You are on page 1of 17

Computers and Chemical Engineering 30 (2006) 14471463

Measurement and control of polymerization reactors


John R. Richards , John P. Congalidis
E. I. du Pont de Nemours and Company, Experimental Station, Wilmington, Delaware 19880, USA Received 9 February 2006; received in revised form 2 May 2006; accepted 16 May 2006 Available online 25 July 2006

Abstract The measurement and control of polymerization reactors is very challenging due to the complexity of the physical mechanisms and polymerization kinetics. In these reactors many important variables, which are related to end-use polymer properties, cannot be measured on-line or can only be measured at low sampling frequencies. Furthermore, end-use polymer properties are related to the entire molecular weight, copolymer composition, sequence length, and branching distributions. This paper surveys the instrumentation technologies, which are of particular interest in polymerization reactors with emphasis on, for example, measurement of viscosity, composition, molecular weight, and particle size. This paper presents a hierarchical approach to the control system design and reviews traditional regulatory techniques as well as advanced control strategies for batch, semibatch, and continuous reactors. These approaches are illustrated by focusing on the control of a commercial multiproduct continuous emulsion polymerization reactor. Finally, the paper captures some of the trends in the polymer industry, which may impact future development in measurement and reactor control. 2006 Elsevier Ltd. All rights reserved.
Keywords: Process control; Polymerization reactors; Solution copolymerization; Emulsion copolymerization; Mathematical modeling; Reactor control

1. Introduction Consistent polymer properties are of paramount importance to end-user manufacturers who must produce the polymer in its nal form and shape for the intended application. These properties are the result of complex polymer architecture and composition formed in reaction and perhaps further inuenced in isolation and extrusion processes. Producing consistent, uniform, and in-specication polymer for the end-user are the tasks of the polymer process measurement and control systems. Polymer processes, whether batch or continuous, rarely run under exactly specied conditions; disturbances move the process away from desired trajectories. However, in order to operate such processes safely and in order to set the characteristics of the products optimally, a set of process manipulated variables must be kept constant or systematically modied over the duration of the reaction or in the course of the various reaction steps. Ray, Soares, and Hutchinson (2004) recently reviewed the main developments of polymer reaction engineering from the early days of polymer science to the current challenges of today.

Corresponding author. Tel.: +1 302 695 4059; fax: +1 302 695 8805. E-mail address: john.r.richards@usa.dupont.com (J.R. Richards).

The purpose of this contribution is to discuss the various measurement and control techniques of importance to engineers and scientists designing and operating polymer reactors and associated equipment. We have attempted to summarize and update the information already provided in our previous work (Congalidis & Richards, 1998; Richards & Congalidis, 2005; Richards & Schnelle, 1988). The framework for our discussion is the hierarchical approach summarized in Fig. 1, which has proved very useful in the successful application of process control in a complex industrial environment as shown in an earlier review by Richards and Schnelle (1988). It has been our own experience and learning that the same hierarchical approach is particularly important in the control of polymer reactors. Process knowledge, which is usually captured in an experimentally validated mathematical model, is the cornerstone of a successful control strategy. This is particularly true for polymerization reactors, where the in-depth knowledge of process operation in terms of the effect of operating variables on polymer properties can be used to great advantage in the design of the control system and can result in a much more straightforward (and therefore easy to maintain) strategy than would have been possible otherwise. This point will be illustrated in Section 3 and particularly by using the examples referred to in Figs. 2 and 3.

0098-1354/$ see front matter 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.compchemeng.2006.05.021

1448

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

Fig. 1. The process control hierarchy (Richards & Congalidis, 2005).

The use of polymerization reactor modeling in conjunction with control design was discussed by the authors in an earlier publication (Congalidis, Richards, & Ray, 1989). Process knowledge together with the appropriate sensors, transmitters, and analyzers are the prerequisites for the design of the basic control system to regulate pressure, temperature, level, and ow (PTLF). Only when the elements of the regulatory control system are in place and are properly designed and maintained can the control engineer attempt, in increasing order of complexity, the implementation of more advanced regulatory control strategies, multivariable model based control algorithms, and on-line scheduling and optimization strategies to compute set points for the regulatory controls. In many instances advanced control applications have failed in an industrial environment

