MECHANISTIC MODELING AND MODEL-BASED STUDIES IN SPONTANEOUS SOLUTION POLYMERIZATION OF ALKYL ACRYLATE MONOMERS

A Thesis Submitted to the Faculty of Drexel University by Felix Suryaman Rantow in partial fulfillment of the requirements for the degree of Doctor of Philosophy August 2006

ii

Copyright 2006 by Felix Suryaman Rantow All Rights Reserved

iii ACKNOWLEDGEMENTS I thank my co-advisors Prof. Masoud Soroush and Dr. Michael Grady for the opportunity, the guidance and the invaluable advice given to me throughout my graduate studies. I was fortunate to have been given the chance to engage in an applied and interactive research work with people at Drexel Chemical Engineering Department and Dupont Marshall Laboratory Process Engineering Group. Undoubtedly the experience has positively influenced my personal and professional growth. Prof. Soroush and Dr. Grady have always emphasized the importance of ingenuity, originality, passion and effective communication in making meaningful scientific contributions. At Drexel, I was fortunate to have been given the chance to interact with fellow graduate students. They have played roles in shaping how I perceive a stimulating work environment should be. I thank students in Prof. Soroushs group, Nasir Mehranbod, Congling Quan, Chanin Panjapornpon, Swa Swarup Metta and Sriraj Srinivasan for contributing positively to the work environment. We as a group have shared work space with Prof. Nily Dans and Prof. Cameron Abrams groups. I thank Veena Pata, Kaveh Komaee, Kevin Towles, Ehsan Jabbarzadeh and Yelena Sliozberg for the stimulating discussions on formal and informal matters. I also thank other students in the Chemical Engineering Department who I have been given the chance to interact with daily, and through many exciting graduate student activities. I believe these are also important to the creation of a healthy work environment. There are too many names to mention. I thank them nonetheless. At Dupont Marshall Laboratory, I was very fortunate to have been given the chance to interact with the process operators, the process engineers and the scientists. I was able to absorb the work culture and the professionalism shown by the people I interact with. I would like to thank Dr. George Kalfas for the invaluable insights on the modeling and optimization studies. I would also like to

iv thank Dr. Thomas Scholz and Mark Karalis for help with the polymer characterization work. I thank Chris Bruni and the Drexel Co-op students Amutha Jeyarajasingam and Erik Amrine for the help with scheduling experimental runs. And I also thank Peter Markos, Mike Darczuk and Mike Rullo for running and tending the experiments. I would also like to express my gratitude to my committee members Prof. Masoud Soroush, Prof. Charles Weinberger, Prof. Giuseppe Palmese of Drexel, and Dr. Michael Grady, Dr. George Kalfas and Dr. Joan Hansen of DuPont. They have given me their valuable time and advice in the process of my thesis writing and defense. I would also like to acknowledge my undergraduate Professor, Prof. Raghunathan Rengaswamy, currently at Clarkson University, whose advice led me to pursue work under the guidance of Prof. Masoud Soroush. On a more personal note, I would not have been able to enjoy and truly experience, or absorb, each day without the unconditional support given by my parents, brother, and a special person in my life, Ervina Widjaja. I dedicate my graduate work, and the way I enjoyed it, to my family; my parents, Joannes and Jeanny Widodo Rantow, and my brother, Giovanni Darmawan Rantow.

Dedicated to my parents, Joannes and Jeanny Widodo Rantow, And my brother, Giovanni Darmawan Rantow

vi

Abstract Mechanistic Modeling and Model-Based Studies in Spontaneous Solution Polymerization of Alkyl Acrylate Monomers Felix Suryaman Rantow Masoud Soroush, Ph.D. and Michael Charles Grady, Sc.D.

High-temperature (above 120C) free-radical polymerization of alkyl acrylate monomers is studied to gain a better quantitative understanding of the governing reaction network behind the observed process measurement dynamics. At higher reaction temperatures, mechanistic models developed based on lowertemperature data fail to quantitatively predict the dynamics of polymer property indices such as monomer conversion and average molecular weights. In recent studies, this failure has been attributed to the occurrence of secondary chain reactions that become increasingly significant with temperature increase.

Experimental data obtained from high-temperature batch polymerization of nbutyl acrylate (nBA) in xylene points to the occurrence of short chain branch formation, as opposed to long chain branch formation. This observation substantiates the backbiting (intra-molecular chain transfer) mechanism postulated in earlier studies on alkyl acrylate polymerization.

vii In this study the activation energy is estimated through multivariate data fitting of several measurements. The estimated value of the activation energy of the postulated self-initiation reaction points to actual occurrence of self-initiation. The measurements include monomer conversion, number- and weight-average molecular weights and microstructure quantities (number-average of terminal solvent group, terminal double bond and chain branch per polymer chain). This provides an explanation to the observed spontaneous polymerization in alkyl acrylate polymerization at higher temperatures.

A free-radical polymerization mechanistic model for the spontaneous solution high-temperature polymerization of n-butyl acrylate in xylene is developed. This effort involves characterizing polymer solutions, postulating reaction networks, screening the networks by performing global parameter identifiability analyses, reducing the complexity of the developed mechanistic model, estimating unknown kinetic parameters, and validating the model. A method of global parameter identifiability analysis is developed for models that are affine in parameters. The model reduction techniques used in this study include the method of moments modeling method, the tendency modeling method, and the quasi-steady-state assumptions on the concentrations of free radicals. The model is used to (a) develop a model-based optimal control policy for a spontaneously-initiated semi-batch polymerization reactor and (b) design of a multi-rate state observer to estimate polymer-chain microstructure properties from currently-available on-line and off-line measurements.

viii

ix NOTATION A, [A] [A]B BP BPH D FA kbb kii kp kpb ktc ktd ktdb kti ktm ktp kts k LCB LCBH mI Concentration of species A in reactor, A = M, TDB, etc. Concentration of species A in feed flowrate of B, B = M, S, I Chain-branching = SCB + LCB (Chapter 3) Chain-branching per 100 monomer unit = SCBH + LCBH (Chapter 3) 'Dead' polymer chain Volumetric flow rate of species A, A = M, S, I Rate constant of backbiting reactions Rate constant of thermal initiator decomposition reactions (Chapter 3) Rate constant of propagation reactions Rate constant of chain-branch formation reactions (Chapter 3) Rate constant of termination by combination reactions Rate constant of termination by disproportionation reactions Rate constant of propagation on terminal double bond reaction Rate constant of self-initiation reactions Rate constant of chain-transfer to monomer reactions Rate constant of chain-transfer to polymer reactions Rate constant of chain-transfer to solvent reactions Rate constant of -scission reactions Long chain-branching Long chain-branching per 100 monomer unit Total mass of initiator used in reaction

x M Mn M N rA P Q R R1 R2 R3 R4 R S SCB SCBH T TC TDB TDBH V MWm Monomer Number-average molecular weight Number of discretized values for different feed flow rates Number of equal batch time intervals (Chapter 3) Rate of reaction of species A, A = M, TDB, etc. Live' polymer chain, mid-chain tertiary radical species Live' polymer chain, near-chain-end tertiary radical species Live' polymer chain, secondary radical species Secondary propagating radical (Chapter 3) Tertiary mid-chain propagating radical (Chapter 3) Tertiary near-chain-end propagating radical (Chapter 3) Unsaturated propagating radical (Chapter 3) Propagating radical = R1 + R2 + R3 + R4 (Chapter 3) Solvent Short chain-branching Short chain-branching per 100 monomer unit Reaction time Tertiary carbon in a polymer chain Terminal double bond Terminal double bond(s) per 100 monomer unit Volume of reactor Molecular weight of monomer

xi xm i Monomer conversion Weighting factor of an objective in the performance index

xii

TABLE OF CONTENTS

INTRODUCTION AND BACKGROUND ....................................................... 1 1.1 1.2 SCOPE OF RECENT STUDIES IN MECHANISTIC MODELING KINETICS 1.2.1 1.2.2 1.2.3 1.3 OF FREE-RADICAL SOLUTION HOMOOF SOLUTION AND BULK FREE-RADICAL HOMO-POLYMERIZATION... 3 POLYMERIZATION ...................................................................................... 5 High-Temperature free-radical solution homo-polymerization High-Temperature free-radical solution homo-polymerization Spontaneous thermal initiation in high-temperature free-radical with depropagation involving styrenic/acrylic monomers ....................... 6 with branching and/or scission involving styrenic/acrylic monomers ..... 7 polymerization involving styrenic/acrylic monomers ............................ 10 MODELING METHODS ............................................................... 12 EXPERIMENTAL AND ANALYTICAL PROCEDURES................ 20 POSTULATION OF REACTION NETWORK ............................... 22 PARAMETER ESTIMATION........................................................ 27 DISCUSSIONS AND ANALYSES................................................ 33 CONCLUSIONS........................................................................... 51 OPTIMAL CONTROL OF A SEMI-BATCH HIGH-

MECHANISTIC MODELING ....................................................................... 17 2.1 2.2 2.3 2.4 2.5

MODEL-BASED 3.1 3.2

TEMPERATURE POLYMERIZATION REACTOR ............................................. 52 EXPERIMENTAL FRAMEWORK................................................. 55 MODEL DEVELOPMENT ............................................................ 55 3.2.1 3.2.2 Kinetic Scheme and Rate Laws ............................................. 55 Process Model ....................................................................... 58

xiii 3.3 3.4 PARAMETER ESTIMATION AND MODEL VALIDATION ........... 60 OPTIMIZATION AND OPTIMIZATION VALIDATION .................. 62 3.4.1 3.4.2 3.4.3 3.5 4 Performance Index................................................................. 62 Optimization Constraints ........................................................ 63 Optimization ........................................................................... 64

CONCLUSIONS........................................................................... 70

REDUCED-ORDER MODEL FOR MONITORING SPECTROSCOPIC AND 4.1 4.2 4.3 4.4 4.5 EXPERIMENTAL PROCEDURE ................................................. 73 4.1.1 Off-Line and On-Line Measurements..................................... 75 REDUCED-ORDER MECHANISTIC POLYMERIZATION MODEL . ..................................................................................................... 78 REDUCED-ORDER MULTI-RATE STATE OBSERVER DESIGN... ..................................................................................................... 84 STATE OBSERVER PERFORMANCE........................................ 86 CONCLUDING REMARKS .......................................................... 90

CHROMATOGRAPHIC PROPERTIES .............................................................. 71

GLOBAL IDENTIFIABILITY OF REACTION RATE PARAMETERS IN 5.1 SCOPE AND PRELIMINARIES ................................................... 94 5.1.1 5.1.2 5.1.3 5.2 5.2.1 5.2.2 5.3 Local and Global Identifiability of Model Parameter and Model Distinguishability of Model Structures .................................... 96 Scope..................................................................................... 96 Example 1 .............................................................................. 99 Example 2 ............................................................................ 101 PARAMETRIC IDENTIFIABILITY OF CHAIN Structure.............................................................................................. 94

MECHANISTIC MODELS FOR THE CHAIN HOMOPOLYMERIZATION .......... 91

CHALLENGES IN FREE-RADICAL POLYMERIZATION............. 97

GLOBAL 5.3.1 5.3.2

HOMOPOLYMERIZATION SYSTEMS ..................................................... 103 Example 3 ............................................................................ 104 Example 4 ............................................................................ 106

5.4

CONCLUSIONS......................................................................... 109

xiv 6 CONCLUSIONS AND FUTURE DIRECTIONS......................................... 110 6.1 PRELIMINARY OBSERVATIONS IN HIGH-TEMPERATURE SPONTANEOUS SOLUTION HOMO-POLYMERIZATION OF SEVERAL ALKYL ACRYLATE MONOMERS ............................................................ 111 6.2 6.3 RECENT STUDIES IN SOLUTION POLYMERIZATION OF 2MODELING AND PARAMETER ESTIMATION ......................... 115 ETHYLHEXYL ACRYLATE AND N-OCTYL ACRYLATE.......................... 114 APPENDIX A .................................................................................................... 118 BIBLIOGRAPHY............................................................................................... 125 VITA ................................................................................................................. 144

xv

LIST OF TABLES

Table 1. Different sets of reaction mechanisms used in free-radical co- and homo-polymerization kinetic studies. IM = initiation by monomer, IO = initiation by peroxides, oxygen, IC = initiation by chain transfer agents, P = propagation, CTM = chain-transfer to monomer, CTA = chain-transfer to modifier, CTP = chaintransfer to polymer, CTS = chain-transfer to solvent, CTC = chain-transfer to Macromonomer, TC = termination by combination, TD = termination by disproportionation, D = depropagation, BS = beta-scission, BB = backbiting, TDB = propagation on a terminal double bond, S = Styrene, E = Ethylene. ................. 4 Table 2. Postulated polymerization reaction network for batch polymerization of nBA in xylene at 140-180C with a monomer content of 20-40 wt.%. The subscripts represent polymer chain length. ........................................................ 26 Table 3. Estimated rate constant values based on measurements of X, Mn, Mw and CBC at 160 and 180C, at an initial monomer content of 40wt.%, and the corresponding Arrhenius parameters. ................................................................ 31 Table 4. Reported and estimated rate constant values for high-temperature nBA polymerization. aThe self initiation rate constant was estimated while assuming that initiation by impurities did not occur
b

Inter-molecular chain transfer to

polymer was not considered in this study ........................................................... 34

xvi Table 5. Local sensitivity of parameter-measurement with respect to nominal parameter and measurement values at 160C, 200 min reaction time and initial monomer content of 40wt.% ............................................................................... 36 Table 6. Rate constant values for decomposition of oxygen and peroxides [Cerinski et al., 2002], used in simulations presented in Figure 8....................... 47 Table 7. Activation energies of self-initiating monomers. aInitial monomer content at 170C, ca. 20 min reaction time, 30% monomer conversion. bBased on the postulated participation of molecular oxygen in the self-initiation mechanism .... 50 Table 8. Values of the estimated parameters ..................................................... 60 Table 9. Optimization Settings............................................................................ 65 Table 10. Optimal feed and temperature profiles................................................ 66 Table 11. Key 1H-NMR and 13C-NMR peak assignments................................. 76 Table 12. Simplified (using the small-molecule representation) reaction network for polymerization of nBA in xylene, with a monomer content of 20-40 wt%, and at 140-180C. ..................................................................................................... 79 Table 13. Simplified (using the small-molecule representation) reaction network for polymerization of nBA in xylene, with a monomer content of 20-40 wt%, and at 140-180C. ..................................................................................................... 85 Table 14. Polymerization reaction networks for initiation-propagation-termination scheme............................................................................................................... 99

xvii Table 15. Polymerization reaction networks for initiation-propagation-transfer to polymer-termination scheme ............................................................................ 104 Table 16. Preliminary observation of alkyl substituent size effect on resulting molecular weight and monomer conversion. Experiments were carried out in batch, in four hours, in methyl amyl ketone in the absence of thermal initiators [Grady, Quan and Soroush, 2003].................................................................... 113 Table 17. Summary of known reaction mechanisms in nBA, nOA and 2-EHA polymerization (low to high temperature).......................................................... 116

xviii

LIST OF FIGURES

Figure 1. Quality of fit for parameter estimation based on measurements of X, Mn, Mw and CBC. Solid circle and square markers represent experimental data at 160 and 180C, with initial monomer content of 40 wt.%, respectively. Solid and dashed lines represent model predictions at 160 and 180C, with initial monomer content of 40 wt.%, respectively. ........................................................................ 30 Figure 2. Surface of the objective function (sum of squared error between model predictions and measurements of monomer conversion, number-average molecular weight, weight-average molecular weight and number-average chain branches per chain) of the parameter estimation as functions of kti and kbb at 160C, initial monomer content of 40wt.%.......................................................... 32 Figure 3. Model predictions vs. experimental data for X and Mw, predictions are based on estimated Arrhenius parameters (Table 1). Solid triangle and circle markers represent experimental data with initial monomer content of 20 and 40 wt.%, at 160C, respectively. Solid and dashed lines represent model predictions at initial monomer content of 20 and 40 wt.%, at 160C, respectively. ............... 38 Figure 4. Model predictions vs. experimental data for X and Mw, predictions are based on estimated Arrhenius parameters (Table 2). Solid triangle and circle markers represent experimental data at 140 and 160C, respectively, with initial monomer content of 40 wt.%. Solid and dashed represent model predictions at 140 and 160C, respectively, with initial monomer content of 40 wt.%............... 39

xix

Figure 5. Model predictions vs. experimental data for TDBC and CBC vs. temperature. predictions are based on estimated Arrhenius parameters (Table 2). Solid markers represent experimental data. Experimental data at 160, 170 and 180C have initial monomer contents of 40, 33, and 40 wt.%, respectively (batch), at reaction time of ca. 40-50 minutes. The experimental data at 140C has a final monomer content of ca. 50 wt.% (starved-feed, 90 minute feed time), at reaction time of 90 minutes. The experimental data at 140C is recalculated from [Peck et al., 2004]. Empty markers represent batch model predictions at ca. 60 minute reaction time, and the solid lines are plotted only to connect the model predictions. ......................................................................................................... 40 Figure 6. (Top) Model predictions vs. experimental data for CBC, predictions are based on estimated Arrhenius parameters (Table 2). Solid square and circle markers represent experimental data at 160 and 180C , with initial monomer content of 40 wt.%, respectively. Solid and dashed lines represent model predictions at 160 and 180C, with initial monomer content of 40 wt.%, respectively. (Bottom) Experimental trends of CBC. The empty circles are reproduced from [Peck et al., 2004], based on 140C starved-feed nBA-xylene experiments. Solid circles are experimental data from this study, based on 160180C batch polymerization experiments. .......................................................... 42 Figure 7. Experimental data for TDBC and TSGC. The empty circles are reproduced from [Peck et al., 2004], based on 140C starved-feed nBA-xylene experiments. Empty triangles are experimental data from this study, based on 170C batch polymerization experiments, with initial monomer content of 33 wt.%.................................................................................................................... 45 Figure 8. (Top) Model predictions vs. experimental data for monomer conversion, X, predictions based on oxygen-initiated polymerization (no self-initiation), with rate constants given in Table 4. Solid triangle markers represent experimental

xx data at 160C, with initial monomer content of 40 wt.%. Solid, dashed, dashdotted and dotted lines represent model predictions for oxygen-initiated polymerization with initial oxygen amount of 800, 600, 400 and 200 ppm, at 160C and initial monomer content of 40 wt.%, respectively. (Bottom) Model predictions vs. experimental data for X, predictions based on impurity-initiated polymerization (no self-initiation), with rate constants given in Table 6. Solid triangle markers represent experimental data at 160C , with initial monomer content of 40 wt.%. Solid, dashed, dash-dotted and dotted lines represent model predictions for oxygen-initiated, DTBP-initiated, DCP-initiated and TBPC-initiated polymerization, with initial impurity amount of 800 ppm, at 160C and initial monomer content of 40 wt.%. ............................................................................. 49 Figure 9. Measurements and model predictions of monomer conversion (xm), number-average molecular weight (Mn), number of terminal double bonds per 100 monomer units (TDBH) and number of branching points per 100 monomer units (BPH) when the reactor is operate isothermally. The dashed and solid lines represent model predictions at 160 and 180C respectively, and the triangles and circles represent measurements at 160 and 180C respectively. ....................... 61 Figure 10. Optimization validation results for 300-minute optimal control recipe. Triangles and lines represent measurements and model predictions, respectively. ........................................................................................................................... 68 Figure 11. Corresponding polymer microstructure profiles for the 300-minute optimal control recipe. Triangles and lines represent measurements and model predictions, respectively. .................................................................................... 69 Figure 12. Profiles of the reactor temperature and the flow rates of the monomer (n-butyl acrylate), the solvent (xylene) and the initiator (t-butyl peroxy acetate) solution feed streams. ........................................................................................ 74

xxi Figure 13. Molecular structure of the synthesized poly(n-butyl acrylate). Molecular structure shows the key atoms used to infer polymer microstructure concentrations. ................................................................................................... 77 Figure 14. Measurements of polydispersity index at different temperatures (constant initial monomer weight percent of 40%) and different initial monomer weight percents (constant temperature of 160C). ............................................. 82 Figure 15. Flow of information between the polymerization reactor and the nonlinear reduced-order multi-rate state observer.............................................. 86 Figure 16. Estimated vs. measured values of the number- and weight-average molecular weights (dashed line = estimate without using feed back measurement, solid line = estimate using feed back measurement, circles = measurement). ................................................................................................... 87 Figure 17. Estimated vs. measured values of the concentrations of the solvent and terminal double bond (dashed line = estimate without using feed back measurement, solid line = estimate using feed back measurement, circles = measurement). Bottom figure shows evolution of the 1H-NMR peaks. ............... 88 Figure 18. Estimated vs. measured values of the concentrations of the short chain branches (dashed line = estimate without using feed back measurement, solid line = estimate using feed back measurement, circles = measurement). Bottom figure shows evolution of the 1H-NMR peaks. ........................................ 89 Figure 19. Different sizes of substituent groups in (I) n-butyl acrylate, (II) 2ethylhexyl acrylate, and (III) n-octyl acrylate..................................................... 112

INTRODUCTION AND BACKGROUND

It is estimated that polymer-based materials (e.g. plastics, synthetic fibers, synthetic rubber, paints and coatings) made via free-radical polymerization comprise over half of the total production in the United States [Cunningham et al., 2002]. Particularly in the organic coatings industry, acrylic- and methacrylicbased resins are conventionally produced through free-radical solution polymerization. These solvent-borne resins are commonly used as the primary binder in coatings formulation, due to their photostability and resistance to hydrolysis [Wicks et al., 1999]. However, as environmental restriction on volatile organic content (VOC) in coatings increases, paint and coatings manufacturers are trying to replace the production of high-molecular-weight low-percent-solids acrylic resins with low-molecular-weight high-percent-solids acrylic resins.

