high-pressure free-radical
in free-radical solution polymerization of n-butyl acrylate studied by NMR polymerization: An NMR analysis. Macromolecules, 30, 246–256.
spectroscopy. Macromolecules, 31, 2822–2827. McManus, N. T., Penlidis, A., & Dube, M. A. (2002). Copolymerization of
Asua, J. M., Beuermann, S., Buback, M., Castignolles, P., Charleux, B., Gilbert, alpha-methyl styrene with butyl acrylate in bulk. Polymer, 43, 1607–1614.
R. G., et al. (2004). Critically evaluated rate coefficients for free-radical Mesrobian, R. B., & Tobolsky, A. V. (1947). Some structural and chemical
polymerization, 5. Propagation rate coefficient for butyl acrylate. Macro- aspects of aging and degradation of vinyl and diene polymers. Polymer
molecular Chemistry and Physics, 205, 2151–2160. Science, 2, 463.
Balchan, A. S., Klein, J. A., & Klein, F. G. (1995). Loss prevention and safety Nikitin, A. N., Castignolles, P., Charleux, B., & Vairon, J. P. (2003). Deter-
promotion in the process industries, Vol. 1. Belgium: Antwerp., pp. 1–14. mination of propagation rate coefficient of acrylates by pulsed-laser
Beuermann, S., & Buback, M. (2002). Rate coefficients of free-radical polymer- polymerization in the presence of intramolecular chain transfer to polymer.
ization deduced from pulsed laser experiments. Progress in Polymer Science, Macromolecular Rapid Communications, 24, 778–782.
27, 191–254. Nising, P., Meyer, T., Carloff, R., & Wicker, M. (2005). Thermal initiation of
Brand, E., Stickler, M., & Meyerhoff, G. (1980). Spontaneous polymerization MMA in high temperature radical polymerizations. Macromolecular Mate-
of methyl methacrylate, 3. Reaction behavior of the unsaturated dimer upon rials and Engineering, 290, 311–318.
polymerization. Makromolekulare Chemie, 181, 913–921. Odian, G. G. (1991). Principles of polymerization (3rd ed.). Hoboken, NJ, USA:
Buback, M., Egorov, M., & Feldermann, A. (2004). Chain-length dependence of John Wiley and Sons, Inc.
termination rate coefficients in acrylate and methacrylate homopolymeriza- Peck, A. N. F., & Hutchinson, R. A. (2004). Secondary reactions in the high-
tions investigated via the sp-plp technique. Macromolecules, 37, 1768–1776. temperature free radical polymerization of butyl acrylate. Macromolecules,
Busch, M., & Muller, M. (2004). Simulating acrylate polymerization reactions: 37, 5944–5951.
Toward improved mechanistic understanding and reliable parameter esti- Peck, A. N. F., Hutchinson, R. A., & Grady, M. C. (2002). Branching and scission
mates. Macromolecular Symposia, 206, 399–418. reactions in high temperature acrylate polymerisations. Polymer Preprints,
Cao, G., Zhu, Z., Zhang, M., & Yuan, W. (2004). Kinetics of butylacrylate 43, 154–155.
polymerization in a starved feed reactor. Journal of Applied Polymer Science, Prane, J. W. (1986). Introduction la polymers and resins. Philadelphia, PA:
93, 1519–1525. Federation of Societies for Coatings Technology., p. 24.
 M. Soroush et al. / Computers and Chemical Engineering 32 (2008) 2155–2167 2167

Raghuram, P. V. T., & Nandi, U. S. (1969). Studies on the polymerization of Superintendent of Documents, Title 1. (1990). US Government Printing Office,
ethyl acrylate. II. Chain transfer studies. Journal of Polymer Science Part Washington, DC, p. 1.
A-1, 7, 2379–2385. Superintendent of Documents. (1990). Clean Air Act Amendments of 1990.
Rantow, F. S. (2006). Secondary reactions in high-temperature free-radical poly- Title 111, US Government Printing Office, Washington, DC, p. 236.
merization of alkyl acrylates. Ph.D. Thesis, Philadelphia, PA, USA: Drexel Tatiraju, T., Soroush, M., & Ogunnaike, B. A. (1999). Multirate nonlinear state
University. estimation with application to a polymerization reactor. AIChE Journal, 45,
Rantow, F. S., Soroush, M., Grady, M. C., & Kalfas, G. A. (2006). Spontaneous 769–780.
polymerization and chain microstructure evolution in high-temperature solu- Villermaux, J., & Blavier, L. (1984). Free radical polymerization engineering.
tion polymerization of n-butyl acrylate. Polymer, 47, 1423–1435. I. A new method for modeling free radical homogeneous polymerization
Reisch, M. S. (1993). Paints and coatings. Chemical Engineering News, 71, reactions. Chemical Engineering Science, 39, 87–99.
34–37 (October). VOC’s Directive, EU Committee of the American Chamber of Commerce in
Schmidt, A. D., Clinch, A. B., & Ray, W. H. (1981). The dynamic behavior of Belgium, ASBL/VZw, Brussels, July 1996.
continuous polymerization reactors-III. Chemical Engineering Science, 39, Wang, S. J., & Matyjaszewski, K. (1995). Controlled “living” radical poly-
419–432. merization. halogen atom transfer radical polymerization promoted by a
Schultz, K.R. Proceedings of Advance Coat. Technology Conference, Chicago, Cu(I)/Cu(II) redox process. Macromolecules, 28, 7901.
IL, 1992. Wayland, B. B., Mukerjee, S., Poszmik, G., Woska, D. C., Basickes, L., Gridnev,
Stickler, M., & Meyerhoff, G. (1978). Thermal polymerization of methyl A. A., et al. (1998). Control of radical polymerizations by metalloradicals,
methacrylate, 1. Polymerization in bulk. Makromolekulare Chemie, 179, ACS Symposium Series, 685, 305–315 (Controlled Radical Polymerization).
2729–2745. Zambare, Z., Soroush, M., & Ogunnaike, B. A. (2003). A method of
Stickler, M., & Meyerhoff, G. (1981). The spontaneous thermal polymerization robust multi-rate state estimation. Journal of Process Control, 13, 337–
of methyl methacrylate. 5. Experimental study and computer simulation of 355.
the high conversion reaction at 130 c. Polymer, 22, 928–933.