not because the algorithms were necessarily faulty but because the basic regulatory control system performed poorly, either because of inadequate design (leading to operation in an openloop mode) or because one of the critical measurements (i.e. a process analyzer) was poorly maintained. In other instances the basic regulatory control may have been in place but some of the elements of advanced regulatory control (for example cascade control and ratio control) were not being implemented resulting in degradation of reactor performance in terms of consistent polymer properties. In this contribution we have attempted to cover control topics that we believe should be of interest to a wide spectrum of engineers and scientists in the polymer industry. We have therefore elected to discuss all elements of the process control hierarchy as they apply to polymer reactor control fully realizing that the lower levels of control would be obvious to the academic community or to experienced industrial practitioners. 2. Measurement techniques The measurement technique to be chosen is principally determined by the measured quantity and by the accuracy by which the variable must be measured. The measuring instrument produces a signal, which must be transformed in such a way that it can be registered by an indicator or recorder and further processed. This requirement is fullled directly by some measuring methods; however, in most cases a measurement transmitter is operated between the sensor and the measurement device. Electrical signals are much more commonly used today than pneumatic signals. We will concentrate our discussions on the

Fig. 2. Solution copolymerization with recycle loop (Congalidis et al., 1989).

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

1449

Fig. 3. Emulsion terpolymerization process with recycle loop (Congalidis & Richards, 1998).

measurement techniques that are more specic to polymer processes. During a measurement stochastic errors can be introduced. The effects of process and measurement noise can be minimized by signal conditioning or ltering (Richards & Congalidis, 2005; Seborg, Edgar, & Mellichamp, 2004). All of the techniques to follow represent polymer measurement state-of-the-art and are listed here due to either their novelty or their frequent utilization, with multiple techniques present even in a single installation. A summary table of on-line hardware sensor classication techniques for polymerization reactors can be found in Kammona, Chatzi, and Kiparissides (1999). 2.1. PTLF measurements Pressure, temperature, level, ow and weight are very important basic measurements for polymer processes. They form the cornerstone for all control strategies both regulatory and advanced. These measurements are reviewed in Richards and Congalidis (2005) and Lipt k (2003). a 2.2. Densitometry, dilatometry and gravimetry The density of liquids is monitored by displacement and oat-type densitometers, hydrometers, and hydrostatic densitometers. More advanced instruments are oscillating Coriolis, radiation, vibration, and ultrasonic densitometers. Many of these instruments can be connected photometrically or mechanically to produce a usable electrical signal (Lipt k, 2003). a

Dilatometers measure the volume shrinkage during the course of liquid polymerization reactions and are mainly used for laboratory measurement of monomer conversion. They are based on the principle that polymers are denser than their monomers. As monomer is converted to polymer, volume changes are monitored by following the change in height of the solution inside a graduated capillary tube. Conversion is monitored with a computer-linked photodetector that tracks the meniscus in the capillary and records the height changes (Rodriguez, Cohen, Ober, & Archer, 2003). The percentage of total solids in a polymer sample can be determined by the gravimetric method through moisture weight loss. The sample is loaded onto a pan and the weight determined. Then it is put into an oven at high temperature for a time to remove all volatiles. It is then reweighed and the percent solids determined. 2.3. Viscosity measurement Viscosities are of interest in polymer technology in order to follow the course of a polymerization reaction or to monitor continuously the quality of a product. Viscosity may be constant (Newtonian), shear thickening (dilatant), or shear thinning (pseudoplastic) with shear rate. For polymer systems, solution or melt, the viscosity can be related to the molecular weight of the polymer (Kammona et al., 1999; Lipt k, 2003; Rodriguez et a al., 2003). In most cases viscosity is measured by capillary viscometers or rotating viscometers. The capillary viscometer may