Recently, the utilization of high-temperature free-radical solution polymerization, in the absence of chain transfer agents, has been recognized as an attractive method for the production low molecular weight, highly functionalized acrylicbased resins [Grady et al., 2002]. However, as a consequence of the high temperature requirement, reactions that were deemed insignificant at low temperatures now have magnified effects at elevated temperatures, causing the

2 inapplicability of simple kinetic schemes previously used for predicting polymer qualities in low-temperature polymerization processes. A fundamental understanding of how the various side reactions affect the resulting polymer properties is therefore essential for the successful implementation of hightemperature polymerization processes.

Several high-temperature polymerization processes have received considerable attention, namely, high-temperature polymerization of styrene and ethylene. Styrene is known for its ability to polymerize spontaneously at elevated temperatures, while ethylene is known for forming branched polymer chains at higher temperatures. Similar, but not all of these, reactions have been observed in acrylic systems, for example, in high-temperature polymerization of methyl methacrylate (MMA). However, our knowledge on high-temperature acrylic polymerization is, in general, still very limited. Recently, chain-transfer reactions leading to polymer chain-branching and spontaneous thermal initiation in hightemperature solution homo-polymerization of n-butyl acrylates (nBA), and depropagation reaction in high-temperature solution homo-polymerization of nButyl Methacrylate (nBMA) were, for the first time, carefully investigated [Grady et al., 2002; Ahmad et al., 1998; Quan et al., 2006]. These reactions are of interest, since the depropagation and scission reactions are naturally occurring reactions that can be utilized as molecular weight regulators [Hirano et al., 2003]. Thermal initiation by monomer allows for the generation of radicals in the absence of initiators. The absence of initiators, combined with lower solvent

3 requirements at high reaction temperatures, leads to a decrease of the cost of the coatings formulation and elimination of unwanted moieties in the final product. A detailed knowledge on such side reactions could therefore improve control of molecular weight and molecular structure.

1.1

SCOPE OF RECENT STUDIES IN MECHANISTIC MODELING OF SOLUTION AND BULK FREE-RADICAL HOMO-POLYMERIZATION

The ultimate aim of mechanistic modeling studies is to identify the underlying mechanistic pathways in a given reaction system. In the process of identification one needs to conform to the chemistry of free-radical polymerization and the dynamics of the process measurements. For free-radical homo-polymerization systems in general, different monomer systems often have similar underlying reaction networks. A thorough examination of the different reaction sets or networks reported in the literature for free-radical polymerization of different monomer systems is given in the following table.

Table 1. Different sets of reaction mechanisms used in free-radical co- and homo-polymerization kinetic studies. IM = initiation by
monomer, IO = initiation by peroxides, oxygen, IC = initiation by chain transfer agents, P = propagation, CTM = chain-transfer to monomer, CTA = chain-transfer to modifier, CTP = chain-transfer to polymer, CTS = chain-transfer to solvent, CTC = chain-transfer to Macromonomer, TC = termination by combination, TD = termination by disproportionation, D = depropagation, BS = beta-scission, BB = backbiting, TDB = propagation on a terminal double bond, S = Styrene, E = Ethylene.

Reference Kotoulas et al., 2003 Campbell et al., 2003 Villermaux et al., 1984 Qin et al., 2002 Hutchinson et al., 2001 Pladis et al., 1998 Becker et al., 1998 Villermaux et al., 1984 Fenouillot et al., 2001 Fenouillot et al., 1998 Hoppe et al., 1998 Nagasubramanian et al., 1970 Villermaux et al., 1984

System S S S E E E E E MMA MMA MMA MAce MAce

IM X X X

IO

IC

P X X X

CTM X

CTA

CTP X

CTS

CTC X

TC X

TD

BS X

BB X X X X X

TDB X

X X X X X X X X X X X X X X X X X X X X X X X X

X X X X X X X

X X X X X X X X X X X X X X X X X X X

X X

X X X X X

X X X X

X X X

X X X X X

1.2

KINETICS OF FREE-RADICAL SOLUTION HOMO-POLYMERIZATION

One of the simplest kinetic schemes in free-radical solution homo-polymerization is a chain polymerization system which consists of initiation, propagation and termination reactions. In deriving the rate of polymerization assumptions such as the long chain hypothesis (LCH), quasi-steady-state assumption (QSSA), and irreversible reactions are often invoked. Consequently, mathematical derivation of the polymerization rate is greatly simplified by making these assumptions. Through the long chain assumption, we are assuming that the reactivity of all propagating polymer chains is only dependent on the end-group monomer. And through the irreversible reaction assumption we are in assuming that depolymerization reactions are insignificant. The QSSA further simplifies the derivation by assuming a relatively long reaction time period, thus neglecting the early dynamics of the radical concentration.

The inadequacy of simple kinetic schemes in describing high-temperature polymerization has prompted the investigation of other significant reactions occurring at higher temperatures. Deviations have been attributed to the occurrences of chain-transfer reactions, which lead to, among others, backbiting and scission reactions, thus ultimately forming branched polymer chains. Furthermore, the classical polymerization kinetics does not take into account the

6 presence of depropagation and spontaneous thermal initiation reactions, which will have significant effects at elevated reaction temperatures. Recent studies on branching, scission, depropagation and spontaneous thermal initiation reactions in several free-radical polymerization systems are reviewed in the following subsections.

1.2.1 High-Temperature free-radical solution homo-polymerization with depropagation involving styrenic/acrylic monomers

The depropagation effect, or in other words, the occurrence of the depolymerization reaction, is more conveniently defined through the concept of the ceiling temperature (Tc) [Sawada, 1976]. By definition, at Tc, the rate of polymerization is exactly equal to the rate of de-polymerization. The Tc is defined for systems with an initial monomer concentration of 1 mol/L. Alternatively, analogous to Tc, the onset of depropagation can also be defined by the equilibrium (i.e. limiting) monomer conversion, with the system held at a constant reaction temperature.

Depropagation in homo-polymerization has been observed and studied recently, in several high temperature polymerization systems, namely, in the solution polymerization of nBMA at ca. 140 C [Grady et al., 2002], in the bulk polymerization of Methyl Acrylate Trimer (MAT) at ca. 130 C [Hirano et al., 2003], and in the bulk and solution polymerization of Methyl Methacrylate (MMA)

7 at 130-165 C [Peck et al., 2002].

In the high temperature polymerization of MAT, the depropagation effect was attributed to the acceleration of the fragmentation reaction at higher temperatures [Hirano et al., 2003]. Alternatively, the depropagation effect was attributed to the limiting monomer conversion, apparent in the conversion profile in the high temperature polymerization of MMA [Hoppe et al., 1998]. In the high temperature polymerization of nBMA, simulation results generated by Predici showed that both the polymer concentration and the temperature have significant effects on the rate of depropagation [Grady et al., 2002].

1.2.2 High-Temperature free-radical solution homo-polymerization with branching and/or scission involving styrenic/acrylic monomers

At high reaction temperatures, chain-transfer reactions become significant. Chain-transfer to polymer, or inter-molecular chain-transfer, leads to the existence of active mid-chain propagating radical center (structure A), which upon further propagation would create long-chain branching points (structure C), or, upon scission reactions would create terminal double bonds (structure B).

Similarly, at high reaction temperatures, intra-molecular chain-transfer, i.e. backbiting, becomes significant. Backbiting reactions (structure D) lead to the formation of near-chain-end propagating radical (structure E), which could propagate further with other monomer to form short-chain branching points (structure F) or undergo scission reactions to form terminal double bonds (structure G). A classical example of branching and scission reactions is that observed in the high-temperature high-pressure polymerization of poly-ethylene.

Polymer chains with terminal double bonds created by both inter- and intramolecular chain-transfer (structure B and structure G) could undergo further propagation with a radical center, which could lead, again, to the creation of terminal double bonds, or long branching points.

Through the identification of these molecular structures, i.e. terminal double bonds (TDB), short-chain branching points (SCB) and long-chain branching points (LCB), the occurrence of various chain fragmentation reactions were recently investigated in several acrylic polymerization systems, namely, solution polymerization of nBA at ca. 140 C [Grady et al., 2002; Quan et al., 2005; Peck et al., 2002], bulk polymerization of MAT at ca. 70-130 C [Hirano et al., 2003], solution polymerization of Di-n-Butyl Itaconate (DBI) at 60-100 C [Hirano et al., 2002] and bulk polymerization of Styrene (ST) at ca. 250-350 C [Campbell et al., 2001; Campbell et al., 2003; Campbell et al., 2003].

Branching and scission reactions were characterized in the starved-feed semibatch solution polymerization of n-Butyl Acrylate (nBA) at ca. 140 C, through the branching frequency terminal double bond characterization [Grady et al., 2002; Quan et al., 2005; Peck et al., 2002]. Fragmentation reactions occurring in the bulk polymerization of methyl acrylate trimer (MAT) at 70-130 C was observed, and was characterized by using 1H- and 13C-NMR spectroscopy techniques, resulting in the formation of a propagating polymer chain with an unsaturated end group (TDB) [Hirano et al., 2003]. In the solution polymerization of Di-n-Butyl

10 Itaconate (DBI) at 60-100 C, unexpected C resonances were observed in the

13

C-NMR spectra of the carbonyl groups at higher reaction temperatures. Further

investigation reveals the dependence of the intensity of the C resonance on the monomer feed level, suggesting the occurrence of intra-molecular chain transfer reaction [Hirano et al., 2004]. Analogous to that observed in BA polymerization, backbiting, followed by scission reactions were also observed in the polymerization of Styrene (ST) at elevated temperatures, ca. 250-350 C, carried out in a continuous stirred tank reactor, in the absence of initiators. Through the identification of oligomers by the use of the 13C-NMR spectra, 1:3 and 1:5 backbiting/scission reactions were characterized and the mid-chain branching was observed to be non-existent, thus yielding only linear polymer chains with 0,1 or 2 TDB. In addition, the measurement of the TDB distribution by the use of the MALDI-TOF MS indicates the low level of chain transfer to polymer, relative to backbiting, yielding only chains with 1 or 2 TDB, showing the dominance of the -scission reactions in the production of polymer chains [Campbell et al., 2001; Campbell et al., 2003; Campbell et al., 2003].

1.2.3 Spontaneous thermal initiation in high-temperature free-radical polymerization involving styrenic/acrylic monomers

Spontaneous thermal initiation, or thermal initiation by monomer, is best described by the ability to polymerize spontaneously in the absence of other

11 initiating species or in the absence of irradiation [North, 1966]. As have been long observed and studied in the polymerization of ST [Odian et al., 1990; BarnerKowollik et al., 2002; Campbell et al., 2001; Campbell et al., 2003; Campbell et al., 2003], spontaneous thermal initiation does occur at appreciable rates at higher reaction temperatures. Self thermal initiation has also been observed in the high-temperature polymerization of MMA [ Odian et al., 1990; BarnerKowollik et al., 2002; Hoppe et al., 1998; Fenouillot et al., 1998; Fenouillot et al., 2001]. However, the exact nature of the initiation mechanism of MMA is still not well understood. Furthermore, due to the difficulty in obtaining reproducible data, the spontaneous polymerization observed in MMA was attributed to the presence of adventitious peroxides, whose existence is difficult to exclude by the usual purification techniques [Lehrle et al., 1988].

The thermal initiation kinetics of the high-temperature polymerization of ST is well established, and is characterized by a third-order monomer dependence on the rate of initiation. ST monomers form a Diels-Alder adduct, which can undergo molecule induced homolysis to produce two radicals, each capable of initiating polymerization. Recently, spontaneous thermal initiation, with a significant rate of initiation, was observed in the high-temperature polymerization of nBA [Quan et al., 2005].

12

1.3

MODELING METHODS

The purpose of constructing a detailed kinetic model is to be able to correlate the reaction conditions (e.g. temperature, initiator concentration, feed time) with the polymer quality (e.g. MWD). The multi-scale modeling aspect of polymerization reactors has been discussed in [Ray, 1986].

Depropagation and branching are commonly observed phenomena, and their existence in free-radical solution polymerization have been discussed in previous sections. Gel effect, on the other hand, takes place only when viscosity effects become significant (e.g. >50 wt% in monomer in MMA polymerization [Odian et al., 1990]). The meso-scale phenomena are only of importance in heterogeneous (e.g. emulsion) polymerization. Thus, in a homogeneous (i.e. single phase) polymerization they are usually not considered. The macro-scale phenomena are readily modeled through the application of mass and energy balance to the polymerization reactor content.

From the species balance equations, one could solve or integrate the differential equations directly to obtain concentrations of polymer chains with different length

13 as a function of time. A more common approach is by integrating a reduced set of balance equations, which are reduced by defining important moments of the MWD [Bamford et al., 1953]. Another popular approach, which does not involve solving differential balance equations is through the use of statistical methods, which view chain growth as a stochastic process having possible states resulting from the kinetic mechanism and state transition probabilities dependent on the kinetic parameters [Ray, 1972]. In any case, all methods depend on an accurate knowledge of the set of kinetic mechanism. Methods that have found wide acceptance and those that were recently proposed, are reviewed in the following paragraphs.

Instantaneous Property Method (IPM) [Bamford et al., 1958]. The Instantaneous property method employs the long chain hypothesis (LCH) and the quasi steady-state assumption (QSSA) in deriving analytical expressions for the rate of polymerization and molecular weight distribution (MWD). However, the IPM is strictly valid for linear co- and homo-polymers. Hence, scission reactions such as that observed in the high-temperature polymerization of nBA [Quan et al., 2005] cannot be accounted for by the IPM.

Method of Moments (MM) [Bamford et al., 1953; Ray 1972; Bamford et al., 1954]. The method of moments transforms the original high-dimensional systems of differential equations into a low-order system of equations by introducing the leading moments of the distributions of interest. The major limitation of models

14 based on the method of moments is that they only track average quantities. While adequate for most situations, the MM cannot examine, for example, the combined effects of chain-scission and long-chain branching on polymer architecture, or to incorporate chain-lengthdependent termination kinetics into the kinetic scheme [Konstadinidis 1992; Achilias et al., 1992].

Property Moments Method (PMM) [Villermaux et al., 1983, 1984; Blavier et al., 1984; Lorenzini et al., 1992, 1992]. The property moments method is also known as the tendency modeling method. The PMM does not consider balances on the individual species present in the reaction mixture. Instead, mass balance equations are derived for a number of selected quantities that characterize the polymer quality, namely, the average concentration of live radicals, the leading moments of the polymer number chain-length distribution (NCLD), and the concentration of structural characters such as short-chain branching (SCB) and long-chain branching (LCB) and terminal double bonds (TDB) irrespective of the macromolecules carrying them. The PMM, originally developed for modeling free-radical homo-polymerization, has also been extended to the description of free-radical co-polymerization reactions [Konstadinidis 1992; Achilias et al., 1992].

Discrete Galerkin method [Wulkow, 1996]. The Discrete Galerkin method is

more commonly used through the software package Predici . Predici allows one

to analyze polymer reactions with an arbitrary number of species and chain-

15 length distributions and arbitrary number of chain-length dependent reaction steps, with no restrictions on the form of MWD and no necessity of model reduction (e.g. by invoking QSSA). The package has also recently been extended to follow branch-point concentrations as a function of chain length [Iedema et al., 2000].

Three Stage Polymerization Model [Qin et al., 2000, 2002, 2002, 2002, 2003]. Recently, an empirical kinetic model called the Three Stage Polymerization Model (TSPM) was proposed. This particular model was proposed in order to provide a general framework for modeling the bulk free radical polymerization. The name three stage polymerization represents the three prominent stages observed in bulk free polymerization; the low conversion stage, the gel effect stage and the glass effect stage. The TSPM satisfactorily depicts the course of polymerization for the bulk thermal polymerization of ST at 100-200 C, and as expected for polymerization above the Tg of the polymer, only the low conversion and the gel effect stages were observed. An extension of TSPM to copolymerization systems was proven to be effective for modeling the copolymerization ST/MMA, MMA/Ethylene Glycol Di-methacrylate (EGDMA), and AN/AMS, even up to elevated reaction temperatures (ca. 60-120 C).

Numerical Fractionation (NF) method [Teymour et al., 1994; Papavasiliou et al., 2002]. The NF method was specifically developed to address the problem of gel formation in polymer systems. The method is based on the definition of

16 different generations of branched polymers, where each generation differs in the number of branches per polymer chain. The population of chains in each generation is then represented through the method of moments. This model has also been extended further in order to calculate LCB number as a function of chain length.

Monte Carlo techniques [Tobita, 1993, 2001, 2003, 2003]. Monte Carlo techniques have been repeatedly used in modeling MWD development in nonlinear free radical polymerization. Statistical simulation studies were focused on the study of the effects of branching, scission, and crosslinking reactions on MWD profile.

17

MECHANISTIC MODELING

High-temperature (ca. above 100C) solution polymerization processes have been considered as a viable alternative to produce low solvent-level, high solidslevel, environmentally-friendly, acrylic resins [Grady et al., 2002]. Market competition in industrial resin production has motivated optimal operation of current commercial processes. The optimal operation requires understanding of how common process variables such as monomer content and reaction temperature influence the final product quality indices.

Recent high-temperature n-butyl acrylate (nBA) solution homo-polymerization studies have been motivated by the inability of 'classical' chain-polymerization kinetic models (e.g., based on initiation-propagation-termination) in describing nBA homo-polymerization at elevated temperatures. This inability has been attributed to significant effects from secondary reactions, such as inter- or intramolecular chain-transfer reactions, which can lead to branching or scission reactions [Peck et al., 2004]. Hence recent studies on nBA homo-polymerization have focused on (i) the identification of dominant polymerization reactions through spectroscopic methods (i.e., polymer characterization studies) [Ahmad et al., 1998; Chiefari et al., 1999; McCord et al., 1997; Peck et al., 2002], (ii)

18 calculation of reaction rate orders [Fernandez-Garcia et al., 2000], (iii) individual estimation of reaction rate constants through pulsed-laser initiated polymerization (PLP) experiments [Asua et al., 2004; Nikitin et al., 2002; Buback et al., 2004; Beuermann et al., 2002], and (iv) simultaneous estimation of a set of reaction rate constants based on advanced (e.g., spectroscopic measurement of the chain branch concentration) and conventional (e.g., gravimetric measurement of polymeric solids content and chromatographic measurement of molecular weight distribution) measurements [Peck et al., 2004; Busch et al., 2004].

At first glance, the approach of simultaneously estimating a set of parameters against a set of measurements is the most practical and direct approach toward obtaining a process model. Such approach, however, is often hindered by; (i) non-convexity of the least-squared-error objective function and ill-conditioning of the minimization problem (i.e., more than one set of rate constants can give comparable model predictions), (ii) over- or under-postulation of polymerization reaction networks (i.e., knowledge of dominant reaction pathways is still evolving), (iii) limited availability in number and type of measurements. Consequently, recent research directions in this area have included (i) identification of novel 'experimental sensors' (measurements) to be included in the parameter estimation [Busch et al., 2004], (ii) identification of optimal parameter-measurement pairs through sensitivity analyses in order to reduce the complexity of the parameter estimation problem [Polic et al., 2004; Li et al., 2004] and (iii) use of global optimization approaches to obtain a unique set of rate

19 constants (e.g., implementation of the simulated annealing algorithm in modeling package PREDICI [Wulkow et al., 1996]).

Hence an alternative to reducing complexities in the model building and the parameter estimation problem is the inclusion of spectroscopic measurements. The inclusion should limit the number of different combinations of rate constants that can give (i) comparable model predictions and (ii) satisfactory predictions for all types of measurements. The availability of these measurements depends on the observability of signature peaks that can be used as a measure of a certain chain microstructure quantity. Particularly in polymer characterization problems, of critical importance is the observance of peaks that are distinguishable from others arising from polymeric backbone atoms. The complexity of the simultaneous parameter estimation problem is also reduced by the more reliable knowledge of the dominant reaction rate constants (e.g., linear propagation and termination) obtained from PLP experiments.

This study is motivated by the observation of near-complete conversion polymerization experiments that were carried out in the absence of thermal initiators [Grady et al., 2005]. Spectroscopic evidence (ESI-FTMS) of the absence of initiator moieties in the polymer chains that are produced from the spontaneous nBA polymerization experiments [Quan et al., 2005] further supports this observation. In addition to the spontaneous polymerization observed at high temperatures (ca. above 140C), secondary reactions occurring

20 at these temperatures are also studied. The aims of this study are (i) to gain insight into initiation mechanism and to probe the significance of the initiation reaction, (ii) to quantitatively explain the effects of process variables on polymer quality indices (conversion, molecular weight, chain microstructures). The understanding of the initiation and secondary mechanisms can potentially lead to initiator-free commercial polymerization processes and can help process engineers in the safety assessment of high-temperature polymerization processes.

Section 2 describes the experimental and analytical procedures, and section 3 describes the postulation of the polymerization reaction network. The parameter estimation approach used in this study is introduced in section 4. Discussions and analyses of results are presented in section 5, followed by concluding remarks in section 6. Derivation of the process model is presented in the Appendix A.

2.1

EXPERIMENTAL AND ANALYTICAL PROCEDURES

n-Butyl acrylate (BASF, 99.5%, with boiling point of 146.8C) was passed through an inhibitor removal column (DHR-4, Scientific Polymer Products, Inc.) prior to use. Xylene (ExxonMobil Chemical Co., an 80/20 mix of xylene isomers and ethyl benzene with a boiling point range of 137 - 143C) and 4-

21 methoxyphenol (99%, ACROS, with boiling point of 243C) were used as received. Polymerizations were carried out in a 1-liter glass RC1 calorimeterreactor (Mettler-Toledo GmbH, Schwerzenbach, Switzerland).

Each experiment started with loading the xylene solvent into the reactor. Nitrogen gas was subsequently purged through the solvent and reactor for 15 to 30 minutes. The monomer was then pumped into the reactor and heated under nitrogen blanket to the desired reaction temperature. Reaction temperature was maintained throughout the course of the polymerization by adjustment of the reactor jacket temperature. Samples drawn from the reaction mass at specified time intervals were diluted (50/50 v/v) in a cold inhibitor solution (1000 ppm 4methoxy phenol in xylene) to prevent further reaction from taking place.

Polymer and residual monomer concentrations were determined by gravimetric and chromatographic (GC) analyses, respectively. GC analyses were carried out by using Hewlett Packard (HP) 6890 GC-FID system with a 30 m DB-5 separation column. The FID (flame ionization detector) temperature was set to 250C, with oven temperature ramp at 10C/min. Measurement of residual monomer by weight was obtained from the integration of monomer peak area, adjusted with a dilution factor when necessary.

Molecular weight distributions were determined by using an HP 1090 HPLC (high performance liquid chromatography) system, equipped with an HP 1047A RI

22 (refractive index) detector and a four-column set configuration (105, 104, 103, 102 30 cm x 7.8mm id microstyragel columns). Tetrahydrofuran was used as the mobile phase at flow rate of 1mL/min at 40C. Data from the RI detector was collected and processed by using a Waters Millennium system. Narrow molecular weight polystyrene standards were used to calibrate the column set.

Polymeric solid samples for NMR analyses were isolated through high vacuum and evaporation at 65C. Quantitative carbon and proton NMR spectra were recorded at 34C on a Bruker DRX-400 MHz spectrometer (1H frequency of 400.13 MHz and 13C frequency of 100.62 MHz). Deuterated chloroform (CDCl3) was used as solvent in all analyses. 13C-NMR spectra were obtained by using a 10 mm broadband probe equipped with 1H decoupler, acquisition time of 1.36 s, spectral width of 24 kHz, relaxation time of 15 s, number of scans between 1600 and 3200, and a 90 pulse width of 13.9 s.1H-NMR spectra were obtained by using a 4 s acquisition time, a 10 kHz spectral width, an 8 s relaxation time, number of scans of 16 and an 11.5 s 90 pulse width. Peak assignments obtained through analyses of DEPT spectra reported by [Quan et al., 2005] is used in this study.

2.2

POSTULATION OF REACTION NETWORK

23 Chain-initiation. Studies in spontaneous polymerization of methyl mecthacrylate (MMA) have cited both self-initiation [Stickler et al., 1978; Brandolin et al., 1980; Lingnau et al., 1980; Stickler et al., 1981; Lingnau et al., 1983; Lingnau et al., 1984; Lingnau et al., 1984] and decomposition of impurities, i.e. oxygen or peroxides in general [Clouet et al., 1993; Lehrle et al., 1988], as the underlying mechanism for the generation of initiating radicals. In a recent study on thermally-initiated MMA polymerization, it is reported that MMA could also undergo reaction with air to form macromolecular peroxides [Nising et al., 2005]. Both peroxide decomposition [McManus et al., 2002] and self-initiation [Cao et al., 2004] have been assumed to be responsible for the generation of initiating radicals in nBA polymerization. In this study, mechanisms of thermal self initiation proposed for methyl methacrylate [Lingnau et al., 1983] and the decomposition of peroxide [Cerinski et al., 2002] are adopted for nBA. The initiation step dictates the amount of live radicals (initially) present, and will eventually dictate the amount of dead polymer chains formed. The initiating radicals formed by the peroxide decomposition is represented by live radical chains of length zero (P0). While in the self-initiation reaction, the initiating radicals are assumed to be the monomer itself, thus represented as live (secondary) radical chains of length one (P1).

24 Chain-propagation, chain-transfer to polymer and chain-fragmentation. Intra-molecular chain-transfer to polymer or backbiting has been postulated to dominate at lower monomer content (ca. below 50 wt.%) [Peck et al., 2004], and is known to occur even at lower temperatures [Asua et al., 2004; Willemse et al., 2005]. The backbiting event allows for the creation of short chain-branches, upon propagation of the tertiary (mid-chain) radical near the end of the chain. This tertiary radical can alternatively undergo -scission, to create a dead polymer chain with a terminal double bond (TDB) and a live polymer chain. As only averages are considered in this study, only the scission reactions which create long dead chains instead of dead trimer chains with TDBs are included in the overall kinetic scheme. The propagation, chain transfer to polymer, branching and scission reactions are presented in unison, since these reactions simultaneously affect the average molecular weights, the number of TDBs and the number of branching points. Dead polymer chains are created through the scission reaction, and are governed by the amount of tertiary radicals (Qn) present. The scission reaction leads to the creation of a dead polymer chain (Dn2)

with a TDB in one step. At the same time the tertiary radicals can also form

chain branches through further propagation. SCB denotes a short chain-branch point formed through such a reaction.

Chain-transfer to monomer and solvent. Chain transfer to monomer allows for the creation of a dead chain with a TDB and a live chain. The rate of chain transfer to monomer is influenced by the amount of monomer and live

25 (propagating) chains present at a given time. The concentration of monomer is much larger compared to the concentration of live chains. The significance of chain transfer to monomer, however, is often discounted due to the low reaction rate constant relative to the scission reaction. A recent estimation of the rate constant for chain transfer to monomer reactions in high-pressure hightemperature polymerization of nBA indicated that transfer to monomer might have a more significant contribution towards the overall kinetic scheme [Busch et al., 2004]. In this study, since temperature effects on the number of terminal double bonds are of interest, -scission and chain transfer to monomer reactions are considered in the model development. Along with chain transfer to monomer [Maeder et al., 1998], nBA has also been reported to undergo chain transfer to solvent [Peck et al., 2002; Raghuram et al., 1969]. Upon chain-transfer to monomer, a live secondary radical of length one or the monomer itself (P1TDB) is created. When this radical propagates further, a TDB is created for the resulting chain. The reactive solvent species (i.e., in this study, xylol radical) is represented as a live chain of length zero (P0), and is assumed to propagate at the same rate as a secondary radical.

Chain-termination. While termination for acrylates usually take place almost exclusively by combination, at high temperatures, the extent of termination by disproportionation can increase [Odian et al., 1991; Vana et al., 2002]. Chain termination by a combination of both modes is assumed, and the estimates

26 reported by [Peck et al., 2004] on the relative extent of each termination mode are used in this study to describe the various termination reactions.

The postulated polymerization reaction network is presented in Table 2. Details of the derivation of the batch polymerization process model are presented in the Appendix A.

Table 2. Postulated polymerization reaction network for batch polymerization of nBA in xylene at 140-180C with a monomer content of 20-40 wt.%. The subscripts represent polymer chain length.

k ti 2M 2P1
O2 R1OOR 2 2P0 p Pn +M Pn+1

k bb Pn Q n
pq Q n +M Pn+1 (+SCB)

Q n D n-2 +P2 (+TDB)

k tm Pn +M D n +P1TDB
p P1TDB +M P2 (+TDB)

k ts Pn +S D n +P0 k tc Pn +Pm D n+m k td Pn +Pm D n +D m

tc Pn +Q m k D n+m sec-tert sec-sec sec-sec

k tc Pn +Q m D n +D m k tc Q n +Q m D n+m k tc Q n +Q m D n +D m
tert-tert tert-tert

sec-tert

27

2.3

PARAMETER ESTIMATION

The objective of the parameter estimation is to obtain reliable estimates for spontaneous initiation and secondary reaction rate constants. The rate constants are simultaneously estimated from a set of measurements. Two different sets of measurements at two different temperatures are considered in the parameter estimation.

A simultaneous parameter estimation problem with a sum of squared errors (between model predictions and measurements) performance index is formulated. The simplex search method (fminsearch) of the Optimization Toolbox of the commercial software MATLAB version 6, is used in the parameter estimation. The parameter non-negativity constraint is implicitly imposed in the simplex method by estimating the exponents of the parameters to be estimated

(i.e., estimate , where = 10 , instead of directly estimating , where is the

rate constant actually being estimated. Note that by estimating the exponents instead of the actual parameters, the algorithm converges faster to a minimum. To minimize the probability of obtaining a local minimum, different initial conditions for the parameter estimates were used in the parameter estimation.

28
Rate constants for the primary reactions available in the literature are not reestimated and are used in the model. The minimization problem is of the form:
y ( t ) -y ( t ) i j i j min J= yi ( t j ) i=1 j=1
4 Ni 2

where y1,,y4 represent the measurements of monomer conversion (X), numberaverage molecular weight (Mn), weight-average molecular weight (Mw), and number of (short) chain branches per chain (CBC), respectively. Ni is the number of measurements of the measured variable yi, and the parameters being estimated 1, , 4 are the rate constants corresponding to the second-order self-initiation (kti), propagation of tertiary radicals (kpq), secondary radical backbiting (kbb), and tertiary radical -scission (k). The measurements were obtained at temperatures of 160 and 180C, each at an initial monomer content of 40 wt.%. The parameters kp, ktm, kts, kt were extrapolated from references [Asua et al., 2004; Maeder et al., 1998; Raghuram et al., 1969; Beuermann et al., 2002], respectively. It was assumed that the combination of measurements X, Mn, Mw, CBC, contains adequate information for the estimation of the rate constants kti, kpq, kbb, k. The relationship between the rate constants and the set of measurements can be seen by inspection of the analytical model structure (Appendix A): the measurement of X is directly influenced by kti (mass balance equation on monomer), CBC is directly influenced by kpq (mass balance equation on chain branches), and the average molecular weights are indirectly influenced by all the parameters being estimated. Experimental trends to be captured in the parameter estimation are: increasing monomer conversion, decreasing average

29
molecular weights and CBC, with increasing temperature. Monomer conversion and molecular weight measurements were obtained through gravimetric and chromatographic measurements, respectively.

The measurements of number average of chain branches per chain (CBC) were obtained by using quantitative nBA 13C-NMR peak area at 47-49 ppm as a measure of number of branch points (quaternary carbon atom [Ahmad et al., 1998]) and the peak area at 165-180 ppm (carbon atom on C=O [Quan et al., 2005]) or the peak area at 65 ppm (carbon atom on C-O [Quan et al., 2005]) as a measure of number of repeat units. One way to report number of branches is to use the ratio of number of branches to number of repeat units. When this ratio is multiplied by 100, the number of branches is expressed as (average) number of branch points per 100 repeat units (CBH). The effect of temperature on chain branches was most evident as the quantity CBC is defined as:

CBC =

SCBH DPn 100

where DPn is the number-average degree of polymerization. The quality of fit at the two different temperatures is presented in Figure 1. Estimates at each temperature, and the corresponding estimates of the Arrhenius parameters are presented in Table 2. To gauge the performance of the simplex search algorithm, the surface of the objective function is presented as functions of kti and kbb, at 160C and initial monomer content of 40wt.% in Figure 2. The parameters kti and kbb represent the two underlying pathways that directly influence all of the measurements (terms in the rate laws that include kpq and k depend on amount

30
of tertiary radicals, which is directly governed by the rate of backbiting). Assuming that the intervals considered for kti and kbb are sufficiently large, it is shown that the global minimum was obtained by using the optimization approach (kti = 1.25 x 10-6 L mol-1 s-1, kbb = 1.92 x 104 s-1).

0.8

2
0.6

0.4

CBC
1 0
20 15

X
0.2 0
10 8

Mn x 103

Mw x 10

10

0 0

20

40

60

80

100

0 0

20

40

60

80

100

Time, min

Time, min

Figure 1. Quality of fit for parameter estimation based on measurements of X, Mn, Mw and CBC. Solid circle and square markers represent experimental data at 160 and 180C, with initial monomer content of 40 wt.%, respectively. Solid and dashed lines represent model predictions at 160 and 180C, with initial monomer content of 40 wt.%, respectively.

31

Table 3. Estimated rate constant values based on measurements of X, Mn, Mw and CBC at 160 and 180C, at an initial monomer content of 40wt.%, and the corresponding Arrhenius parameters. Rate Constant kti (L.mol-1s-1) kpq (L.mol-1s-1) kbb (s-1) k (s-1) 160C 1.25E-06 5.16E+01 1.95E+04 7.40E-01 180C 3.05E-06 5.20E+01 2.42E+04 4.42E+00 A 8.27E+02 5.99E+01 3.87E+06 2.88E+17 E (J.mol-1) 73,152 534 19,114 145,853

32

Figure 2. Surface of the objective function (sum of squared error between model predictions and measurements of monomer conversion, number-average molecular weight, weight-average molecular weight and number-average chain branches per chain) of the parameter estimation as functions of kti and kbb at 160C, initial monomer content of 40wt.%.

33

2.4

DISCUSSIONS AND ANALYSES

Parameter estimation results and model validation. From Figure 1, it is shown that a satisfactory fit was obtained. Based on the parameter estimates at the two different temperatures, Arrhenius parameters were estimated and are presented in Table 2. Calculated secondary reaction parameter values (based on the estimated Arrhenius parameters) were then compared with literature values at temperatures 140 and 170C. The comparison is presented in Table 3. Note that the nominal values of kpq and kbb are in agreement with reported values. Several different values of k have been reported in the literature, particularly the one reported by [Busch et al., 2004]. It should be noted that the simultaneous parameter estimation study in [Busch et al., 2004] considers different sets of measurements, parameters and reaction paths (e.g., formation of long chain branches preceded by inter-molecular chain transfer is considered). As will be discussed further in the following subsection, the observed trend of chain branches vs. number average degree of polymerization suggests the formation of short chain branches, as opposed to long ones. The estimate for kti is the first report on the postulated second-order self-initiation reaction. The apparently high energy activation value for k is currently understood as a high temperature

34
phenomenon, as scission is not observed to be prevalent at lower reaction temperatures.

Table 4. Reported and estimated rate constant values for high-temperature nBA polymerization. aThe self initiation rate constant was estimated while assuming that initiation by impurities did not occur polymer was not considered in this study
b

Inter-molecular chain transfer to

Rate Constant kti (L.mol-1.s-1) kp (L.mol-1.s-1) kpq (L.mol .s ) ktm (L.mol .s ) kts (L.mol .s ) kbb (s )
-1 -1 -1 -1 -1 -1 -1

A 8.27E+02 2.21E+07 -kpx1.20E-03 5.99E+01 kpx1.60E-02 -kpx1.95E+01 3.50E+07 --3.87E+06

E (J.mol-1) 73,152 17,900 --534 15,200 -32,175 29,300 --19,114 ---145,853 5,590 --

140C 4.66E-07 1.21E+05 -1.45E+02 5.13E+01 2.31E+01 -2.01E+02 6.91E+03 -4.00E+03 1.48E+04 -6.00E+00 -1.04E-01 5.05E+07 --

170C 1.97E-06 1.91E+05 1.67E+05 -5.18E+01 4.43E+01 1.81E+05 5.39E+02 1.23E+04 2.13E+04 -2.16E+04 5.00E+03 -6.28E+04 1.85E+00 5.64E+07 1.10E+07

Ref. This Studya Asua et al., 2004 Busch et al., 2004 Peck et al., 2004 This Study Maeder et al., 1998 Busch et al., 2004 Raghuram et al., 1969 Plessis et al., 2003 Busch et al., 2004 Peck et al., 2004 This Study Busch et al., 2004b Peck et al., 2004 Busch et al., 2004 This Study Beuermann et al., 2002 Busch et al., 2004

ktp (L.mol .s ) k (s-1)

-1

-1

---2.88E+17

kt (L.mol .s )

-1

-1

2.57E+08 --

35

To quantify the reliability or variability of the reported estimates, we consider the local sensitivity of the parameters with respect to the measurements (we assume that the parameters are linearly related to the measurements within one order of magnitude of the nominal values). The parameter-measurement sensitivity, with respect to measurements of monomer conversion and weight-average molecular weight, is presented in Table 4. The variability of the estimates can then be calculated as follows:

exp erimental = y

yexp erimental

theoretical

hence a conservative approach towards quantifying the variability of the reported estimates is by using measurement with the largest variability (e.g. monomer conversion, +/- 10% off the average) as basis for the above calculation. By using such approach, the variability with respect to the nominal values at 160C are approximately: kti x 5 kti kti / 3, kpq x 5 kpq kpq / 2, kbb x 3 kbb kbb / 2 and k x 60 k k / 60.

36
Table 5. Local sensitivity of parameter-measurement with respect to nominal parameter and measurement values at 160C, 200 min reaction time and initial monomer content of 40wt.%

Parameter () kti

X10 /10

X (%) +17.75 -26.49 +19.50 -19.72 -34.90 +19.57 +1.48 -0.15

Mw (%) -46.54 +96.48 -123.89 +53.79 -81.95 +129.49 -11.44 +1.40

kpq

X10 /10

kbb

X10 /10

X10 /10

The model was then validated by comparing model predictions with experimental data at extrapolated conditions (at experimental conditions outside the range used in the parameter estimation). In Figure 3, the model predictions are compared with experimental data at 160C, and at initial monomer content of 20wt.% (the model was based on subset of data obtained at 160-180C, initial monomer content of 40wt.%). Based on separate triplicate experiments, the variability of the monomer conversion data can be captured by using constant error bars on the average value. The constant error bars represent +/- 10% off the average for the monomer conversion data and represent +/- 5% off the

37
average for the weight-average molecular weight data. The comparison shows that the model was able to capture the monomer conversion trend quantitatively and the weight-average molecular weight trend qualitatively (slight overprediction). Similarly, the model predictions are compared with experimental data at 140C, at initial monomer content of 40wt.%. The comparisons are presented in Figure 4. The model was over-predicting the extent of monomer conversion at 140C, and slightly under-predicting the weight-average molecular weight profile. To understand the cause of the discrepancies, the model was then validated against molecular structure quantities as function of temperature. This validation is presented in Figure 5. From the figure, it is shown that the model is able to capture the temperature-chain branch trend, however, is not able to capture the temperature-terminal double bond trend. This suggests that the model has not fully captured the interplay between the chain branch and terminal double bond formation. This also explains the high variability of the estimate for k. It should also be mentioned that many efforts have been focused on understanding the effects of chain length on the termination rate coefficient [Buback et al., 2002]. In this study, it is assumed that the rate of chain-termination does not depend on chain length at low initial monomer content (shorter polymer chains throughout reaction time). The rate constant for chain-termination is therefore not reestimated, instead, the chain termination parameters reported in [Peck et al., 2004] is adopted.

38
1

0.8

0.6

X
0.4 0.2 0
2 x 10
4

1.5

Mw

0.5

0 0

50

100

150

200

Time, min

Figure 3. Model predictions vs. experimental data for X and Mw, predictions are based on estimated Arrhenius parameters (Table 1). Solid triangle and circle markers represent experimental data with initial monomer content of 20 and 40 wt.%, at 160C, respectively. Solid and dashed lines represent model predictions at initial monomer content of 20 and 40 wt.%, at 160C, respectively.

39
1

0.8

0.6

X
0.4 0.2 0 4 x 10 3

w
1 0 0

50

100

150

200

Time, min
Figure 4. Model predictions vs. experimental data for X and Mw, predictions are based on estimated Arrhenius parameters (Table 2). Solid triangle and circle markers represent experimental data at 140 and 160C, respectively, with initial monomer content of 40 wt.%. Solid and dashed represent model predictions at 140 and 160C, respectively, with initial monomer content of 40 wt.%

40
5

CBC

0 1

0.8

TDBC

0.6

0.4

0.2

0 120

140

160

180
o

200

Temperature, C

Figure 5. Model predictions vs. experimental data for TDBC and CBC vs. temperature. predictions are based on estimated Arrhenius parameters (Table 2). Solid markers represent experimental data. Experimental data at 160, 170 and 180C have initial monomer contents of 40, 33, and 40 wt.%, respectively (batch), at reaction time of ca. 40-50 minutes. The experimental data at 140C

41
has a final monomer content of ca. 50 wt.% (starved-feed, 90 minute feed time), at reaction time of 90 minutes. The experimental data at 140C is recalculated from [Peck et al., 2004]. Empty markers represent batch model predictions at ca. 60 minute reaction time, and the solid lines are plotted only to connect the model predictions.

Microstructural evolution and dominant modes of chain-initiation and chain-termination. In this study, temperature was found to have the largest effect on microstructures, and this effect was most evident when microstructural quantities are represented as quantities per chain. The linear increase of average number of chain branches per chain as a function of DPn (Figure 6) suggests the formation of short chain branches and confirms the reaction network postulated in [Peck et al., 2004]. This trend suggests a constant 'rate of formation' of chain branches (i.e., if there is a shift from intra- to inter-molecular chain transfer, the 'rate of formation' should not be constant as DPn increases). This also supports studies that attribute the (short) chain branch formation to backbiting. Data from [Peck et al., 2004], recalculated and presented in Figure 6, also support this observation.

42
3

CBC
1 0

CBC

0 0

20

40

60

80

100

DPn
Figure 6. (Top) Model predictions vs. experimental data for CBC, predictions are based on estimated Arrhenius parameters (Table 2). Solid square and circle markers represent experimental data at 160 and 180C , with initial monomer content of 40 wt.%, respectively. Solid and dashed lines represent model predictions at 160 and 180C, with initial monomer content of 40 wt.%,

43
respectively. (Bottom) Experimental trends of CBC. The empty circles are reproduced from [Peck et al., 2004], based on 140C starved-feed nBA-xylene experiments. Solid circles are experimental data from this study, based on 160180C batch polymerization experiments.