1450

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

also be employed in-line for monitoring of molecular weight in polymerizations as described in Vega, Lima, and Pinto (2001). An indirect method to obtain a measure of molecular size that is quick and inexpensive is the Melt Indexer (Rodriguez et al., 2003). The Melt Index is dened as the number of grams of polymer extruded in 10 min through a capillary 2.1 mm in diameter and 8 mm in length at a certain temperature and pressure (ASTM D1238). It is evident that the Melt Index varies inversely proportionally to the polymer molecular weight. Among the different possibilities to measure viscosities in rotating viscometers, the coaxial cylinder apparatus is the most commonly used in practice. The measurement of the angular velocity of the cup and the angular deection of the bob makes it possible to determine the viscosity (Lipt k, 2003). Beside the a coaxial device the cone-and-plate viscometer is also used. In this device, an inverted cone faces a solid plate and the apex of the cone just touches the plate. The measured liquid is in the free gap. The viscosity of the measured uid is computed from the torque on the driving shaft (Lipt k, 2003). The Mooney a viscometer, particularly used in the rubber industry, is a variant of the cone-and-plate viscometer, which restricts the sample to a disc-shaped cavity (ASTM D1646) (Lipt k, 2003). a 2.4. Measurement of composition The composition of raw materials, nished products, and samples of the various steps of a reaction is normally measured at the laboratory using the appropriate physical and chemical analytical methods. However, sampling and analysis are time consuming and, in many cases, the result of the analysis is only of current interest and too late for control decisions to be made. In order to monitor compositions continuously, one needs automatically functioning analytical instruments that can continuously obtain the composition of a mixture. Optical methods are common (Kammona et al., 1999; Lipt k, a 2003) as infrared spectrographic analysis (IR) permits in many cases to follow the appearance or the disappearance of one or more characteristic absorption frequency bands. These frequency bands correspond to frequencies of vibrations of the bonds in the molecules. One must rst analyze the spectrum of the IR radiation and then measure the corresponding frequencies. More recently the Fourier transform infrared technique (FTIR) has been used for faster data acquisition and handling than traditional IR spectrographic analysis. IR and FTIR can be applied to polymer solutions or solid lms for composition analysis and are particularly useful for copolymer composition determination. Near IR spectroscopy has been used to control a polymerization reactor to produce solution polymers with well-dened molecular weight (Othman, Fevotte, Peycelon, Egraz, & Suau, 2004). Optical analytical devices are also built for measuring radiation in the ultraviolet (UV) and the visible spectral region, but the spectra absorption bands obtained here are usually so broad that these devices are only of limited use. The refractive index (RI) of a mixture is a function of the composition of the mixture and their respective refractive indices (Kammona et al., 1999). Operational measuring instruments are usually differential refractometers or critical angle refractome-

ters (Lipt k, 2003). A differential refractometer is commonly a used as a concentration detector in the efuent of a gel permeation chromatography (GPC) column for molecular weight determination. Raman spectroscopy is dependent on the collision of incident light quanta with the molecule, inducing the molecule to undergo a change (Rodriguez et al., 2003). It is now being used to provide a means of studying pure rotational and vibrational transitions in molecules. Raman scattering of light by molecules may be used to provide chemical composition and molecular structure and is currently being applied to polymers (Elizalde, Leiza, & Asua, 2004; Kammona et al., 1999; Leffew et al., 2005; Reis, Ara jo, Sayer, & Giudici, 2004). u Apart from optical methods, magnetic and electrical methods can also be used for composition measurement. Examples of the latter are conductivity measurements (of ionic liquids, e.g. purity of boiler feeding water), ionization methods (e.g. the ame ionization detector in gas chromatographs or the photo-ionization of gases with UV light as tracking measuring instrument of hydrocarbons in air), electrochemical potential methods (e.g. pH measurements), and occasionally polarographic methods. Nuclear magnetic resonance (NMR) is based on the principle that when a hydrogen containing compound is in a strong magnetic eld and exposed to radio frequency signals, the compound absorbs energy at discrete frequencies (Rodriguez et al., 2003). This technique can be used to measure chain molecular structure, copolymer composition, and copolymer sequence lengths. It can also deduce isotacticatactic ratios and other structure variations as shown for example in Wyzgoski, Rinaldi, McCord, Stewart, and Marshall (2004). Mass spectrometry and NMR are currently not in routine on-line process use but can be used to calibrate other on-line methods. Many methods depend on the separation of a uid mixture. Among these the process gas chromatography (GC) stands out (Kammona et al., 1999). Suitable devices for online control were developed from laboratory gas chromatographs and operate very reliably. The principle of the GC is that a carrier gas (helium) is passed over a tubular column of a ne solid. A sample is injected into the carrier gas stream and the gas efuent from the column is run past a detector such as a ame ionization detector. Calibration is based on the fact that all conditions being equal, a given hydrocarbon will require the same length of time to pass through the column to the detector (elution time) (Lipt k, 2003). a A mass spectrometer source produces ions and information about a sample may be obtained by analyzing the dispersion of ions when they interact with the sample using the mass-to-charge ratio. Sometimes mass spectrometers are used after a separation step such as gas chromatography or liquid chromatography for fraction identication. 2.5. Surface tension In emulsion polymerizations, particularly it may be of interest to measure the surface tension of the emulsion. The surface tension can give an indication of whether or not micelles are present, which is important in particle nucleation above the critical micelle concentration (CMC) (Schork, 1993; Schork,