Figure 7 presents the microstructural data of number-average terminal double bonds per chain (TDBC) and number-average terminal solvent groups per chain (TSGC). TDBC and TSGC measurements were obtained from quantitative nBA
1

H-NMR analyses. The 1H-NMR peak area at 5.5 and 6.2 ppm (2 hydrogen

atoms on the terminal double bond [Chiefari et al., 1999]) was used as a measure of the concentration of terminal double bonds, and the peak area at 7.1 ppm (4 hydrogen atoms on the terminal solvent group [Quan et al., 2005]) was used as a measure of the concentration of terminal solvent groups. Number of repeat units was measured by using the peak area at 4.1 ppm (2 hydrogen atoms on CH2 next to COO [Quan et al., 2005]). Number of terminal double bonds per 100 repeat unit (TDBH) was calculated by dividing the peak area at either 5.5 or 6.2 ppm by half of the peak area at 4.1 ppm, multiplied by 100; and the number of terminal solvent groups per 100 repeat unit (TSGH) by dividing one-fourth of the peak area at 7.1 ppm by half of the peak area at 4.1 ppm, multiplied by 100. Similarly, the number of terminal double bonds per chain (TDBC) and the number of terminal solvent groups per chain (TSGC) are given by:

44
TDBH DP TDBC = n 100 TSGH DP TSGC = n 100 At the moment there are not enough data to draw conclusions on the evolution of terminal solvent groups per chain, however, the values of TSGC agree with previous studies. FTMS studies by Grady et al. [Grady et al., 2002], Peck and Hutchinson [Peck et al., 2002] and Quan et al. [Quan et al., 2005] reported that at elevated temperatures most of chains are xylol-initiated or -scission initiated/terminated. Experimental trends show that approximately 3 out of 10 chains are xylol-initiated, and 7 out of 10 chains are -scission terminated at 170C and at initial monomer content of 33wt.% (Figure 7). A more thorough study on the evolution of chain microstructures for broader experimental ranges of temperature and monomer content is currently being carried out.

45
1

0.8

TDBC
TSGC

0.6

0.4

0.2

0
1

0.8

0.6

0.4

0.2

0 0

20

40

60

80

100

DPn
Figure 7. Experimental data for TDBC and TSGC. The empty circles are reproduced from [Peck et al., 2004], based on 140C starved-feed nBA-xylene experiments. Empty triangles are experimental data from this study, based on 170C batch polymerization experiments, with initial monomer content of 33 wt.%.

46

Spontaneous initiation mechanism. While this study reports the estimated rate constant for self-initiation, direct evidence for self-initiation is still absent. Thus we also consider the possibility of initiation by decomposition of impurities (e.g. oxygen and peroxides). The Arrhenius expressions reported by [Cerinski et al., 2002] are used to represent rate of decomposition of common peroxides (Table 5). In Figure 8, two cases are considered: (i) in the case that an unknown amount of impurities initiate the polymerization and the decomposition rate is approximately equal to the decomposition rate of molecular oxygen, and (ii) when the amount of impurities is known, however the rate at which they decompose is not known. The first plot (top plot, Figure 8) shows that trace amount of impurities, ca. 800 ppm, is able to initiate and sustain a considerable rate of polymerization, with a slow rate of impurity decomposition. It should be noted, however, that solubility of oxygen in organic medium is well below 800 ppm. Hence the hypothetical scenario of oxygen-initiated polymerization (in conjunction with the reported secondary constants) is deemed unrealistic. The second plot (bottom plot, Figure 8) shows that faster impurity decomposition rates lead to lower final monomer conversion. Thus much higher amounts (>> 800ppm), of impurities with faster decomposition rates, need to be present to initiate and sustain observed monomer conversion profiles.

47
Table 6. Rate constant values for decomposition of oxygen and peroxides [Cerinski et al., 2002], used in simulations presented in Figure 8.

Type of impurity Oxygen (O2) di-t-butyl peroxide (DTBP) di-cumyl peroxide (DCP) t-butyl perisopropyl carbonate (TBPC)

A 6.00E+09

E (J.mol-1) 116,000

Value at 160C (s-1) 6.15E-05

6.64E+15

156,000

1.02E-03

1.12E18

170,000

3.53E-03

1.42E+15

141,000

1.41E-02

From the reported estimates, we are also able to analyze the activation energy of the postulated second-order self-initiation, by comparing the estimated activation energy for n-butyl acrylate with other self-initiating monomers. The comparison is presented in Table 6. Note that the amount of initial monomer required to achieve 30% monomer conversion in ca. 20 minute reaction time at 170C, increases with increasing energy of activation. In other words, the lower the initial monomer content required to achieve a certain degree of monomer conversion (at a given temperature and reaction time) the lower the activation energy associated with the self-initiation reaction. Thus the observed trend of the self-initiation activation energy and the high amount of impurities required to obtain observed conversion profiles suggest the occurrence of self-initiation. However, at the moment there is

48
still no direct evidence for self-initiation. It should also be noted that removal of oxygen/peroxide-type impurities through nitrogen bubbling of solvent prior to monomer feed, might not be adequate [Albertin et al., 2005]. A more direct approach toward identifying the initiating molecules or quantifying amount of impurities will be presented in a future study.

49

0.8

0.6

X
0.4 0.2 0

0.8

0.6

X
0.4 0.2 0 0

40

80

120

160

200

Time, min

Figure 8. (Top) Model predictions vs. experimental data for monomer conversion, X, predictions based on oxygen-initiated polymerization (no selfinitiation), with rate constants given in Table 4. Solid triangle markers represent

50
experimental data at 160C, with initial monomer content of 40 wt.%. Solid, dashed, dash-dotted and dotted lines represent model predictions for oxygeninitiated polymerization with initial oxygen amount of 800, 600, 400 and 200 ppm, at 160C and initial monomer content of 40 wt.%, respectively. (Bottom) Model predictions vs. experimental data for X, predictions based on impurity-initiated polymerization (no self-initiation), with rate constants given in Table 6. Solid triangle markers represent experimental data at 160C , with initial monomer content of 40 wt.%. Solid, dashed, dash-dotted and dotted lines represent model predictions for oxygen-initiated, DTBP-initiated, DCP-initiated and TBPC-initiated polymerization, with initial impurity amount of 800 ppm, at 160C and initial monomer content of 40 wt.%.

Table 7. Activation energies of self-initiating monomers. aInitial monomer content at 170C, ca. 20 min reaction time, 30% monomer conversion. bBased on the postulated participation of molecular oxygen in the self-initiation mechanism

Monomer nBA MMA Styrene

E (J.mol-1) 73,152 85,600b 108,500

Monomer contenta (wt.%) 33 50 100

Ref. This study Nising et al., 2005 Hui et al., 1972

51
2.5 CONCLUSIONS

A mechanistic model for high-temperature batch polymerization of nBA was developed and presented. The model was able to capture the process dynamics between 140 and 180C and between 20 and 40 wt.%. Quality of estimated rate constants, along with the predictive capabilities of the developed model, is presented and discussed. New experimental observations that support previous studies on high-temperature solution of alkyl acrylates are also presented. This study sheds light on the nature of the spontaneous initiation, backbiting, scission and branch formation reaction mechanisms. It should also be noted that there are limitations to the extents by which extrapolations can be made based on the reported parameters. Future research directions in this field include; (i) identification of novel 'experimental sensors' or measurements for parameter estimation studies, (ii) design of experiments that provide direct evidence of self initiation (e.g. better impurity removal techniques), (iii) development of efficient parameter estimation approaches and (iv) development of iterative and datadriven approaches towards identification of polymerization reaction networks and the corresponding rate constants.

52

MODEL-BASED OPTIMAL CONTROL OF A SEMI-BATCH HIGHTEMPERATURE POLYMERIZATION REACTOR

Batch reactors are widely used in the commercial production of polymer solutions (resins), especially in low-volume production of high-quality resins [Srinivasan et al., 2003; Grady et al., 2002]. While traditional batch processes are equipped with minimum instrumentation [Srinivasan et al., 2003], efforts have been made to use advanced non-conventional analytical methods such as near-infra-red spectroscopy to measure monomer conversion and polymer average molecular weights in batch solution polymerization of styrene [Fontoura et al., 2003].

Optimal control of batch polymerization reactors has received considerable attention over the past several decades. Choices of monomer(s), solvent(s), initiator(s) and operating conditions dictate the underlying polymerization reaction mechanism and the complexity of the dynamics of the polymerization system (e.g. homo- vs. co-polymerization, solution vs. emulsion polymerization, absence vs. presence of gel effect, etc.). Recent optimal control studies on batch polymerization systems include: solution and emulsion polymerization of styrene [Karagoz et al., 2000; Gentric et al., 1997], solution and bulk polymerization of

53
methyl methacrylate [Zhang, 2004; Garg et al., 1999], and semi-batch emulsion copolymerization of vinyl acetate/butyl acrylate [Immanuel et al., 2002].

Optimal control problems in batch polymerization reactors are inherently multiobjective. Optimizing variables often have competing effects on an objective function. The effect of an optimizing variable on one objective function can be in reverse of the effect of the same optimizing variable on another objective function. While the optimization problem can be formulated and solved in a number of ways, choices are limited by computational expense [Srinivasan et al., 2003]. Constraints of the optimization problem are feasible range of optimizing variables, balance equations, product quality limits, instrument limitations, and safety limits. Advances have been made to improve numerical algorithm (e.g., genetic algorithm [Garg et al., 1999; Immanuel et al., 2002; Tian et al, 2001] and to develop alternative optimization approaches (e.g., extended iterative dynamic programming [Wang et al., 1997] and measurement inclusion [Srinivasan et al., 2002]).

Most optimization studies are based on first-principle models, derived from knowledge of a predominant set of polymerization reaction mechanism (mechanistic models). The need for reliable process models in the presence of process-model mismatch and disturbances has motivated the use of hybrid models such as a combination of neural networks and a first-principles model [Zhang 2004].

54

This paper presents a study on optimal control of high-temperature (ca. above 140C) semi-batch solution polymerization of alkyl acrylates. Challenges in the high-temperature polymerization include the inadequacy of low-temperature mechanistic models for the high temperatures and the limited industrial experience in high-temperature batch operation. The inadequacy of the lowtemperature mechanistic models is a consequence of the profound effect of secondary reactions that are negligible at low temperatures on the reactor dynamics at the high temperatures. Motivated by these, in this work, a mechanistic semi-batch reactor model based on a proposed reaction mechanism for n-butyl acrylate is derived. The model parameters (reaction rate constants) are estimated from off-line measurements of conversion, average molecular weight, and number of terminal double bonds and branching points per 100 monomer units. The model is validated against measurements made when the reactor is operated in regions different from those in which parameter estimation was based on. By using the model, optimal solvent, monomer solution, and initiator solution feed and reactor temperature profiles that minimize a multiobjective performance index, are calculated. The reactor is operated in real-time according to the optimal recipe and measurements are made to validate the optimality of the recipe.

55
3.1 EXPERIMENTAL FRAMEWORK

The polymerization system being considered is the thermal solution polymerization of n-butyl acrylate in xylene. Polymerizations were carried out at 160 and 180C, low monomer concentration (20-40 monomer wt.%), in the absence of thermal initiators, in a 2-Liter calorimeter. Details of the experimental setup and spectroscopic analyses can be found in [Peck et al., 2004]. Experiments were carried out at DuPont Marshall Laboratory, Philadelphia, PA.

3.2

MODEL DEVELOPMENT

3.2.1 Kinetic Scheme and Rate Laws

Reaction mechanism postulated for the polymerization of n-Butyl Acrylate is based on the characterization results on different polymer micro-structures. The approach introduced in tendency modeling method [Quan et al., 2005] is used in this work. The approach assumes that rate of reactions are only dependent upon concentrations of micro-structural entities that are involved in the reaction. Reactivity of propagating radicals is also assumed to be independent of chain length.

56
Thermal self-initiation and propagation. n-Butyl Acrylate has been observed to undergo polymerization in the absence of thermal initiators. The self-initiation reaction is currently assumed to be second-order with respect to monomer concentration [Rantow et al., 2006]:
k ii I 2R1 k ti 2M 2R1
p R 1 +M R 1

Initiation by initiator Initiation by monomer Propagation

Inter- and Intra-molecular chain transfer. Backbiting and chain-transfer to polymer (CTP) have been observed to be dominant at low and high initial monomer concentrations, respectively [Rantow et al., 2006].
k bb R1 R 3
tp R1 +D+TC R 2 +D

Backbiting CTP

Chain-transfer to small molecules. Assuming that amount of impurities in solution polymerization is kept to a minimum, the only chain-transfer reactions to be considered are chain-transfer to monomer (CTM) and solvent (CTS). A terminal double bond (TDB) is produced every time chain-transfer to monomer occurs.
k ts R+S R1 +D

CTS CTM

R+M R 4 +D

k tm

Long and short chain-branching formation. Formation of long chain-branching (LCB) and short chain-branching (SCB) have been observed in solution polymerization of n-Butyl Acrylate at 140C [Grady et al., 2002].

57
pb R 2 +M R1 +LCB pb R 3 +M R1 +SCB

k k

LCB Formation SCB Formation

Terminal double bond formation. Scission reactions are in competition with the branching-formation reactions [Grady et al., 2002]. Notice that TDB is also formed in CTD.
p R 4 +M R1 +TDB R 2 R1 +D+TDB R 3 R1 +D+TDB

TDB Formation Beta-scission Beta-scission

TDB propagation. Existence of TDBs allows further polymerization to take place. TDB propagation is in competition with the scission reactions [Grady et al., 2002].
k tdb R1 +TDB+D R 2

TDB Propagation

Chain termination. Both known modes of chain-termination is assumed to occur in this work; termination by combination (CTC) and by disproportionation (CTD).
k td 2R 2D+TDB k tc 2R D

CTD CTC

Based on the postulated reaction mechanism, the following rate laws are derived:

58 rM =-2k ti [M]2 -k p [M][R1 ]-k pb [M]([R 2 ]+[R 3 ]) -k tm [M][R]-k p [M][R 4 ] rI =-k ii [I] rS =-k ts [S][R] rTDB =k p [M][R 4 ]+k ([R 2 ]+[R 3 ]) -2k tdb [TDB][R 1 ]+k td {[R 1 ][R]+[R 2 ]([R 2 ]+[R 3 ])+[R 3 ]2 } rSCB =k pb [M][R 3 ] rLCB =k pb [M][R 2 ] rD =k tm [M][R]+k ts [S][R]+k ([R 2 ]+[R 3 ]) +(k tc +2k td ) {[R1 ][R]+[R 2 ]([R 2 ]+[R 3 ])+[R 3 ]2 -2k tdb [TDB][R 1 ]}
where by invoking the quasi-steady-state-assumption on all the radical concentrations, we end up with:

k [M]2 +2k ii [I] [R]= ti k tc +k td [R1 ]= [R] 1++

0.5

[R 2 ]=[R1 ] [R 3 ]=[R1 ] [R 4 ]= = = k tm [R] k tp

(k tp [TC]+k tdb [TDB]) (k pb +k tm )[M]+k ts [S]+k + (k tc +k td )[R]

(k bb +k tdb [TDB]) (k pb +k tm )[M]+k ts [S]+k + (k tc +k td )[R]

[TC]=[M]f -[M]-[TDB]-[SCB]-[LCB]

3.2.2 Process Model

59
An isothermal semi-batch solution polymerization process model can then be derived:

dV =FT dt d[M] 1 = (FM [M]M +rM V-[M]FT ) dt V d[I] 1 = (FI [M]I +rI V-[I]FT ) dt V d[S] 1 = (FS [S]S +FI [S]I +rS V-[S]FT ) dt V d[TDB] 1 = (rTDB V-[TDB]FT ) dt V d[SCB] 1 = (rSCB V-[SCB]FT ) dt V d[LCB] 1 = (rLCB V-[LCB]FT ) dt V d[D] 1 = (rD V-[D]FT ) dt V
where FT = FM + FS + FI. The measured variables are given by:
xm = [M]f -[M] [M]f 1 FM [M]M dt V 0
t

[M]f = Mn =

[M]f -[M] MWm [D] [TDB] 100 [M]f -[M]

TDBH= BPH=

[SCB]+[LCB] 100 [M]f -[M]

60
3.3 PARAMETER ESTIMATION AND MODEL VALIDATION

Reaction rate constants are then estimated from the four types of measurements shown in Figure 9. The parameter values are given in Table 8. The rate of thermal initiator decomposition that is used in this study is that of t-butyl peroxide; kii = 3.15 x 1014 exp(-1.38 x 105/ RT) s-1. The quality of the model predictions is shown in Figure 9.

Table 8. Values of the estimated parameters

Parameter kti kp ktp kbb ktm kts kpb k ktdb ktc ktd

Unit L.mol-1.s-1 L.mol-1.s-1 L.mol-1.s-1 s-1 L.mol-1.s-1 L.mol-1.s-1 L.mol-1.s-1 s-1 L.mol-1.s-1 L.mol-1.s-1 L.mol-1.s-1

160C 1.33E-10 1.73E+02 1.57E-01 2.15E+01 3.38E-04 1.18E-01 8.47E+00 1.84E+01 3.97E+02 5.76E+00 1.76E+01

180C 1.33E-09 9.25E+02 1.69E+00 1.20E+02 9.05E-03 1.47E+00 1.14E+01 3.06E+01 1.63E+03 2.02E+02 7.96E+01

61
1

10000

0.8

8000

xm

0.4

M
0.2 0 0 50 100 150 200

0.6

6000

4000

2000

time, min
8 7 6

0 0

50

100

150

200

time, min
8 7 6

TDBH

5 4 3 2 1 0 0 20 40 60 80 100 120

BPH

5 4 3 2 1 0 0 20 40 60 80 100 120

time, min

time, min

Figure 9. Measurements and model predictions of monomer conversion (xm), number-average molecular weight (Mn), number of terminal double bonds per 100 monomer units (TDBH) and number of branching points per 100 monomer units (BPH) when the reactor is operate isothermally. The dashed and solid lines represent model predictions at 160 and 180C respectively, and the triangles and circles represent measurements at 160 and 180C respectively.

62
3.4 OPTIMIZATION AND OPTIMIZATION VALIDATION

3.4.1 Performance Index

The performance index to be minimized is formulated by considering typical resin recipes and general knowledge of industrial polymerization processes. In particular, it is desirable to achieve the following objectives: Monomer conversion. A monomer conversion of 100% is desirable, since unreacted monomers can undergo further (unwanted) polymerization. Number-average molecular weight. Low viscosity resins are desired for automotive coatings applications. This generally translates to lower number-average molecular weight resins. In a typical batch production of resins, a certain number-average molecular weight is desired. Number of branching points. Effect of branching on polymer properties in general is still not well understood. Thus, currently production is based on experience or knowledge of the 'optimum' recipe. The number of branching points is kept constant in each recipe. Number of terminal double bonds. The existence of terminal double bonds creates opportunities for macromonomer production. Manufacturers

63
would ultimately like to have control over number of terminal double bonds in the end product. Initiator. Thermal initiators are the most expensive component in the resin production recipe. Thus, it is desirable to minimize the amount of initiator used in the resin production. Based on the above considerations, the following performance index is formulated:
2 2 M n -1 + BPH -1 + TDBH -1 + m I J=1 (1-x m ) +2 3 4 5 M n,des BPH des TDBH des m I,t 2 2 2

where 1, , 5 are the weights on the individual objectives. Their values depend on the goal of optimization.

3.4.2 Optimization Constraints

The above performance index is minimized subject to:

Optimizing Variable Constraints. These constraints represent the feasible ranges of the optimization variables.

Balance equations. These are the conservation equations that govern the process.

64
Hardware and safety limits. Safety concerns and instrumentation capabilities are represented as limits on how steep change in temperatures can be.

the optimization constraints are as follows:

f
1

i,j

=f j , fI

j=M,S

f
1

i,I

0 f i,j f j , Ti -Ti+1 (T )des

j=M,S,I

i=1,,N

i=1,,N i=1,,N

393.2 Ti 573.2

3.4.3 Optimization

Different batch times are selected and evaluated. Each batch time is divided into N equal time intervals, and each feed flowrate and temperature profile is discretized into M equally-spaced constant values. Thus, the feed flowrate and temperature profiles are piece-wise constant. Since we have 3 different flow rates (monomer, solvent and initiator), we end up with 4 x M x N optimizing variables. When choosing N and M one should take into account the

65
computational cost of the optimization. In other words, N and M should be large enough such that the performance index is minimized adequately.

In this study, the performance index minimum is obtained by simulating all possible profiles subject to the defined constraints (simulation), and subsequently, searching through the list of the resulting performance indices for a minimum (optimization). An optimal reactor feed policy subject to an isothermal condition is first calculated. Subsequently, based on the optimal feed policy, different temperature profiles are simulated to obtain a better optimum. Polymerization recipe used in the optimization and optimization condition is given in Table 9. Optimization results are shown in Table 10.

Table 9. Optimization Settings

Monomer (n-butyl acrylate) Solvent (xylene) Initiator (t-butyl peroxide) 1 = 2 = 3 = 4 = 5 N=M Mn,des TDBHdes BPHdes (T)des

= 300 g = 600 g = 0.01% x Monomer =1 =4 = 2000 =5 = 10 = 5C

66
Table 10. Optimal feed and temperature profiles

Time(min) Fraction of monomer Added

100 0.5 0.5 0 0 1 0 0 0 0 0.5 0 0 170 165 160 155 100% 2253 5.4 7.8 0.8089

200 0.5 0.5 0 0 1 0 0 0 0 0 0.5 0 170 175 175 180 100% 2053 5.1 8.1 0.7708

300 0.75 0.25 0 0 1 0 0 0 0 0.5 0 0 170 175 180 185 100% 2035 4.6 9.7 0.7455

400 0.75 0.25 0 0 0.5 0.5 0 0 0 0.5 0 0 170 175 180 175 100% 1978 4.8 10.2 0.7387

Fraction of solvent Added

Fraction of initiator Added

Temperature (C)

xm Mn TDBH BPH J

Optimization results were validated experimentally. The optimization validation results are shown in the Figure 10. The validation is a comparison between experimental data and model predictions from the method of moments model presented in the mechanistic modeling section. Optimal feed policy from tendency model and method of moments models yield the same results. The microstructures are presented in terms of quantities per chain, calculated as follows:

67 TDBH DPn 100 BPH BPH= DPn 100

TDBC=

Challenges in experimental validation include; (i) limited process operator experience in carrying out high-temperature experiments, (ii) possible existence of process-model mismatch, (iii) sub-optimal control policies and (iv) limited experience in utilization of modern polymer characterization techniques in optimal control studies.

68

0.8

Conversion
Mn

0.6

0.4

0.2

6E3

4E3

2E3

2.5

Polydispersity

1.5

1 0

100

200

300

Time, minutes

Figure 10. Optimization validation results for 300-minute optimal control recipe. Triangles and lines represent measurements and model predictions, respectively.

69

0.2

0.15

[SCB], mol/L
[TDB], mol/L

0.1

0.05

0.1

0.08

0.06

0.04

0.02

0 0

100

200

300

Time, minutes

Figure 11. Corresponding polymer microstructure profiles for the 300-minute optimal control recipe. Triangles and lines represent measurements and model predictions, respectively.

70

3.5

CONCLUSIONS

In this study, a class of batch polymerization reactor was investigated. A mechanistic model for an isothermal semi-batch reactor was developed based on a proposed set of reaction mechanism of high-temperature solution polymerization of alkyl acrylates. Optimal control policies were calculated based on minimization of a proposed performance index. The performance index was formulated based on desired micro-structural properties such as number of terminal double bonds and branching points, and more conventional objectives such as maximization of monomer conversion, minimization of thermal initiator and obtainment of a pre-determined number-average molecular weight. Optimization results revealed non-intuitive control policies which allow for the optimization of end-product properties and the minimization of production cost.

71

REDUCED-ORDER MODEL FOR MONITORING SPECTROSCOPIC AND CHROMATOGRAPHIC PROPERTIES

With increasing accessibility to advanced measurement techniques, recent studies focus on improving process modeling and monitoring approaches in a variety of applications. The unifying theme of recent studies in kinetic modeling is the integration of chemical information obtained from a variety of measurements. The goal of these studies has been to elucidate the dynamics governing the behavior of a process. For example, a combination of mass action kinetic model and artificial neural network has recently been applied to monitor different levels of pesticide contents in fruit and vegetable samples, whose levels were obtained by using spectrophotometry techniques [Nising et al., 2004]. Similar studies focused on (a) identifying and estimating a unique set of rate coefficients by simultaneously fitting a set of measurements [Bezemer et al., 2001; Zogg et al., 2004; Rantow et al., 2006], (b) understanding how changes in measurements affect model parameter values and initial rate of reactions [Carlson et al., 1993], (c) decreasing parameter search space by including parametric non-negativity constraints in the parameter estimation [Jaumot et al., 2005], and (d) understanding the effect of the initial guesses on an unconstrained optimization approach to parameter estimation [Zhu et al., 2002]. These studies emphasize the need for a systematic approach to using different measurements in modeling and monitoring strategies. Mechanistic models often provide inaccurate extrapolating predictions in complex polymerization systems. This can be attributed to phenomena that occur at the extrapolated conditions but are not accounted for in the model. To address this

72
problem, one usually resorts to continual improvement of the mechanistic models or development of process monitoring strategies that use off-line and on-line measurements to correct the predictions. Research directions in this area include multivariate monitoring or data processing strategies by using statistical tools [Kourti et al., 1995; Negiz et al, 1997; Neogi et al., 1998], model-based monitoring strategies [Soroush et al., 1997; Tatiraju et al., 1999] and characterization of batch reactions with in situ spectroscopic measurements, calorimetry and dynamic modeling [Ma et al., 2003]. Particularly in process monitoring of classical batch polymerization reactors, with the advancements made in `hardware' sensors [Kammona et al., 1999], the current challenge is in integrating data from emerging measuring devices into existing monitoring systems (e.g., monitoring of polymer chain end groups by using nuclear magnetic resonance spectroscopy [Lutz et al., 2005]). This study was motivated by the observation of complex phenomena [Grady et al., 2002] in high-temperature free-radical polymerization of acrylates. One such phenomenon is spontaneous thermal polymerization of n-butyl acrylate, i.e. polymerization without the addition of known free-radical initiators. Reproducible, sustained conversion of monomer to polymer in a few hours was observed [Quan et al., 2005]. Other phenomena such as back-biting, betascission and long chain branching also occur and complicate the underlying polymerization kinetics. The method-of-moments model presented in [Rantow et al., 2006] is able to successfully predict monomer conversion and numberaverage molecular weight, but it over-predicts the polymer polydispersity index and under-predicts the concentration of terminal double bonds (subsection 3.4.3). In view of these predictive deficiencies, this study aims to (i) develop a simplified reduced-order polymerization reactor model for high-temperature polymerization of n-butyl acrylate, (ii) develop a multi-rate state estimator using the method described in [Zambare et al., 2003] and the reduced order model, and (iii) evaluate the performance of the state observer in predicting numberand weight-average molecular weights (chromatographic properties) and polymer

73
chain microstructures (spectroscopic properties). The approach of modeling

macromolecular kinetics as small-molecular kinetics [Villermaux et al., 1984] is adopted as a model-reduction technique. The polymerization reaction network in study contains 15 irreversible reaction pathways and 10 rate coefficients. The reduced-order reactor model is of order 7 (has 7 state variables), compared to the 20 state variables of the original model. The measurements of the reactor temperature and feed flow rates of monomer, solvent and initiator are the frequent measurements, and the measurements of molecular weight averages and microstructural quantities are the infrequent measurements. The multi-rate state estimator estimates from these measurements.

4.1

EXPERIMENTAL PROCEDURE

The information of the raw material suppliers and preparation of raw materials used in the experiments are reported in [Rantow et al., 2006]. The polymerization is carried out in a 1-liter Mettler-Toledo RC1 Reactor-Calorimeter, which is a jacketed semi-batch glass reactor. The feed and temperature profiles are the optimal profiles already calculated in [Rantow et al., 2005] to achieve certain endproduct properties. Initially, 600 g of solvent (xylene) is loaded to the reactor at room temperature. The reactor temperature is then increased to 170C. Subsequently after the temperature is reached, 300 g of monomer (n-butyl acrylate) is fed to the reactor. At this time, the cooling system embedded in the reactor-calorimeter automatically decreases the jacket temperature (by adjusting the flow rate of cooling fluid to the jacket), to balance the heat generated by the polymerization reaction, thus maintaining the reactor temperature at 170C. This initial temperature drop can be seen in Figure 12.

74

Figure 12. Profiles of the reactor temperature and the flow rates of the monomer (n-butyl acrylate), the solvent (xylene) and the initiator (t-butyl peroxy acetate) solution feed streams. After 75 minutes into the reaction, the reactor temperature is then increased to 175C. Once the temperature is reached, 100 g of monomer is then fed to the reactor, along with approximately 40 mg of initiator solution (10 wt.% of t-butyl peroxy acetate in xylene). This results in a milder temperature drop (Figure 12). Then 75 minutes later, the temperature is again increased to 180C, and 75 minutes after that, the temperature is again increased to 185C. The reaction temperature is then maintained at about 185C for another 75 minutes before the reactor temperature is dropped back to room temperature. It should be noted that in the absence of the thermal initiation phenomenon of n-butyl acrylate this could lead to a monomer pooling (accumulation of unreacted monomer in the reactor)

75
event, although in this particular study the total monomer content is reasonably low (40% by weight). Thus care needs to be taken in terms of knowing whether polymerization is occurring or has occurred prior to the addition of the free-radical initiators.

4.1.1 Off-Line and On-Line Measurements

A sample is taken every 30 minutes. The drawn samples are diluted (50/50 v/v) in a cold inhibitor solution (1000 ppm 4-methoxy phenol in xylene). Polymer solids content in the samples is determined by gravimetry, while molecular weight distributions are determined by using an HP 1090 high performance liquid chromatography (HPLC). Polymer-chain microstructure properties are quantified by using proton nuclear magnetic spectroscopy (1H-NMR) and carbon NMR (13CNMR). More details on the analytical procedures can be found in [Rantow et al., 2006]. Time delay associated with each of the infrequent measurements is 30 minutes. The solvent concentration in the reactor is inferred from 1H-NMR measurement of the hydrogen atoms on the terminal solvent (xylol) groups. The concentration of the solvent in the reactor is calculated from the number of solvent groups per 100 repeat units (SH) in the polymer:

SH =

number of solvent groups in NMR sample number of repeat units in NMR sample area of 1H-NMR peak at H = 7.1 ppm 4 100 = area of 1H-NMR peak at H = 4.1 ppm 2
t

FS dt + ms 0
0

M wS V M wM V
t

[S ] 100

F
0

dt [ M ]

76
where ms0 is the mass of the solvent in the reactor at time t = 0 (right after initial reactor loading and before the feeding starts). Thus, from the above equation, [S] can be inferred from:

t F dt + m SH S [S] = s0 100 M w M V M wS V 0 1

FM dt [M]

Note that this is a bulk average quantity. Also note that the other quantities, i.e. concentrations of terminal double bonds and short chain branches can also be inferred in the same way. Key peak assignments are given in Table 11 and the corresponding molecular structure of poly(n-butyl acrylate) is given in Figure 13. Table 11. Key 1H-NMR and 13C-NMR peak assignments H , ppm
4.1 5.5 6.2 7.1
1

H-NMR Peak Assignment

Atom(s) in Figure
i b a d,e,f,g

Reference
Quan et al., 2005 Chiefari et al., 1999 Chiefari et al., 1999 Quan et al., 2005

H on CH2 of ester linkage next to COO H on CH2 or terminal double bond H on CH2 of terminal double bond H on terminal xylol group

C , ppm
47.2-48.4 165-180

13C-NMR Peak Assignment

Quaternary backbone C C on C=O

Atom(s) in Figure
c h

Reference
Ahmad et al., 1998 Quan et al., 2005

77

H3 C

CH2 H2 C H CH2 O C CH2 H O d C C CH2 O O H2C CH C O CH2 C e C

C C H g

CH3

H C C C O CH2 CH h C O O i i H C H H2 C CH2 CH2 O

CH2 c C C

b H O H2 C

CH2 CH3

CH2 CH2 H3 C CH2

H2C

H3 C

CH2 H2C CH3

Figure 13. Molecular structure of the synthesized poly(n-butyl acrylate). Molecular structure shows the key atoms used to infer polymer microstructure concentrations.

The concentration of solvent can also be readily measured using and on-line or at-line gas chromatography. The reactor temperature, the volume of monomer added to the reactor, the volume of the reacting mixture inside the reactor, and the feed flow rates are measured at a sampling rate of one minute with no time delay; these are frequent measured inputs.

78
4.2 REDUCED-ORDER MECHANISTIC POLYMERIZATION MODEL

The polymerization reaction network relevant to polymerization reactions of alkyl acrylates at higher temperatures has been reported in [Peck et al., 2004; Rantow et al., 2006]. In this study, we simplify the polymerization reaction network given previously in Table 2 by using a small-molecule representation. The underlying assumption behind the simplification of the reaction network is in the classifications of the reacting species. In the reduced-order model we assume that the polymer chain microstructures are formed as if new small molecules are formed, and reactive end-group radicals are viewed as regular radical-containing small molecules. This idea was first put forth by [Villermaux et al., 1984]. This approximation becomes more accurate as the polymer chain length and the concentration of polymer decrease. For this particular study, the polymer chains produced have 30 repeating units on average. The resulting reaction network is given in Table 12, which leads to a reduced-order kinetic model.

79
Table 12. Simplified (using the small-molecule representation) reaction network for polymerization of nBA in xylene, with a monomer content of 20-40 wt%, and at 140-180C.

kti 2 M 2 P
p P + M P

kbb P Q
pq Q + M P + SCB Q D + P + TDB

ktm P + M D + PTDB
p PTDB + M P + TDB

kts P + S D + P ktc 2 P D ktd 2 P 2 D ktc P + Q D ktd P + Q 2 D ktc 2Q D ktd 2Q 2 D

terttert terttert sec tert sec tert sec sec sec sec

The reduced-order reactor model is then developed by writing mass and structural-character balances for the semi-batch reactor. An energy balance is not needed to be included in the process model, because the measured reactor temperature is an input to the model (is a measured input). It should be noted that the density of reacting solution is assumed to be constant. This assumption is deemed appropriate as polymerization is carried out at the lower end of the concentration range (less than 40wt.%), the operating temperature range is narrow, and that the solvent (xylene) and the monomer (n-butyl acrylate) have similar densities. The resulting ordinary differential equations (ODEs) describing the dynamics of the reactor are:

80

[I ] f I (T , V ,[ I ], FI , FM , FS ) [M ] f M (T , V ,[ M ],[ I ], FI , FM , FS ) d [ S ] f S (T , V ,[ S ],[ M ],[ I ], FI , FM , FS ) = dt [TDB ] fTDB (T ,V ,[ M ],[TDB],[ I ], FI , FM , FS ) [ SCB] f (T ,V ,[ M ],[ SCB],[ I ], F , F , F ) I M S SCB f T ,V ,[ M ],[ S ],[ D],[ I ], F , F , F [ D] D ( I M S )
where
1 [ I ]( F + F + F ) I F f I (T ,V ,[ I ], FI , FM , FS ) = RI ([ I ], T ) I M S I V M wI f M (T , V ,[ M ],[ I ], FI , FM , FS ) = RM ([ M ],[ P],[Q],[ PTDB ], T ) 1 [ M ]( F + F + F ) M F I M S M V M wM f S (T , V ,[ S ],[ M ],[ I ], FI , FM , FS ) = RS ([ S ],[ P], T ) 1 [ S ]( F + F + F ) S F S I M S V M wS TDB fTDB (T , V ,[ M ],[TDB],[ I ], FI , FM , FS ) = RTDB ([Q],[ M ],[ P ], T ) FI + FM + FS V f SCB (T , V ,[ M ],[ SCB],[ I ], FI , FM , FS ) = RSCB ([Q],[ M ], T ) [TDB] FI + FM + FS V f D (T , V ,[ M ],[ S ],[ D],[ I ], FI , FM , FS ) = RD ([Q],[ M ],[ S ],[ P ], T ) [ SCB ] [ D] FI + FM + FS V

where the state variables, [I], [M], [S], [TDB], [SCB] and [D] are the concentration of the initiator (t-butyl peroxy acetate), concentration of the monomer (n-butyl acrylate), concentration of the solvent (xylene), concentration of terminal double bonds, concentration of short chain branches, and concentration of `dead' polymer chains, respectively. The measured inputs V, FI, FM, FS, and T are the volume of the reacting mixture, the reactor temperature, and the volumetric flow

81
rates of the initiator solution, the monomer and the solvent, respectively. The vector of infrequent measured outputs:

M n H M n (VM ,V,[M],[D]) M w H M (VM ,V,T,[M],[D]) w [TDB] = [TDB] [SCB] [SCB] [S] [S]
where

V MW m H M n (VM ,V,[M],[D]) = M M -[M] M V [D] wM V V -[M] H M w (VM ,V,T,[M],[D]) = c1 +c 2 M M +c3T M M VM [D] V wM

t

VM = FM dt
0

VM is also a frequent measured input. Note that weight-average molecular weight is calculated by multiplying numberaverage molecular weight by polydispersity index, PDI, given by the empirical correlation:

V PDI=c1 +c 2 M M +c3T V

where the correlation constants c1=2.66031, c2=0.03337 and c3=-0.00929. The second term on the right hand side of the correlation is c2 times the weight fraction of the monomer fed to the reactor. The correlation constants are obtained by fitting the correlation to the measurements shown in Figure 14.

82

3 140oC 160 C 2.5 180oC

o

Polydispersity

1.5

1 0

10

20

30

40

50

Initial Monomer Content, wt.%

3 10wt.% 20wt.% 30wt.% 40wt.%

2.5

Polydispersity

1.5

1 120

140

160

180
o

200

Temperature, C

Figure 14. Measurements of polydispersity index at different temperatures (constant initial monomer weight percent of 40%) and different initial monomer weight percents (constant temperature of 160C).

83
The reaction rate laws are:

R I =-k d [I]
tert R M =-2k ti [M]2 -k sec [M]([P]+[P TDB ]) -k p [M][Q]-k tm [M][P] p

R S =-k ts [S][P] R TDB =k [Q]+k sec [M][P TDB ] p

tert R SCB =k p [M][Q]

R D =k [Q]+ ( k tm [M]+k ts [S])[P]+ ( k sec-sec +2k sec-sec )[P]2 tc td

tert-tert + (k sec-tert +2k sec-tert )[P][Q]+ ( k tert-tert +2k td )[Q]2 tc td tc

where the concentrations of the free radicals are given by the following algebraic equations, obtained by applying the quasi-steady-state assumption:
2 FI +FM +FS FI +FM +FS +8k (k [M]2 +2k [I]) + t ti d V V [R]= 2 4 (k ti [M] +2k d [I]) 0.5

[P]= =

[R] 1++ k bb FI +FM +FS V

tert k p [M]+k +k t [R]+

k tm [M] F +F +F k sec [M]+ I M S p V [Q]=[P] = [P TDB ]=[P]

84
4.3 REDUCED-ORDER MULTI-RATE STATE OBSERVER DESIGN

On the basis of the developed reduced-order model, a reduced-order multi-rate state observer is designed using the method described in [Tatiraju et al., 1999; Zambare et al., 2003]. The estimator is of the form: & [ I ] = f I T , V ,[ I ], FI , FM , FS

) )

& [ M ] = f M T ,V ,[ I ],[ M ], FI , FM , FS

& [ S ] = f S T , V ,[ S ],[ M ],[ I ], FI , FM , FS + L31 [ S ]* [ S ]

{
S

}
*

& [TDB ] = fTDB T ,V ,[TDB],[ M ],[ I ], FI , FM , FS + L42 [TDB]* [TDB] & [ SCB] = f SCB
I M
53

( (T ,V ,[SCB],[M ],[ I], F , F

& [ D] = f D T , V ,[ D],[ S ],[ M ],[ I ], FI , FM

) { } , F ) + L {[ SCB] [ SCB]} , F ) + L {M H (V ,V ,[ M ],[ D])}

*

64

Mn

where each term with an asterisk represents the predicted present value of the variable with the corresponding infrequent measurement. This present value is obtained by the least-squared-error fit of a line to the three most recent measurements of the infrequent measured output. The second term in the right hand side of the estimator equation is simply a corrective feedback term; as long as there are discrepancies between the predicted present value of a variable (variable with asterisk) and the corresponding estimate (variable with hat), the corrective action is made. The estimator calculates continuous estimates of the concentrations 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

85
initiator solution feed streams, and the volume of monomer fed to the reactor; and The infrequent measured outputs: the concentrations of the solvent, terminal double bonds, short polymer chain branches, and the numberaverage molecular weight of the polymer. . An analysis of the local observability of the system (Table 13) led to setting all of the estimator gain entries except for the entries L31, L42, L53 and L64 to zero. Table 13. Simplified (using the small-molecule representation) reaction network for polymerization of nBA in xylene, with a monomer content of 20-40 wt%, and at 140-180C. Measurement State Variable V [I] [M] [S] [TDB] [SCB] [D] Yes Yes Yes Yes No No No Yes Yes Yes No Yes No No Yes Yes Yes No No Yes No Yes Yes Yes Yes No No Yes Yes Yes Yes Yes No No Yes [S] [TDB] [SCB] Mn Mw

The non-zero observer gain entries L31 = 1 x 10-1, L42 = 1 x 10-1, L53 = 1 x 10-1, and L64 = -5 x 10-6, which ensure that all eigenvalues of the Jacobian of the observer error dynamics lie in the left half of the complex plane. The non-zero terms in effect indicate which slow measurement is used to correct a state variable estimate. A more thorough discussion on how the estimation error (between the estimates and the true values) depends on the estimator gain matrix entries is given in [Zambare et al., 2003].

86
4.4 STATE OBSERVER PERFORMANCE

The flow of information between the reactor and the multi-rate observer is depicted in Figure 15. Measurements of the reactor temperature and the feed flow rates were fed to the nonlinear reduced-order multi-rate state estimator. The estimator initial conditions were set to that of the actual reactor.

Figure 15. Flow of information between the polymerization reactor and the nonlinear reduced-order multi-rate state observer. The performance of the estimator in providing estimates of the number-average and weight-average molecular weights is shown in Figure 16.

87

8E3

6E3

4E3

2E3

x 10

1.5

Mw

0.5

0 0

100

200

300

Time, minutes
Figure 16. Estimated vs. measured values of the number- and weight-average molecular weights (dashed line = estimate without using feed back measurement, solid line = estimate using feed back measurement, circles = measurement).

88
In the estimator implementation, the entries L31, L42, L53 and L64 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. The latter case, a line is fitted to two measurements.

0.1

5.5

0.08

[TDB], mol/L
100 200 300

[S], mol/L

0.06

0.04

4.5

0.02

4 0

0 0

100

200

300

Time, minutes

Time, minutes

Figure 17. Estimated vs. measured values of the concentrations of the solvent and terminal double bond (dashed line = estimate without using feed back measurement, solid line = estimate using feed back measurement, circles = measurement). Bottom figure shows evolution of the 1H-NMR peaks.

89
The estimates of the polymer microstructure quantities calculated by the multirate state estimator are shown in Figures 17 and 18, which demonstrate that the state estimator provides accurate continuous 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.

Figure 18. Estimated vs. measured values of the concentrations of the short chain branches (dashed line = estimate without using feed back measurement, solid line = estimate using feed back measurement, circles = measurement). Bottom figure shows evolution of the 1H-NMR peaks.

90
Selection of the non-zero estimator gain entries based on the local observability analyses is shown to provide adequate information to the state observer. Selection of the non-zero entries show that the reactor volume, monomer concentration and initiator concentration can be estimated solely based on the fast measurable inputs. And there is a one-to-one pairing between the slow measurable outputs and the rest of the state variables.