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

1451

Deshpande, & Leffew, 1993). The online method used is usually the bubble pressure method (Schork et al., 1993). 2.6. Molecular weight distribution (MWD) It is widely recognized that a reliable method of monitoring molecular weight distribution, and the various molecular weight averages (Mn , Mw , and Mz ) during the polymerization process is of importance to nal polymer quality. Traditionally gel permeation chromatography (GPC) or size-exclusion chromatography (SEC) have been used to determine MWD (Rodriguez et al., 2003; Kammona et al., 1999). In GPC/SEC a polymer solution is injected into one or more columns in series packed with porous particles. The packing has small pores and during elution the polymer molecules may or may not, depending on their size, penetrate into the pores. Therefore, smaller molecules have access to a larger fraction of pores compared to the larger ones, and the chains elute in a decreasing order of molecular weights. For each type of polymer an empirical correlation exists between molecular weights and elution volumes. This can be used to calibrate the GPC/SEC, which allows the evaluation of average molecular weights and molecular weight distributions. Direct column calibration for a given polymer requires the use of narrow MWD samples of that polymer. The chromatograms of these standards give narrow peaks and each standard is associated with the retention volume of the peak maximum. There are a number of polymers for which narrow MWD standards are commercially available. More recently triple-detector instruments have been designed, which include a differential viscometer, a light-scattering instrument, and a differential refractometer that monitors the column efuent. A calibration curve can be obtained from this arrangement as long as all signals are calibrated (Rodriguez et al., 2003). For online purposes, the viscosity measures previously mentioned in Section 2.3 have been used as a substitute for molecular weight averages in online control. Some vendors are commercializing more rapid GPC/SEC instruments for online control with some instruments already available. 2.7. Particle size distribution (PSD) The particle size distribution can have a fundamental effect on the physical properties of dispersions that are common polymer products. The measurement of just the average particle size may not be sufcient. For example, the presence of different size populations resulting in a multimodal distribution could have a strong inuence on nal properties and may need to be controlled. There are several particle size measurement techniques used such as optical imaging, electron imaging, optical diffraction and scattering, electrical resistance changes, sieving, sedimentation, and ultrasonic attenuation (Lipt k, 2003). a Optical (larger than 1 m) and scanning electron microscopy (SEM) techniques literally give the clearest picture of a PSD. The two principal light scattering technologies commercially available are light scattering intensity measurement (also known as static or Rayleigh scattering) and dynamic light scattering measurement (also known as quasi-elastic light scattering (QELS) or

photon correlation spectroscopy (PCS)). Dynamic light scattering provides a relatively fast and simple method for submicron particle sizing (Kammona et al., 1999). Turbidimetry, which is a measure of the attenuation of a beam of light passing through a suspended particle sample, has been used traditionally in industry to obtain a measure of average particle size and even the entire PSD (Kammona et al., 1999). Acoustic attenuation measurements can be made without the need for sample dilution and can be used in the particle size range of 10 nm100 m. As sound travels through a slurry or colloid, it is attenuated. The level of attenuation is related to the particle size distribution (Hipp, Storti, & Morbidelli, 2002a,2002b). Capillary hydrodynamic fractionation (CHDF) is a hydrodynamic method for measurement of nanometer-sized particles. In this method, slurry containing the particles is forced through a capillary. The measurable particle size range for CHDF is about 15 nm2 m. For online control, these techniques must be evaluated for speed, reliability and sample dead time (Lipt k, 2003). a As in the MWD techniques, manufacturers are moving more towards online implementation of the more recent methods, but many of these techniques still are practiced off-line in the process analytical laboratory. 3. Advanced regulatory control 3.1. Controllers and actuators Measurement instruments supply information on the current operating conditions of a plant. These form the basis to control the process and to keep conditions constant so that the optimal quality results are obtained. For this purpose, controllers are used, which are devices that are designed to keep a specic controlled variable constant despite outside disturbances. In the past, the large majority of controllers were continuous controllers, but the digital controller has now become commonplace due to the widespread use of computers and distributed control systems. Digital signals are discrete in nature and arise from sampling continuous measurements at equal time intervals or they may arise from naturally discrete signals as, for example, from analyzers. A more extensive discussion can be found in Richards and Congalidis (2005). 3.2. Polymer reactor control issues The polymerization reactor is usually at the heart of the manufacturing process impacting both downstream processing and nal customer related polymer properties. The following factors have contributed to the industrial signicance of polymer reactor control: (1) The need to improve xed asset productivity by optimizing reactor yield and uptime. (2) The trend towards shorter manufacturing campaigns for the different polymer grades manufactured in the same reactor or towards more frequent on-line product transitions to reduce product inventories and hence working capital.