4.5

CONCLUDING REMARKS

A reduced-order high-temperature mechanistic polymerization model was presented. The model is capable of predicting spectroscopic and chromatographic polymer properties. On the basis of the model, a reduced-order multi-rate nonlinear state observer was developed for a semi-batch, hightemperature, n-butyl acrylate, solution polymerization reactor. Selection of the non-zero estimator gains based on the local observability analysis reveals optimal 'pairings' between the observer estimates and measurements of the chromatographic and spectroscopic variables. Real-time results from the on-line implementation of the estimator (soft sensor) confirmed the reliability of the estimator to provide accurate estimates of spectroscopic and chromatographic polymer properties. The feedback feature of `closed-loop observers/estimators provide process models with much needed robustness to process-model mismatch.

91

GLOBAL IDENTIFIABILITY OF REACTION RATE PARAMETERS IN MECHANISTIC MODELS FOR THE CHAIN HOMOPOLYMERIZATION

Recent efforts in the modeling and monitoring of polymerization reactors have been focused on understanding the underlying reaction mechanisms in broader ranges of process operation. For example, such efforts have been motivated by the increasing average temperature used in industrial polymerization reactors (e.g., batch chain-polymerization reactor for acrylics [Grady et al., 2002] and continuous chain-polymerization reactor for styrene [Campbell et al., 2003]). Broadening the range of process operations often leads to increasing complexities in the underlying polymerization chemistry (e.g., in alkyl acrylate polymerization at 25-140C [Peck et al., 2004]). The added complexity requires one to revisit/re-evaluate models that were developed in the narrower range of process operation, and subsequently, obtain a more representative set of reaction mechanisms (reaction network) and the corresponding parameter (rate constant) values.

The parameter estimation problem has been formulated as follows. Given a set of process measurements, postulate a set of reaction mechanisms and estimate the corresponding set of parameters. While such formulation may seem straightforward, the simultaneous parameter estimation is often hindered by (i) limited

92
number and type of available measurements, (ii) limited knowledge of dominant reaction mechanisms, and (iii) the obtaining of more than one set of parameters that give comparable model predictions. Hence, what at first seems to be a simple parameter estimation problem is also a kinetic model identification problem. The problem is inherently complex, as one needs to simultaneously consider the nature of the measurements (outputs), the underlying polymerization chemistry, the modeling techniques, and all the elements of the parameter estimation problem (e.g., optimization/search algorithms [Maria, 2004], optimization problem formulation [Bindlish, 2003]). A more comprehensive treatment of the problem has been called design of experiments [Jacquez, 1998], where the quality (e.g., sampling frequency) of output measurements and the type of input sequences are also of interest. Often times, estimates are solely based on available (conventional) measurements (e.g., in polymerization of styrene [Hui et al., 1972] and methyl methacrylate [Stickler et al., 1981]). Currently, estimates are based on both advanced and conventional measurements (e.g., in high-temperature acrylates polymerization [Busch et al., 2004; Rantow et al., 2006]). Spectroscopic measurements help reveal characteristic polymer chain structures that are formed by specific reactions.

What is not always understood is how such inclusion help alleviate the parameter estimation or kinetic model identification problem. Therefore, this question brings to light other questions such as (i) whether a unique set of parameter values can be inferred from the available measurements (e.g., parametric identifiability

93
[Walter et al., 1996], parametric observability [Vidal et al., 2002; Vecchio et al., 2003]) and (ii) how sensitive a measurement is to a parameter (e.g., sensitivity analyses [Polic et al., 2004]). Recent case studies mostly deal with biological/biochemical systems [Audoly et al., 2001; Vajda et al., 1994], and a few with catalytic chemical systems [Yao et al., 2003; Ben-Zvi et al., 2004]. Several methods for analyzing identifiability have been put forth; they include: analyses of explicit solutions through series expansions [Walter et al., 1996], analyses of the state isomorphism matrix [Walter et al., 1996; Vajda et al., 1994], and analyses of the linearized structures of the output (or input-output) functions through differential algebra manipulations [Ljung et al., 1994; Denis-Vidal et al., 2000].

Motivated by the need to understand the nature of parametric identifiability, particularly in relevance to kinetic model identification and parameter estimation in polymerization systems, this study aims to (i) analyze the degree of identifiability of the individual parameters (reaction rate constants) from state-ofart measurements [Kammona et al., 1999] and (ii) assess the distinguishability of process models that arise from different reaction networks. Chain homopolymerization systems considered in this study are those that represent the high-temperature polymerization kinetics of alkyl acrylates.

94
5.1 SCOPE AND PRELIMINARIES

This section describes the scope of this study and reviews the notions of local and global identifiability of model structure and model parameter, and distinguishability of model structures.

5.1.1 Local and Global Identifiability of Model Parameter and Model Structure

Consider the class of processes that can be represented by a parameterized non-linear state-space model of the form:

x=f ( x,,u ) & M (): y=h ( x,,u )

where x = [x1 xn]T denotes the vector of model state variables, = [ 1 i]T denotes the vector of model parameters, u = [u1 up]T denotes the vector of model inputs, and y = [y1 ym]T denotes the vector of model outputs. The vector of output measurements is represented by = [1 m]T. It is assumed that continuous measurements of the outputs are available; that is, = y(t).

Model parameter values are commonly obtained by solving the least squarederror (non-linear regression) problem:

95
N T

min h i ( x,,u ) -yi ( t )

i=1 0

Definition 1 [Bellman et al., 1970]: A model structure is locally identifiable from a given set of output measurements, 1(t) m(t), generated by the vector
of parameter values , if = is a local minima of the non-linear regression problem. The model structure is globally identifiable, if = is a global minima of

the non-linear regression problem. In other words, if the model is not identifiable, then there are more than one parameter vectors that correspond to exactly the same input-output behavior, and it is impossible to eliminate them from the data alone.

Remark 1: The model structure generally refers to the vector function f(x,,u) which represents analytical relationships between model states, model parameters and model inputs.

Remark 2: The notion of structural identifiability is associated with the nature of the solution to the parameter estimation problem. Hence in this study the notion of parametric identifiability is interchangeably used in place of structural identifiability.

96
5.1.2 Distinguishability of Model Structures

% Definition 2 [Walter et al., 1996]: A model structure, M (.) , is distinguishable from

% a model structure, M (.) , if and only if for almost any parameter vector value, , there is no other parameter vector value, , such that

% % M () M ()
% % where M denotes the model structure, and the above equation reads M () has
% the same input-output behavior as M () . In other words, if M (.) is not

distinguishable from M (.) , then there is no way to discover from experimental data that the structure M(.) of the model does not correspond to that M(.) of the process that has generated the data.

5.1.3 Scope

In this study, we limit the scope to the special form of the general nonlinear model put forth in 3.4.1.1:

x = f ( x) & M ( ) : y = h ( x)
where is a parameter matrix and m=n. It is assumed that the model outputs are not redundant; that is, the nxn matrix

97

h ( x) x
is nonsingular locally. Then it is possible to write:

G ( y ) = Hx
where H is a constant nonsingular nxn matrix, and the states can be expressed in terms of the outputs:

x = H 1G ( y )

The model in terms of the outputs takes the form:

d 1 1 H G ( y ) = f ( H G ( y )) dt
Then linear regression can be used to estimate the parameter matrix [Ljung et al., 1994]. The parameter matrix will be unique (model will be globally identifiable), if the right hand side of the above equation depends on elements of only.

5.2

CHALLENGES IN FREE-RADICAL POLYMERIZATION

Current issues in mechanistic models for free radical polymerization chemistry include:

98
Measurements of the concentration of intermediate radicals are not available, as the intermediate radicals exist in very low concentrations and have very short life time. Batch free-radical models are usually not cast in a linear form (with respect to the parameter matrix); the optimization problem of the parameter estimation is non-convex. Role of measurements on parametric identifiability is not well understood yet. More than one postulated network is able to capture the dynamics of a given set of measurements; more than one set of parameters is able to capture the given set of measurements.

Before carrying out polymerization and collecting data on the reactor, one should check to see if there is any possibility of selecting the best model structure and estimating uniquely its parameters from the experiments considered, irrespective of the quality of the experimental data. To this end, the global identifiability of mechanistic models with different sets of measurements and the distinguishability of globally identifiable model structures are assessed. A review of current state-of-art on-line measurements for polymerization reactors can be found in [Kammona et al., 1999].

99

Table 14. Polymerization reaction networks for initiation-propagation-termination scheme R1

kd I 2 R
p R + M R

R2
kd I 2 R
p R + M R

kt 2 R D

kt 2 R 2 D + TDB

5.2.1 Example 1

Consider a polymerization system in which only the primary chain polymerization reactions take place; i.e., chain initiation, chain propagation and chain termination by combination (reaction network R1 in Table 14). Using species mass balances, one can develop the chain-polymerization batch-reactor model:

& x1 = 1 x1 & x2 = 2 x2 x4
2 & x3 = 3 x4 2 & x4 = 1 x1 3 x4

x1 (0) = x1,0 x2 (0) = x2,0 x3 (0) = 0 x4 (0) = 0

y1 = x1 y2 = 1 y3 = x2 x2,0 x3

x2,0 x2

where x = [x1 x2 x3 x4]T = [ [I] [M] [D] [R] ]T is the vector of state variables, = [1 2 3]T = [kd kp kt]T is the vector of parameters, and y = [y1 y2 y3]T = [ [I] xm DPn ]T is the vector of measurements. Invoking quasi steady state assumption (QSSA) for

100
the molar concentration of radicals, x4, and writing the states in terms of the outputs, we obtain:
1 0 x1 0.5 x1 d x2 = 0 2 1 0.5 x x dt 3 1 2 x3 1 0

= x1 G ( y) = y1 x2,0 (1 y2 ) x2 y2 x3 x2,0 y3
The following observations can be made:

The parameters 1 and 2 1 are globally identifiable from the set of

3

0.5

measurements. The individual parameters of 2 and 3 are not identifiable from the set of measurements. Inclusion of measurement of y3 does not improve the parametric identifiability of the system.

Considering the above observations, one can experimentally determine the globally identifiable lumped parameters, and not rely on the measurement of number-average molecular weight (Mn) on improving the quality of the parameter estimates. Also note that the measurement of y1 (concentration of initiator) is a

101
measurement where the solution to the globally identifiable parameter 1 and

lumped parameter 2 1 are dependent upon.

3

0.5

5.2.2 Example 2

One may also consider a reaction network that is the same as the above, except for the chain termination mechanism (reaction network R2 in Table 14). Such a case is considered when one is not certain of the path by which the propagating chain terminates. To our advantage, terminal double bonds (TDBs) are created whenever the chain termination by disproportionation occurs. Therefore, the TDBs can be used as a measure of how often (reaction rate constant) the chain termination by disproportionation occurs. The structure of TDB is typically measured by using proton nuclear magnetic spectroscopy (1H-NMR) [Peck et al., 2004; Quan et al., 2005; Rantow et al., 2006]. By using the same modeling approach we arrive at the following model:

& x1 = 1 x1 & x2 = 2 x2 x4
2 & x3 = 23 x4 2 & x4 = 1 x1 3 x4 2 & x5 = 3 x4

x1 = x1,0 x2 = x2,0 x3 = 0 x4 = 0 x5 = 0

102
y1 = x1 y2 = 1 y3 = y4 = x2 x2,0 x3 x5 x2,0 x2

x2,0 x2

where x = [x1 x2 x3 x4 x5]T = [ [I] [M] [D] [R] [TDB] ]T is the vector of state variables, = [1 2 3]^T = [kd kp kt]T is the vector of parameters, and y = [y1 y2 y3 y4]T = [ [I] xm DPn TDB]T is the vector of measurements. Making QSSA for the molar concentration of the intermediate radicals, x4, we obtain:

1 0 0.5 x 1 d 0 2 1 x1 x2 = 3 0.5 dt x1 x2 x3 21 0 x5 1 0 = x1 G ( y) = y1 x2,0 (1 y2 ) x2 y2 x3 x2,0 y3 x 5 x2,0 y2 y4

Note that the two models are indistinguishable from the measurements and the addition of the spectroscopic measurement of TDB does not improve the parametric identifiability of the system. Such prior knowledge is useful as the measurement of TDB requires a time-consuming procedure and is not always obtainable on-line. It should also be noted, however, that the experimental

103
evidence of the existence of TDBs does suggest that the network R2 is the more accurate description of the system.

5.3

GLOBAL PARAMETRIC IDENTIFIABILITY OF CHAIN HOMOPOLYMERIZATION SYSTEMS

Considered here are models that are postulated for homopolymerization systems which show polymer chain branch formations. Based on current NMR techniques, short- and long-chain branches are not always experimentally distinguishable [Quan et al., 2005]. Two reaction networks are considered. The first one attributes branching formation to the preceding intra-molecular chain transfer reaction (reaction network R3 in Table 15), and the second reaction network attributes the branching formation to the preceding inter-molecular chain transfer (reaction network R4 in Table 15). In the reaction network R1, short chain branches (SCB) are formed through the intra-molecular chain transfer (backbiting) followed by propagation of the tertiary radical (Q). While in the reaction network R2, the long chain branches (LCB) are formed through the intermolecular chain transfer (chain transfer to polymer) followed by the propagation of the tertiary radical (P). The two reaction networks are typical representations of how chain branches are formed. From experimental measurements, it is difficult to distinguish one reaction network from the other. Currently, branch quantities are determined through carbon nuclear magnetic spectroscopy (13C-

104
NMR) [Peck et al., 2004; Quan et al., 2005; Rantow et al., 2006]. Often times, both reaction networks can capture the dynamics of the concentration of chain branches.

Table 15. Polymerization reaction networks for initiation-propagation-transfer to polymer-termination scheme

R3
kd I 2 R
p R + M R

R4
kd I 2 R
p R + M R

kt 2 R D kbb R Q
pq Q + M R + SCB

kt 2 R D
tp R + TC + D P pp P + M R + TC + LCB

5.3.1 Example 3

Similar to Examples 1 and 2, based on species mass balances, we can develop an isothermal chain-polymerization batch-reactor model [Rantow et al., 2006]:

& x1 = 1 x1 & x2 = 2 x2 x4 5 x2 x5
2 & x3 = 3 x4 2 & x4 = 21 x1 4 x4 + 5 x2 x5 23 x4

x1 = x1,0 x2 = x2,0 x3 = 0 x4 = 0 x5 = 0 x6 = 0

& x5 = 4 x4 5 x2 x5 & x6 = 5 x2 x5

105
y1 = x1 y2 = 1 y3 = y4 = x2 x2,0 x3 x6 x2,0 x2

x2,0 x2

where x = [x1 x2 x3 x4 x5 x6]T = [ [I] [M] [D] [R] [Q] [SCB] ]T is the vector of state variables, = [1 2 3 4 5]^T = [kd kp kt kbb kpq]T is the vector of parameters, and y = [y1 y2 y3 y4]T = [ [I] xm DPn SCB]T is the vector of measurements. Again, making QSSA for the concentration of free radicals, we obtain:

1 0 0 0.5 0.5 x 0 1 4 1 x1 2 1 3 3 0.5 d x1 x2 x2 = 0 0 dt 1 x 0.5 x3 0.5 1 1 x6 0 0 4 3

= x1 G ( y) = y1 x2,0 (1 y2 ) x2 y2 x3 x2,0 y3 x 6 x2,0 y2 y4

Observations:

106 The parameters 1, 2 1 and 4 1 are globally identifiable from the 3 3

measurements. The parameters 2, 3, and 4 are not identifiable from the measurements. The measurements y3 and y4 do not improve the parametric identifiability of the model. The parameter 5 does not have any effect on the measurements; it is not identifiable due to the model structure.
0.5 0.5

As in Example 2, measurement of the number-average molecular weight (y3) does not improve the global identifiability of the model. The spectroscopic measurement of the short chain branches (SCB) also does not improve the identifiability of the model. By analyzing the parameter matrix, we can also see that the parameter 5 does not at all affect the set of measurements. This is contrary to the common intuition which assumes that once we have included the measurement of branch quantities (y4)that we would obtain a more reliable measurement of the rate of branch formation (5).

5.3.2 Example 4

One can also postulate a reaction network where the branch formation is attributed to the inter-molecular chain transfer to polymer (reaction network R4 in

107
Table 2). Note that in the reaction network, TC (tertiary carbon) is not a reactive species, it is only used to represent how the rate of chain transfer to polymer reaction is governed by the number of tertiary carbons on the polymer chain. Again species mass balances lead to [Rantow et al., 2006]:

& x1 = 1 x1 & x2 = 2 x2 x4 5 x2 x5
2 & x3 = 3 x4 4 x4 x6 2 & x4 = 21 x1 4 x4 x6 + 5 x2 x5 23 x4

x1 = x1,0 x2 = x2,0 x3 = 0 x4 = 0 x5 = 0 x6 = 0 x7 = 0

& x5 = 4 x4 x6 5 x2 x5 & x6 = 5 x2 x5 4 x4 x6 + 5 x2 x5 & x7 = 5 x2 x5

y1 = x1 y2 = 1 y3 = y4 = x2 x2,0 x3 x7 x2,0 x2

x2,0 x2

where x = [x1 x2 x3 x4 x5 x6 x7]T = [ [I] [M] [D] [R] [P] [TC] [LCB]] ]T is the vector of state variables, = [1 2 3 4 5]^T = [kd kp kt ktp kpp]T is the vector of parameters, and y = [y1 y2 y3 y4]T = [ [I] xm DPn LCB]T is the vector of measurements. Upon the rearrangement of the output equations:

108 1 0 0 0.5 0.5 0 2 1 4 1 x 3 3 x1 1 d 0.5 x10.5 x2 x2 = 1 dt 1 0 4 0.5 x ( x + x ) x3 3 1 2 7 0.5 x7 4 1 0 0 3

= x1 G ( y) = y1 x2,0 (1 y2 ) x2 y2 x3 x2,0 y3 x 7 x2,0 y2 y4

Observations:

The parameters 1, 2 1 , and 4 1 are globally identifiable from the 3 3

set of measurements.

0.5

0.5

The parameters 2, 3, and 4 are not globally identifiable from the set of measurements.

The measurements y3 and y4 do not improve the parametric identifiability of the system.

The parameter 5 does not have any effect on the measurements; it is not identifiable due to the model structure.

109
Note that the dynamics of state x6 can be represented by a combination of states x2 and x7. The reaction networks R3 and R4 are therefore indistinguishable from one another from the set of outputs considered.

5.4

CONCLUSIONS

A class of chain-homopolymerization systems has been analyzed in terms of its global parametric identifiability. This study considers identifiability analyses as an integral part of parameter estimation efforts in polymerization systems. The advantages or disadvantages of including measurements were evaluated and presented. Globally identifiable parameters or lumped parameters were obtained, and the implications of their global identifiability properties were discussed. The study can be readily extended to more complex polymerization systems, and can also consider alternate models that are developed by different modeling techniques. The importance of the identifiability and distinguishability analyses in developing complex polymerization models is also re-emphasized, particularly for parameter estimation and modeling efforts in high-temperature polymerization systems. Lastly, the study also signifies the need for both non-redundant measurements and identifiable model structures, to avoid issues associated with different sets of models and parameters that can explain a given set of measurements.

110

CONCLUSIONS AND FUTURE DIRECTIONS

This study led to the following contributions:

Estimates of the Arrhenius parameters for the secondary reactions observed at high reaction temperatures (self-initiation, backbiting, scission and propagation of tertiary radicals)

First report on estimate of the activation energy of the self-initiation reaction Experimental evidence of short chain branch formation

The study also led to the following open questions:

How to isolate experimentally the self initiation species? How to calculate theoretically the energetically-favorable molecular structure of the self-initiation species?

Based on the current knowledge on high-temperature spontaneous solution polymerization of n-butyl acrylates (nBA), one could establish a fundamental and general understanding of the underlying pathways in high-temperature spontaneous solution and bulk polymerization of alkyl acrylate monomer family,

111
particularly of alkyl acrylate monomers that contain longer substituent alkyl groups.

Currently, experimental observations that shed light on the underlying mechanistic pathways in polymerization of 2-ethylhexyl acrylates (2-EHA) and noctyl acrylate (nOA) are still scattered and non-unified.

6.1

PRELIMINARY OBSERVATIONS IN HIGH-TEMPERATURE

SPONTANEOUS SOLUTION HOMO-POLYMERIZATION OF SEVERAL ALKYL ACRYLATE MONOMERS

When considering different alkyl acrylate monomers, a fundamental question that comes to mind is how the size of the substituent group affects the rate of polymerization. The molecular structures of the monomers are depicted in the figure below.

112

Figure 19. Different sizes of substituent groups in (I) n-butyl acrylate, (II) 2ethylhexyl acrylate, and (III) n-octyl acrylate

It has been reported based on pulsed-laser initiated polymerization studies that rate of chain propagation increases with increasing size of acrylate substituent (Beuermann, Paquet, McMinn and Hutchinson, 1996). This study was carried out at low temperatures (ca. 5-30C). In contrast at higher temperatures, preliminary experiments carried out at DuPont Marshall Laboratory indicated that resulting weight-average Molecular weight decreases, and % monomer conversion increases as size of acrylate substituent group increases (i.e., methyl, propyl, butyl, octyl, etc.).

113
Table 16. Preliminary observation of alkyl substituent size effect on resulting molecular weight and monomer conversion. Experiments were carried out in batch, in four hours, in methyl amyl ketone in the absence of thermal initiators [Grady, Quan and Soroush, 2003]

Monomer methyl acrylate ethyl acrylate n-butyl acrylate 2-ethylhexyl acrylate

Resulting Mw 53887 25755 9965 6898

%Conversion 85 94 98 99

While the increase in propagation rate could explain the increasing trend in %conversion, it does not explain the decreasing trend in molecular weight. One possible explanation is the increasing tendency for chain transfer events to take place with increasing alkyl substituent size. In a separate study, the propagation rate constant of 2-EHA was reported to be lower to that of nBA [Plessis et al., 2001].

114
6.2 RECENT STUDIES IN SOLUTION POLYMERIZATION OF 2ETHYLHEXYL ACRYLATE AND N-OCTYL ACRYLATE

Recent studies report scattered observations on occurrences of various reactions in different reaction conditions and temperature regimes. In a high-temperature bulk polymerization of n-octyl acrylate, self-initiation was reported at

temperatures above 190 C, with order of dependence of rate of polymerization with respect to monomer of 0.8 [Katime et al., 1988]. It was also reported that the dependence of the rate of polymerization on monomer concentration was observed to be in the order of 1.5 in the solution polymerization of 2-EHA at 50C [Kothandaraman et al., 1993]. n-octyl acrylate has also been used in atom transfer radical polymerization under microwave irradiation [Xu et al., 2003]. Systematic modeling study on polymerization of 2-EHA and nOA has not been reported in the open literature. It has been observed that n-octyl acrylate (nOA) polymerizes spontaneously at ca. 190C. While 2-EHA spontaneously

polymerizes at ca. 150C. We are thus interested in understanding how initial monomer content and temperature affects the tendency of the monomers (nOA and 2-EHA) to polymerize spontaneously.

Based on 1H-NMR spectra of nOA-Styrene copolymer, terminal double bonds and terminal solvent groups were not observed [Bisht et al., 2003]. On the other hand, chain transfer to polymer was observed based on
13

C-NMR spectra

measurement in the polymerization of 2-EHA in cyclohexane, with branching

115
levels much higher than those reported for nBA [Heatley et al., 2001]. Another study determined that the chain transfer to polymer event in emulsion polymerization of 2-EHA is preceded by an intra-molecular chain transfer reaction (backbiting), based on
13

C-NMR analyses [Plessis et al., 2001]. It was

suggested that the reaction scheme proposed for n-BA also applies to 2-EHA. Chain transfer to solvent (cyclohexane) was also observed in the pulse radiolysis-initiated polymerization of 2-EHA at room temperature [Takacs et al., 1999]. In this study, peak assignments of
13

C- and 1H-NMR spectra of poly-OA

and poly-2-EHA will be re-confirmed and reported.

6.3

MODELING AND PARAMETER ESTIMATION

The mechanistic modeling approach used in the preliminary study on hightemperature spontaneous solution polymerization of n-butyl acrylate will be used to study polymerization of 2-EHA and nOA. The following table summarizes known reaction mechanisms in nBA, nOA and 2-EHA solution polymerization, from low (ca. room temperature) to high temperature (ca. 200C).

116
Table 17. Summary of known reaction mechanisms in nBA, nOA and 2-EHA polymerization (low to high temperature)

Reaction Chain Initiation (Spontaneous) Chain Propagation Chain Termination Chain Transfer to Solvent Chain Transfer to Monomer Chain Transfer to Polymer (Intra) Chain Transfer to Polymer (Inter) Chain Fragmentation (Scission)

nBA X X X X X X

2-EHA X X X X

nOA X X X

It should be re-emphasized that the current knowledge on underlying mechanistic pathways in high-temperature spontaneous solution polymerization of 2-EHA and nOA is still very limited.

Hence the following are open research questions that can be pursued in the future:

The nature of branching in high-temperature solution and bulk polymerization of n-butyl acrylate, 2-ethylhexyl acrylate and n-octyl acrylate

117
The effect of size of alkyl substituent group on primary and secondary reaction pathways, and on the overall rate of polymerization Effect of temperature on the dynamics of polymer chain microstructure formation The rate of reactions of primary and secondary free-radical polymerization pathways

118

APPENDIX A

DERIVATION OF THE BATCH POLYMERIZATION PROCESS MODEL

A.1

RATE LAWS FOR SMALL MOLECULES.

Rate laws for dimers, oxygen radical, xylol radical, monomer, solvent and microstructural properties such as short chain-branching points (SCBs) and terminal double bonds (TDBs) were derived based on the postulated reaction network presented in Section 3. They are:

119
rM = 2kti [ M ]2 k p [ M ] P0 + P + P2 + PTDB + Pn 1 1 n =3

k pq [ M ] Qn ktm [ M ] Pn
n =3 n= 3

rS = kts [ S ] Pn
n =3

rR1OOR2 = kO2 [ R1OOR2 ] rSCB = k pq [ M ] Qn

n =3

rTDB = k Qn + k p [ M ]PTDB 1
n =3

rP0 = 2k R1OOR2 [ R1OOR2 ] + kts [ S ] Pn k p [ M ]P0

n= 3

rP1 = 2kti [ M ]2 + k sec [ M ]R0L k sec [ M ]R1L p p rP2 = k Qn + k p [ M ]PTDB + k p [ M ]P k p [ M ]P2 1 1

n= 3

rP1TDB = ktm [ M ] Pn k p [ M ]PTDB 1

n =3

A.2

RATE LAWS FOR `LIVE' AND `DEAD' POLYMER CHAINS.

Rate laws for live secondary propagating radicals, live tertiary propagating radicals and dead chains of length n, based on postulated reaction network presented in Section 3, take the following form:

120 rPn = k p [ M ]( Pn1 Pn ) + k pq [ M ]Qn1 (kbb + ktm [ M ] + kts [ S ]) Pn

sec sec sec sec 2 (ktc sec + ktd sec ) Pn Pm (ktc tert + ktd tert ) Pn Qm

m= 3

m =3

tert tert rQn = kbb Pn (k pq [ M ] + k ) Qn 2 (ktc tert + ktd tert ) Qn Qm m =3

sec sec (ktc tert + ktd tert ) Qn Pm

m= 3

sec rDn = (ktm [ M ] + kts [ S ]) Pn + k Qn+2 + ktc sec Pnm Pm

m =3

sec sec + ktc tert PnmQm + ktc tert Qnm Pm

m= 3

m =3

sec tert + ktc tert QnmQm + 2ktd sec Pn Pm

m =3

m =3

sec sec + 2ktd tert Pn Qm + 2ktd tert Qn Pm

m =3

m= 3

tert + 2ktd tert Qn Qm m =3

A.3

METHOD OF MOMENTS

The kth moment of the live and dead chains are defined as:

k = n k Pn
n =3

k = n k Qn
n =3

k = n k Dn
n =3

121
When the above moment definitions were applied to the rate laws, the following equations were obtained;
rM = 2kti [ M ]2 k p [ M ]( P0 + P + P2 + PTDB + 0 ) 1 1 k pq [ M ]0 ktm [ M ]0 rS = kts [ S ]0 rR1OOR2 = kO2 [ R1OOR2 ] rSCB = k pq [ M ]0 rTDB = k 0 + k p [ M ]PTDB 1 rP0 = 2k R1OOR2 [ R1OOR2 ] + kts [ S ]0 k p [ M ]P0 rP1 = 2kti [ M ]2 + k sec [ M ]R0L k sec [ M ]R1L p p rP2 = k 0 + k p [ M ]PTDB + k p [ M ]P k p [ M ]P2 1 1 rP1TDB = ktm [ M ]0 k p [ M ]PTDB 1

122
r0 = k p [ M ]P2 + k pq [ M ]0 (kbb + ktm [ M ] + kts [ S ])0
sec sec sec sec 2 (ktc sec + ktd sec )02 (ktc tert + ktd tert )00 tert tert r0 = kbb0 (k pq [ M ] + k ) 0 2 (ktc tert + ktd tert ) 02 sec sec (ktc tert + ktd tert )00 sec r0 = (ktm [ M ] + kts [ S ])0 + k 0 + (ktsecsec + 2ktd sec )02 c sec sec + (2ktc tert + 4ktd tert )00 tert tert + (ktc tert + 2ktd tert ) 02

r1 = k p [ M ]( P2 + 0 ) + k pq [ M ]1 (kbb + ktm [ M ] + kts [ S ])1

sec sec sec sec 2 (ktc sec + ktd sec )01 (ktc tert + ktd tert )10 tert tert r1 = kbb1 (k pq [ M ] + k ) 1 2 (ktc tert + ktd tert ) 10 sec sec (ktc tert + ktd tert )01 sec sec r1 = (ktm [ M ] + kts [ S ])1 + k 1 + (ktc sec + 2ktd sec )01 tert sec tert sec + (2ktc tert + 2ktd tert )(01 + 10 ) + (ktc tert + 2ktd tert ) 01

r2 = k p [ M ]( P2 + 0 + 21 ) + k pq [ M ]2 (kbb + ktm [ M ] + kts [ S ])2

sec sec sec sec 2 (ktc sec + ktd sec )02 (ktc tert + ktd tert )20 tert tert r2 = kbb2 (k pq [ M ] + k ) 2 2 (ktc tert + ktd tert ) 20 sec sec (ktc tert + ktd tert )02 sec r2 = (ktm [ M ] + kts [ S ])2 + k 2 + 2ktc sec (02 + 12 ) sec tert + 2ktc tert (02 + 211 + 20 ) + 2ktc tert (02 + 12 ) sec sec tert + 2ktd sec02 + 2ktd tert (02 + 20 ) + 2ktd tert 02

123
A.4 BATCH PROCESS MODEL

The rate laws were then incorporated in conservation (balance) equations for a batch polymerization reactor model. Perfect mixing and constant volume (negligible volume changes in reaction medium during reaction time) were assumed. Quasi steady-state assumption on the live chains is not made. The batch polymerization reactor model is of the form;

d[M ] [ M ](0) = [ M ]0 = rM , dt d[S ] [ S ](0) = [ M ]0 = rS , dt d [ R1OOR2 ] [ R1OOR2 ](0) = [ R1OOR2 ]0 = rR1OOR2 , dt d [TDB] [TDB](0) = 0 = rTDB , dt d [ SCB] [ SCB](0) = 0 = rSCB , dt d [ P0 ] [ P0 ](0) = 0 = rP0 , dt d[P ] 1 = rP1 , [ P ] ( 0) = 0 1 dt d [ P2 ] = rP2 , [ P2 ](0) = 0 dt d [ PTDB ] 1 = rP1TDB , [ PTDB ](0) = 0 1 dt dk k (0) = 0, k = 0,1, 2 = rk , dt d k = rk , k (0) = 0, k = 0,1, 2 dt d k = rk , k (0) = 0, k = 0,1, 2 dt

124
The measured variables are related to the state variables of the preceding model according to:

Mn = Mw =

1 MWm 0 2 MWm 1 [M ] [ M ]0 [TDB] 100 1 [ SCB] 100 1 [ S ]0 [ S ] 100 1

X = 1

TDBH = SCBH = TSGH =

125

BIBLIOGRAPHY

D. S. Achilias and C. Kiparissides, Toward the development of a general framework for modeling molecular-weight and compositional changes in free-radical co-polymerization reactions, Journal of Macromolecular Science-Reviews in Macromolecular Chemistry and Physics, C32(2): p. 183-234, 1992.

N. M. Ahmad, F. Heatley, P. A. Lovell, "Chain transfer to polymer in free-radical solution polymerization of n-butyl acrylate studied by NMR spectroscopy," Macromolecules, vol. 31, pp. 2822-2827, 1998.

L. Albertin, M. H. Stenzel, C. Barner-Kowollik, L. J. R. Foster and T. P. Davis, "Solvent and oxygen effects on the free radical polymerization of Solvent and oxygen effects on the free radical polymerization of 6-O-vinyladipoylD-glucopyranose, POLYMER, 46 (9): 2831-2835, 2005

J. M. Asua, S. Beuermann, M. Buback, P. Castignolles, B. Charleux, R. G. Gilbert, R. A. Hutchinson, J. R. Leiza, A. N. Nikitin, J. Vairon, A. M. van Herk, "Critically evaluated rate coefficients for free-radical polymerization, 5. Propagation rate coefficient for butyl acrylate," Macromolecular Chemistry and Physics, 205, 2151-2160, 2004

G. Arzamendi, C. Plessis and J. M. Asua, "Effect of the intramolecular chain transfer to polymer on PLP/SEC experiments of alkyl acrylates," Macromolecular Theory and Simulation, 12, 315-324, 2003

S. Audoly, G. Bellu, L. D'Angio, M. P. Saccomani, C. Cobelli, Global identifiability of nonlinear models of biological systems," IEEE Transactions on Biomedical Engineering, 48(1), 55--65, 2001

126
C.H. Bamford, W. G. Barb, Jenkins, A.D. and Onyon, D.F., Kinetics of vinyl polymerization by radical mechanisms, London: Butterworths, 1958

C.H. Bamford, H. Tompa, The calculation of molecular-weight distributions from kinetic schemes, Transactions of the Faraday Society, 50: p. 1097-1115, 1954

C. H. Bamford, H. Tompa, "Calculation of molecular-weight distributions from kinetic schemes," Journal of Polymer Science, 10, 345-350, 1953

C. Barner-Kowollik, P. Vana, T. P. Davis, The kinetics of free radical polymerization, in Handbook of Radical Polymerization, K. Matyjaszewski, T. P. Davis, Editors. John Wiley and Sons, Inc.: Hoboken, NJ. p. 187-262, 2002

R.

Bellman, K. J. Astrom, "On Biosciences, 7, 329-339, 1970

structural

identifiability,"

Mathematical

S. Beuermann, D. A. Paquet Jr., J. H. McMinn, R. A. Hutchinson, Determination of free-radical propagation rate coefficients of butyl, 2-ethylhexyl and dodecyl acrylates by pulsed-laser polymerization, Macromolecules, 29, 4206-4215, 1996

A. Ben-Zvi, K. McAuley, J. McLellan, Identifiability study of a liquid-liquid phasetransfer catalyzed reaction system," AIChE J., 50(10), 2493--2501, 2004

S. Beuermann, M. Buback, "Rate coefficients of free-radical polymerization deduced from pulsed laser experiments," Progress in Polymer Science, 27, 191-254, 2002

E. Bezemer, S. C. Rutan, "Multivariate curve resolution with nonlinear fitting of kinetic profiles," Chemometrics and Intelligent Laboratory Systems, 59, 19-31, 2001

127

R. Bindlish, J. B. Rawlings, R. E. Young, "Parameter estimation for industrial polymerization processes," AICHE J., 49(8), 2071-2078, 2003

H. S. Bisht, S. S. Ray, A. K. Chatterjee, Conventional and atom transfer radical copolymerization of n-octyl acrylate-styrene: chemoselectivity and monomer sequence distribution by 1H-NMR, European Polymer Journal, 39, 1413-1420, 2003

L. Blavier, J. Villermaux, Free radical polymerization engineering. II. Modeling of homogeneous polymerization of styrene in a batch reactor, influence of initiator, Chemical Engineering Science, 39((1)): p., 101-10, 1984

E. Brand, M. Stickler, G. Meyerhoff, Spontaneous polymerization of methyl methacrylate. 3. Reaction behavior of the unsaturated dimer upon polymerization," Makromolekulare Chemie, 181, 913-921, 1980

M. Buback, M. Egorov, A. Feldermann, "Chain-length dependence of termination rate coefficients in acrylate and methacrylate homopolymerizations investigated via the SP-PLP technique," Macromolecules, 37, 1768-1776, 2004

M. Buback, M. Egorov, R. G. Gilbert, V. Kaminsky, O. F. Olaj, G. T. Russell, P. Vana, G. Zifferer, "Critically evaluated termination rate coefficients for freeradical polymerization, 1 - The current situation," Macromolecular Chemistry and Physics, 203, 2570-2582, 2002

M. Busch, Modeling kinetics and structural properties in high-pressure fluidphase polymerization, Macromolecular Theory and Simulations, 10(5): p. 408-429, 2001

128
M. Busch, M. Muller, "Simulating acrylate polymerization reactions: toward improved mechanistic understanding and reliable parameter estimates," Macromolecular Symposia, 206, 399-418, 2004

J. D. Campbell, M. Morbidelli, F. Teymour, A theoretical and experimental investigation of high temperature polymerization, DECHEMA Monographs - WILEY-VCH Verlag GmbH, 137: p. 191-202, 2001

J. D. Campbell, F. Teymour, M. Morbidelli, High temperature free radical polymerization. 1. Investigation of continuous styrene polymerization, Macromolecules, 36(15): p. 5491-5501, 2003

J. D. Campbell, F. Teymour, M. Morbidelli, High temperature free radical polymerization. 2. Modeling continuous styrene polymerization, Macromolecules, 36(15): p. 5502-5515, 2003

G. Cao, Z. Zhu, M. Zhang, W. Yuan, "Kinetics of butyl acrylate polymerization in a starved feed reactor," Journal of Applied Polymer Science, 93, 15191525, 2004

R. Carlson, A. K. Axelsson, A. Nordahl, "Optimization in organic synthesis. An approach to obtaining kinetic information by sequential response surface modeling. Outline of the principles," Journal of Chemometrics, 7, 341-367, 1993

B. Cerinski, J. Jelencic, "Modeling of high-pressure ethylene polymerization. I. Kinetic parameters of oxygen initiation," Journal of Applied Polymer Science, 83, 2043-2051, 2002

E. G. Chatzi, O. Kammona, C. Kiparissides, Recent developments in hardware sensors for the on-line monitoring of polymerization reactions, Journal of Macromolecular Science Reviews in Macromolecular Chemistry and Physics, C39, 57-134, 1999

129

J. Chiefari, J. Jeffery, R. T. A. Mayadunne, G. Moad, E. Rizzardo, S. H. Thang, Chain transfer to polymer: a convenient route to Macromonomers, Macromolecules, 32, 7700-7702, 1999

G. Clouet, P. Chaumont, P. Corpart, Studies on bulk-polymerization of methylmethacrylate .1. Thermal polymerization," J. Pol. Sci. Part A Polymer Chemistry, 31, 2815-2824, 1993

M. L. Coote, T. P. Davis, Co-polymerization kinetics, in Handbook of Radical Polymerization, K. Matyjaszewski, T. P. Davis, Editors. John Wiley and Sons, Inc.: Hoboken N J. p. 263-300, 2002

M. F. Cunningham, R. Hutchinson, Industrial applications and processes, in Handbook of Radical Polymerization, K. Matyjaszewski, T. P. Davis, Editors, John Wiley and Sons Inc.: Hoboken, NJ. p. 333-360, 2002

L.

Denis-Vidal, G. Joly-Blanchard, An easy to check criterion for (un)identifiability of uncontrolled systems and its aplications," IEEE Transactions on Automatic Control, 45(4), 768-771, 2000

F. Fenouillot, J. Terrisse, T. Rimlinger, "Thermal polymerization of methyl methacrylate at high temperature," International Polymer Processing, 13, 154-161, 1998

F. Fenouillot, J. Terrisse, T. Rimlinger, Experimental study and simulation of the polymerization of methyl methacrylate at high temperature in a continuous reactor, Journal of Applied Polymer Science, 2001. 79(11): p. 2038-2051.

M. Fernandez-Garcia, M. Fernandez-Sanz, E. L. Madruga, "A kinetic study of butyl acrylate free radical polymerization in benzene solution," Macromolecular Chemistry and Physics, 201, 1840-1845, 2000

130
J. M. R. Fontoura, A. F. Santos, F. M. Silva, M. K. Lenzi, E. L. Lima, J. C. Pinto, "Monitoring and control of styrene solution polymerization using NIR spectroscopy," J. Applied Polymer Science, vol. 90, pp. 1273-1289, 2003

S. Garg, S. K. Gupta, "Multiobjective optimization of a free radical bulk polymerization reactor using genetic algorithm," Macromolecular Theory and Simulation, vol. 8, pp. 46-53, 1999

C. Gentric, F. Pla, J. P. Corriou, "Experimental study of the nonlinear geometric control of a batch emulsion polymerization reactor," Computers and Chemical Engineering, vol. 21, pp. S1043-S1048, 1997

M. C. Grady, C. Quan, M. Soroush, "Thermally initiated polymerization process for acrylates," U.S. Pat. Appl. Publ., 7pp., US20050003094, 2005

M. C. Grady, W. J. Simonsick, R. A. Hutchinson, "Studies of higher temperature polymerization of n-Butyl Methacrylate and n-Butyl Acrylate," Macromolecular Symposia, 182, 149-168, 2002

P.D. Iedema, M. Wulkow, H. C. J. Hoefsloot, Modeling molecular weight and degree of branching distribution of low-density polyethylene, Macromolecules, 33(19): p. 7173-7184, 2000

T. Hirano, R. Takeyoshi, M. Seno, T. Sato, Chain-transfer reaction in the radical polymerization of di-n-butyl itaconate at high temperatures, Journal of Polymer Science Part a-Polymer Chemistry, 40(14): p. 2415-2426, 2002

A. W. Hui, A. E. Hamielec, Thermal polymerization of styrene at high conversions and temperatures. Experimental study, Journal of Applied Polymer Science, 16, 749-69, 1972

131
R. A. Hutchinson, Modeling of chain length and long-chain branching distributions in free-radical polymerization, Macromolecular Theory and Simulations, 10(3), 144-157, 2001

S. Hoppe, A. Renken, Modelling of the free radical polymerisation of methylmethacrylate up to high temperature, Polymer Reaction Engineering, 6(1): p. 1-39, 1998

T. Hirano, B. Yamada, Macromonomer formation by sterically hindered radical polymerization of methyl acrylate trimer at high temperature, Polymer, 44(2): p. 347-354, 2003

C. D. Immanuel, F. J. Doyle III, "Open-loop control of particle size distribution in semi-batch emulsion copolymerization using a genetic algorithm," Chemical Engineering Science, vol. 57, pp. 4415-4427, 2002

J. A. Jacquez, Design of experiments," J. Franklin Institute, 335(2), 259-279, 1998

J. Jaumot, P. J. Gemperline, A. Stang, Non-negativity constraints for elimination of multiple solutions in fiting of multivariate kinetic models to spectroscopic data," Journal of Chemometrics, 19, 97-106, 2005

O. Kammona, E. G. Chatzi, C. Kiparissides, Recent developments in hardware sensors for the on-line monitoring of polymerization reactions," J.M.S.Rev. Macromol. Chem. Phys., C39(1), 57--134, 1999

A. R. Karagoz, H. Hapoglu, M. Alpbaz, Generalized minimum variance control of optimal temperature profiles in a polystyrene polymerization reactor, Chemical Engineering and Processing, vol. 39, pp. 253-262, 2000.