1452

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

(3) Global competition, which not only imposes tough requirements for polymer grade uniformity, but also requires that for new polymer types the time to commercialization be as short as possible. (4) Safety and environmental considerations regarding the stable operation of a potentially thermally unstable reaction. The measurement and control techniques discussed in the previous sections, many of which are generic in nature, are not always adequate to meet practical considerations specic to a polymerization reactor, which pose additional challenges to the successful application of a control strategy. We will highlight several control engineering techniques, some of which are known in the literature as advanced control, which can be deployed to meet some of the challenging aspects of polymer reactor control. These techniques have been extensively discussed in literature reviews (Congalidis & Richards, 1998; Dimitratos, Elicabe, & Georgakis, 1994; Dub , Soares, Penlidis, & Hamielec, 1997; e Embirucu, Lima, & Pinto, 1996; Richards & Schnelle, 1988), and in books (Schork, 1993; Schork et al., 1993). For example, Dimitratos et al. (1994) reviewed the major issues related to control of emulsion polymerization. The authors point out that although emulsion polymerization has been studied and used for several decades, progress has been slow. The special focus in the Embirucu et al. (1996) survey was optimal control theory, nonlinear control, adaptive control, and predictive control. In our experience several issues specic to polymer reactors are: (1) Polymerization reactors are known both theoretically and experimentally to exhibit multiple steady states (Adebekun, Kwalik, & Schork, 1989; Ray & Villa, 2000)and in some cases may also exhibit oscillations in terms of monomer conversion and polymer particle diameter (Meira, 1981). Furthermore, polymerization reactors can be highly exothermic and may result in reactor thermal runaway unless an effective control strategy is implemented. In the case where multiple steady states are present, as shown for example in the case of polyethylene reactors by Villa, Dihora, & Ray (1998), the two stable steady states at very low and very high monomer conversion respectively cannot be chosen as the reactor operating point either for practical, economic, or for safety reasons. It may be therefore necessary to choose the unstable steady state at intermediate conversion as the reactor operating point. In this case the design of an appropriate stabilizing closed loop reactor controller is essential to smooth reactor operation. (2) As discussed previously, on-line measurements of polymer architecture such as composition, molecular weight and degree of branching are not always available and for some polymer systems may be simply unavailable. In many cases the control engineer may have to rely on polymer properties inferred from infrequent laboratory analysis of reactor samples or from laboratory analysis of the nal polymer after it has experienced signicant post reactor processing thus introducing large dead times in the control loop. For new specialty polymer types, which are designed for specic applications, the composition and molecular weight reactor

(3)

(4)

(5)

(6)