132
I. A. Katime, T. Nuno, Determination of kinetic constants for bulk polymerization of octyl acrylate at different temperatures, Makromol. Chem., Macromol. Symp., 20/21, 99-104, 1988

A. Keramopoulos, C. Kiparissides, Development of a comprehensive model for diffusion-controlled free-radical co-polymerization reactions, Macromolecules, 35(10): p. 4155-4166, 2002.

K. Konstadinidis, D. S. Achilias, C. Kiparissides, Development of a unified mathematical framework for modeling molecular and structural-changes in free-radical homo-polymerization reactions, Polymer, 33(23): p. 50195031, 1992

H.

Kothandaraman, M. S. Devi, N. S. Kumar, Radical initiated homopolymerization of 2-octyl acrylate, J. Indian Chem. Soc., 70, 270271, 1993

C. Kotoulas, A. Krallis, P, Pladis, C. Kiparissides, A comprehensive kinetic model for the combined chemical and thermal polymerization of styrene up to high conversions, Macromolecular Chemistry and Physics, 204(10): p. 1305-1314, 2003

T. Kourti, P. Nomikos, J. F. Macgregor, Analysis, monitoring and fault-diagnosis of batch processes using multiblock and multiway PLS, Journal of Process Control, 5, 277-284, 1995

R. S. Lehrle, A. Shortland, A study of the purification of methyl methacrylate suggests that the 'thermal' polymerisation of this monomer is initiated by adventitious peroxides,, European Polymer Journal, 24, 425-429, 1988

R. J. Li, M. A. Henson, M. J. Kurtz, Selection of model parameters for off-line parameter estimation, IEEE Transactions On Control Systems Technology, 12, 402-412, 2004

133

J. Lingnau, M. Stickler, G. Meyerhoff, The spontaneous polymerization of methyl methacrylate. IV. Formation of cyclic dimers and linear trimers, European Polymer Journal, 16, 785-791, 1980

J. Lingnau, M. Stickler, G. Meyerhoff, The spontaneous polymerization of methyl methacrylate. 6. Polymerization in solution: participation of transfer agents in the initiation reaction, Polymer, 24, 1473-1478, 1983

J. Lingnau, G. Meyerhoff, Spontaneous polymerization of methyl methacrylate. 8. Polymerization kinetics of acrylates containing chlorine atoms, Macromolecules, 17, 941-945, 1984

J. Lingnau, G. Meyerhoff, The spontaneous polymerization of methyl methacrylate, 7. External heavy atom effect on the initiation, Makromolekulare Chemie, 185, 587-600, 1984

L. Ljung, T. Glad, On global identifiability for arbitrary model parametrizations, Automatica, 30(2), 265-276, 1994

J. F. Lutz, K. Matyjaszewski, Nuclear magnetic resonance monitoring of chainend functionality in the atom transfer radical polymerization of styrene, Journal of Polymer Science Part A Polymer Chemistry, 43, 897-910, 2005

P. Lorenzini, M. Pons, J. Villermaux, Free-radical polymerization engineering. III. Modeling homogeneous polymerization of ethylene: mathematical model and new method for obtaining molecular-weight distribution, Chemical Engineering Science, 47, 3969-3980, 1992

P. Lorenzini, M. Pons, J. Villermaux, Free-radical polymerization engineering .4. Modeling homogeneous polymerization of ethylene - determination of

134
model parameters and final adjustment of kinetic coefficients, Chemical Engineering Science, 47(15-16): p. 3981-3988, 1992

P. A. Lovell, T. H. Shah, F. Heatley, Chain transfer to polymer in emulsion polymerization of n-butyl acrylate studied by 13C-NMR spectroscopy and GPC, Polymer Communications, 32, 98-103, 1991

B. Ma, P. J. Gemperline, E. Cash, M. Bosserman, E. Comas, Characterizing batch reactions with in-situ spectroscopic measurements, calorimetry and dynamic modeling, Journal of Chemometrics, 17, 470-479, 2003

S. Maeder, R. G. Gilbert, Measurement of transfer constant for butyl acrylate free radical polymerization, Macromolecules, 31, 4410-4418, 1998

G. Maria, A review of algorithms and trends in kinetic model identification for chemical and biochemical systems, Chem. Biochem. Eng., 18(3), 195222, 2004

F. R. Mayo, The dimerization of styrene, Journal of American Chemical Society, 90, 1289-1295, 1968

E. F. McCord, W. H. Shaw Jr., R. A. Hutchinson, Short-chain branching structures in ethylene copolymers prepared by high-pressure free-radical polymerization: an NMR analysis, Macromolecules, 30, 246-256, 1997

N. T. McManus, A. Penlidis, M. A. Dube, Copolymerization of alpha-methyl styrene with butyl acrylate in bulk, Polymer, 43, 1607-1614, 2002

K. Nagasubramanian, W. W. Graessley, Continuous reactors in free radical polymerization with branching. I. Theoretical aspects and preliminary considerations, Chemical Engineering Science, 25(10), 1549-58, 1970

135
A. Negiz, A. Cinar, Statistical monitoring of multivariable dynamic processes with state-space models, AIChE Journal, 43, 2002-2020, 1997

D. Neogi, C. E. Schlags, Multivariate statistical analysis of an emulsion batch process, Industrial and Engineering Chemistry Research, 37, 3971-3979, 1998

A. M. North, The kinetics of free radical polymerization, Pergamon Press Ltd.. 1966

Y. Ni, C. Huang, S. Kokot, Application of multivariate calibration and artificial neural networks to simultaneous kinetic-spectrophotometric determination of carbamate pesticides, Chemometrics and Intelligent Laboratory Systems, 71, 177-193, 2004

A. N. Nikitin, P. Castignolles, B. Charleux, J. P. Vairon, Determination of propagation rate coefficient of acrylates by pulsed-laser polymerization in the presence of intramolecular chain transfer to polymer, Macromolecular Rapid Communications, 24, 778-782, 2003

P. Nising, T. Meyer, R. Carloff, M. Wicker, Thermal initiation of MMA in high temperature radical polymerizations, Macromolecular Materials and Engineering, 290, 311-318, 2005

G. G. Odian, Principles of polymerization - 3rd ed., John Wiley and Sons, Inc., Hoboken, NJ, USA, 1991

G. Papavasiliou, I. Birol, F. Teymour, Calculation of molecular weight distributions in non-linear free-radical polymerization using the numerical fractionation technique, Macromolecular Theory and Simulations, 11(5): p. 533-548, 2002

136
A. N. F. Peck, R. A. Hutchinson, M. C. Grady, Branching and scission reactions in high temperature acrylate polymerizations, Polymer Preprints, 43, 154155, 2002

A. N. F. Peck, R. A. Hutchinson, Secondary reactions in the high-temperature free-radical polymerization of butyl acrylate, Macromolecules, 37, 59445951, 2004

P. Pladis, C. Kiparissides, A comprehensive model for the calculation of molecular weight long-chain branching distribution in free-radical polymerizations, Chemical Engineering Science, 53(18), 3315-3333, 1998

C. Plessis, G. Arzamendi, J. M. Alberdi, A. M. van Herk, J. R. Leiza, J. M. Asua, Evidence of branching in poly(butyl acrylate) produced in pulsed-laser polymerization experiments, Macromolecular rapid communications, 24, 173-177, 2003

A. L. Polic, L. M. F. Lona, T. A. Duever and A. Penlidis, A protocol for the estimation of parameters in process models: case studies with polymerization scenarios, Macromolecular Theory and Simulations, 13, 115-132, 2004

J. G. Qin, W. P. Guo, Z. Zhang, A kinetic study on bulk thermal polymerization of styrene, Polymer, 43(26): p. 7521-7527, 2002

J. G. Qin, W. P. Guo, Z. Zhang, Modeling of the bulk free radical polymerization up to high conversion - three stage polymerization model. II. Numberaverage molecular weight and apparent initiator efficiency, Polymer, 43(18): p. 4859-4867, 2002

J. G. Qin, W. P. Guo, Z. Zhang, Modeling of the bulk free radical polymerization up to high conversion - three stage polymerization model I. Model

137
examination and apparent reaction rate constants, Polymer, 43(4): p. 1163-1170, 2002

J. G. Qin, H. Li, Z. Zhang, Modeling of high-conversion binary copolymerization, Polymer, 44(8): p. 2599-2604, 2003

C. L. Quan, M. Soroush, M. C. Grady, J. E. Hansen, W. J. Simonsick, Hightemperature homopolymerization of ethyl acrylate and n-butyl acrylate: polymer characterization," Macromolecules, 38, 7619--7628, 2005

P. V. T. Raghuram, U. S. Nandi, Studies on the polymerization of ethyl acrylate. II. Chain transfer studies, Journal of Polymer Science Part A1, 7, 23792385, 1969

F. S. Rantow, M. Soroush, M. C. Grady, G. A. Kalfas, Spontaneous polymerization and chain microstructure evolution in high temperature polymerization of n-butyl acrylate, Polymer, 47(4), 1423-1435, 2006

F. S. Rantow, M. Soroush, M. C. Grady, Optimal control of a high-temperature semi-batch polymerization reactor, Proceedings of American Control Conference, 5, 3102-3107, 2005

W. H. Ray, On the mathematical modeling of polymerization reactors, Journal of Macromolecular Science, 8, 1-56, 1972

W. H. Ray, Modelling of polymerization phenomena, Ber. Bunsenges. Phys. Chem., 90: p. 947-955, 1986

W. H. Ray, T. L. Douglas, E. W. Godsalve, Molecular weight distributions in copolymer systems. II. Free radical co-polymerization, Macromolecules, 4: p. 166-174, 1971

138
H. Sawada, Thermodynamics of polymerization, New York, NY: Marcel Dekker, 1976

M. Skowronski, X. M. Meyer, M. Pons, X. Joulia, Dynamic simulation of radical polymerisation reactors in an object oriented environment, Recents Progres en Genie des Procedes, 13(69): p. 89-96, 1999

M. Soroush, Nonlinear state-observer design with application to reactors, Chemical Engineering Science, 52, 387-404, 1997

B. Srinivasan, S. Palanki, D. Bonvin, Dynamic optimization of batch processes I. Characterization of the nominal solution, Computers Chem. Eng, vol. 27, pp. 1-26, 2003

B. Srinivasan, D. Bonvin, E. Visser, S. Palanki, Dynamic optimization of batch processes II. Role of measurements in handling uncertainty, Comp. Chem. Eng., vol. 27, pp. 27-44, 2002

M. Stickler, G. Meyerhoff, The spontaneous thermal polymerization of methyl methacrylate. 5. Experimental study and computer simulation of the high conversion reaction at 130C, Polymer, 22, 928-933, 1981

M. Stickler, G. Meyerhoff, Thermal polymerization of methyl methacrylate, 1. Polymerization in bulk, Makromolekulare Chemie, 179, 2729-2745, 1978

E. Takacs, S. S. Emmi, L. Wojnarovits, Pulse Radiolysis Study on Polymerization Kinetics in Organic Solvent: 2-Ethylhexyl Acrylate, Nuclear Instruments and Methods in Physics Research B, 151, 346-349, 1999

S. Tatiraju, M. Soroush, B. A. Ogunnaike, Multirate nonlinear state estimation with application to a polymerization reactor, AIChE Journal, 45, 769-780, 1999

139

F. Teymour, J. D. Campbell, Analysis of the dynamics of gelation in polymerization reactors using the numerical fractionation technique, Macromolecules, 27(9): p. 2460-2469, 1994

Y. Tian, J. Zhang, J. Morris, Modeling and optimal control of a batch polymerization reactor using a hybrid stacked recurrent neural network model, Ind. Eng. Chem. Res., vol. 40, pp. 4525-4535, 2001

H. Tobita, Molecular weight development during simultaneous chain scission, long-chain branching and crosslinking, 2 - free-radical polymerization, Macromolecular Theory and Simulations, 12(1): p. 32-41, 2003

H. Tobita, Simultaneous long-chain branching and random scission: I. Monte carlo simulation, Journal of Polymer Science Part B-Polymer Physics, 39(4): p. 391-403, 2001

H. Tobita, Molecular-weight distribution in free-radical polymerization with longchain branching, Journal of Polymer Science Part B-Polymer Physics, 31(10): p. 1363-1371, 1993

H. Tobita, Molecular weight development during simultaneous chain scission, long-chain branching and crosslinking, 1 - general matrix formula, Macromolecular Theory and Simulations, 12(1): p. 24-31, 2003

S. Vajda, H. Rabitz, Identifiability and distinguishability of general reaction systems," J. Phys. Chem., 98, 5265--5271, 1994

M.H.C.M. van Boxtel, M. Busch, A general model describing molecular weight distribution and branching indices in co-polymerizations demonstrated by the high-pressure free-radical co-polymerization of ethene and methyl acrylate, Macromolecular Theory and Simulations, 10(1): p. 25-37, 2001

140
M.H.C.M. van Boxtel, M. Busch, S. Lehmann, Long-chain and short-chain branches in ethene-methyl acrylate copolymers studied by quantitative C13 NMR spectroscopy, Macromolecular Chemistry and Physics, 201(3): p. 313-319, 2000

P. Vana, T. P. Davis, C. Barner-Kowollik, End-group analysis of polymers by electrospray ionization mass spectrometry: 2-methyl-1-[4(methylthio)phenyl]-2-morpholinopropan-1-one initiated free-radical photopolymerization, Australian Journal of Chemistry, 55, 315-318, 2002

D. D. Vecchio, R. M. Murray, Observability and local observer construction for unknown parameters in linearly and nonlinearly parameterized systems," Proc. of American Control Conference, 2003

R. Vidal, A. Chiuso, S. Soatto, Observability and identifiability of jump linear systems,'' Proc. of 41st IEEE Conference on Decision and Control, 36143619, 2002

J. Villermaux, L. Blavier, M. Pons, A new method for modeling free radical polymerization and predicting polymer quality, in Polymer reaction engineering: influence of reaction engineering on polymer properties, K. H. Reichert, W. Geiseler, eds., pp. 1-20, 1983

J. Villermaux, L. Blavier, M. Pons, A new method for modeling free radical polymerization and predicting polymer quality, Polym. React. Eng.: Influence React. Eng. Polym. Prop., Workshop Proc., 1983

J. Villermaux, L. Blavier, Free radical polymerization engineering. I. A new method for modeling free radical homogeneous polymerization reactions, Chemical Engineering Science, 39, 87-99, 1984

E. Walter, L. Pronzato, On the identifiability and distinguishability of nonlinear parametric models," Mathematics and Computers in Simulation, 42, 125134, 1996

141

F. S. Wang, T. L. Shieh, Extension of iterative dynamic programming to multiobjective optimal control problems, Ind. Eng. Chem. Res., vol. 36, pp. 2279-2286, 1997

R. X. E. Willemse, A. M. van Herk, E. Panchenko, T. Junkers, M. Buback, PLPESR monitoring of midchain radicals in n-butyl acrylate polymerization, Macromolecules, 38, 5098-5103, 2005

M. Wulkow, The simulations of molecular weight distributions in polyreaction kinetics by discrete galerkin methods, Macromolecular Theory and Simulations, 5, 393-416, 1996

Z. W. Wicks Jr., F. N. Jones, S. P. Pappas, Organic coatings science and technology, 1999, Hoboken, NJ: John Wiley and Sons, Inc.

R. X. E. Willemse, A. M. van Herk, E. Panchenko, T. Junkers, M. Buback, PLPESR monitoring of midchain radicals in n-butyl acrylate polymerization, 38, 5098-5103, 2005

W. Xu, X. Zhu, Z. Cheng, G. Chen, J. Lu, Atom transfer radical polymerization of n-octyl acrylate under microwave irradiation, European Polymer Journal, 39, 1349-1353, 2003

K. Z. Yao, B. M. Shaw, B. Kou, K. B. McAuley, D. W. Bacon, Modeling ethylene/butene copolymerization with multi-site catalysts: parameter estimability and experimental design," Polymer Reaction Engineering, 11(3), 563-588, 2003

R.C.M. Zabisky, W. M. Chan, P. E. Gloor, A. E. Hamielec, A kinetic-model for olefin polymerization in high-pressure tubular reactors - a review and update, Polymer, 33(11): p. 2243-2262, 1992

142
N. Zambare, M. Soroush, B. A. Ogunnaike, A method of robust multi-rate state estimation, J. Proc. Control, 13 (4): 337-355, 2003

J. Zhang, A reliable neural network model based optimal control strategy for a batch polymerization reactor, Ind. Eng. Chem. Res., vol. 43, pp. 10301038, 2004

Z. L. Zhu, W. Z. Cheng, Y. Zhao, Iterative target transformation factor analysis for the resolution of kinetic-spectral data with an unknown kinetic model, Chemometrics and Intelligent Laboratory Systems, 64, 157-167, 2002

A. Zogg, U. Fischer, K. Hungerbuhler, "A new approach for a combined evaluation for calorimetric and online infrared data to identify kinetic and thermodynamic parameters of a chemical reaction, Chemometrics and Intelligent Laboratory Systems, 71, 165-176, 2004

143

144

VITA

FELIX S. RANTOW

PERSONAL Birth date: 16 July 1980 Birth place: Jakarta, Indonesia Citizenship: Indonesian

EDUCATION Ph.D. Candidate, Drexel University Chemical Engineering, September 2002 to September 2006 Thesis: Mechanistic Modeling and Model-Based Studies in Spontaneous Solution Polymerization of Alkyl Acrylate Monomers Advisors: Prof. Masoud Soroush and Dr. Michael C. Grady M.S., Drexel University Chemical Engineering, September 2002 to December 2005 B. S., Purdue University Chemical Engineering, August 1998 to May 2002 Undergraduate Research Project: Development of a JAVA-XML based process scheduling language Advisors: Dr. Seza Orcun and Prof. Joseph F. Pekny

145
AWARDS Graduate Student Research Award, Drexel University, 2006 Best Paper Presentation Award, American Control Conference, 2005

EXPERIENCE DuPont Marshall Laboratory, Process Engineering Group, 2004 to 2006 Visiting Research Assistant. Designed and analyzed batch solution polymerization experiments. Chemtura Quality Assurance, Summer 2001 Database System Developer. Designed and developed an online database system for document request. Purdue University Civil Engineering Department, 2000 to 2002 Undergraduate Research Assistant. Designed and developed a visual basic-based decision support system for choosing trenchless technology equipments. Designed and developed an online roadmap application for the Indiana department of transportation. Purdue Computer Science Department, Spring 1999 Lab Instructor. Teach introductory JAVA programming language. Present lab material.

JOURNAL PUBLICATIONS F. S. Rantow, M. Soroush, M. C. Grady, G. A. Kalfas, Spontaneous Polymerization and Chain Microstructure Evolution in High-Temperature Polymerization of n-Butyl Acrylate, 47(4), 1431-1443, Polymer, 2006

F. S. Rantow, M. Soroush, M. C. Grady, A Reduced-Order Kinetic Model for Process Monitoring of Polymer Microstructure Spectroscopic Data, to be submitted to Journal of Chemometrics, 2006

146

F. S. Rantow, M. Soroush, M. C. Grady, Global Parameter Identifiability of Affine-In-Parameter Polymerization Mechanistic Models, in preparation

M. Soroush, F. S. Rantow, Y. Dimitratos, Control Quality Loss in Analytical Control of Input-Constrained Processes, Accepted for publication in Industrial and Engineering Chemistry Research, 2006

CONFERENCE PROCEEDINGS F. S. Rantow, M. Soroush, M. C. Grady, Global Parametric Identifiability of Mechanistic Models in Chain Polymerization, Proceedings of American Control Conference, 3068-3073, 2006

F. S. Rantow, M. Soroush, M. C. Grady, Optimal Control of a High-Temperature Semi-Batch Polymerization Reactor, Proceedings of American Control Conference, 3102-3107, 2005

F. S. Rantow, M. Soroush, M. C. Grady, Resin Design and Optimization: HighTemperature (140-200C) n-Butyl Acrylate Polymerization, Proceedings of Foundations of Computer-Aided Process Design, 577-580, 2004

PRESENTATIONS American Control Conference 2006. Global Parametric Identifiability of Mechanistic Models in Chain Polymerization AICHE Annual Meeting 2005. Reaction Network Identification in HighTemperature Polymerization of n-Butyl Acrylate: A Simultaneous Parameter Estimation Approach AICHE Annual Meeting 2005. Model for Polymer Microstructure Monitoring and Control in Polymerization of Acrylates American Control Conference 2005. Optimal Control of a High-Temperature Semi-Batch Polymerization Reactor

147
AICHE Annual Meeting 2004. High-Temperature Acrylate Polymerization: Decentralized Parameter Estimation and Multi-Rate State Estimation Foundations of Computer-Aided Process Design 2004. Resin Design and Optimization: High-Temperature (140-200C) n-Butyl Acrylate Polymerization

CERTIFICATION Safety and Chemical Engineering Education (SACHE) Faculty Workshop. What Chemical Engineering Students Need To Know, Rohm and Haas Company Engineering Division, Croydon, PA, September 2005

EXTRACURRICULAR AND PROFESSIONAL AFFILIATIONS President. Drexel University Chemical Engineering Graduate Student Association, 2004 to 2005 Member, American Institute of Chemical Engineers, 2001-present Cantor. St. Agatha-St. James Church, 2003-2005 Social Events Committee. Indonesian Students Association, 1999-2002 Intramural Soccer Captain. Drexel University, 2004