control specications may not be immediately available and are usually determined by lengthy trial and error processes involving product trials at the nal customer as discussed by Leffew et al. (2005). The relationship between reactor operating conditions such as monomer conversion, temperature, residence time, polymer composition, and viscosity and the customer related nal properties such as tensile strength, elongation at break, and processibility in an injection molding machine may not always be well dened. Variability in the polymer isolation process and long term polymer structure changes such as aging may result in the fact that although in some cases the reactor may be operating on-aim within well dened manufacturing specications, the nal polymer may not process satisfactorily when delivered to the customer. The control system for a polymerization reactor must be sufciently robust to handle unmeasured disturbances, which impact polymer reactor operation. These disturbances typically result either from trace amount of polymerization inhibitors left over after monomer purication prior to the polymerization reaction or from trace amounts of other compounds which may be present in a typical polymerization recipe and which may be affecting the reaction. A polymerization reactor often produces several grades (in terms of composition and viscosity) of the same polymer and therefore the control strategy must be easily adapted to a multi-product plant and in some cases to on-line grade transitions. In the case of a multi-product plant, it may be necessary to operate the reactor in terms of rather short campaigns in order to minimize nished product inventory and thus working capital. In these cases, the reactor control system must be designed in such a way as to achieve fast startups while minimizing off-specication polymer formation. A good understanding of the polymerization kinetics is essential in designing a robust and effective reactor control strategy. In most cases, the nature of the kinetics (free radical versus condensation versus living polymerization) and the choice of reactor type (batch versus CSTR, homogeneous versus multiphase) have a direct bearing on the distribution of the nal polymer properties as shown by Ray (2003).

The discussion will be facilitated by focusing on the copolymerization process shown in Fig. 2. The owsheet captures many of the elements of actual free radical copolymerization reactor installations. This process, rst discussed by Congalidis et al. (1989), has as of the time of this paper resulted in over 60 citations, and has been used as a benchmark to test various control and estimation schemes (BenAmor, Doyle, & McFarlane, 2004; Bindlish & Rawlings, 2003; Doyle, Ogunnaike, & Pearson, 1995; Harris & Palazo lu, 1998, 2003; Lee, Han, & Chang, g 1997, 1999; Lewin & Bogle, 1996; Ling & Rivera, 1998; Maner & Doyle, 1997; Maner, Doyle, Ogunnaike, & Pearson, 1996; Ogunnaike & Ray, 1994; Ozkan, Kothare, & Georgakis, 2003; Parker, Heemstra, Doyle, Pearson, & Ogunnaike, 2001; Regnier, Defaye, Caralp, & Vidal, 1996). The main motivation stated by Ozkan et al. (2003) in selecting this process as a benchmark is its complexity, hence its similarity to an industrial problem. The

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

1453

characteristics of this reactor that are most relevant and interesting are that it is representative of a wide class of free radical polymer reactors with recycle, exhibits multiple steady states, is highly nonlinear, is MIMO, and involves multiple sensor issues such as the necessity of online molecular weight and composition measurements or their estimates and dead-time issues such as gas chromatograph delays. However, the problem achieves all this without being overly complex. As will be discussed in the subsequent sections, the implementation of various control algorithms using multiple models on this process has proven very relevant in the area of polymerization reactor control. As shown in Fig. 2, monomers A and B are continuously added with initiator, solvent, and chain transfer agent. In addition, an inhibitor may enter with the fresh feeds as an impurity. These feed streams are combined (stream 1) with the recycle (stream 2) and ow to the reactor (stream 3), which is assumed to be a jacketed well-mixed tank. A coolant ows through the jacket to remove the heat of polymerization. Polymer, solvent, unreacted monomers, initiator and chain transfer agent ow out of the reactor to the separator (stream 4) where polymer, residual initiator, and chain transfer agent are removed. Unreacted monomers and solvent (stream 7) then continue on to a purge point (stream 8), which represents venting and other losses and is required to prevent accumulation of inerts in the system. After the purge, the monomers and solvent (stream 9) are stored in the recycle hold tank, which acts as a surge capacity to smooth out variations in the recycle ow and composition. The efuent (stream 2) recycle is then added to the fresh feeds. Fig. 3 represents an actual industrial emulsion terpolymerization process described by Congalidis and Richards (1998), where the techniques of feedforward control, ratio and cascade control, as well as composition and viscosity feedback control, which are discussed in subsequent sections, were successfully implemented by the authors. The emulsion polymerization process owsheet of Fig. 3 is very similar to the solution polymerization process of Fig. 2, with the exception that water replaces the solvent and the reactor operates adiabatically.

Feedforward control was developed to counter some of these limitations. Its basic premise is to measure the important disturbance variables and then take corrective compensatory action based on a process model. The quality of control is directly related to the delity and accuracy of the process model. Two implementations of feedforward control will be discussed, which are widely used in polymer reactor control, namely feedforward control design based on steady state models and ratio control. Furthermore, the powerful combination of feedforward and feedback control will be discussed, because it utilizes the best of both approaches since feedforward control works by reducing the effects of measured disturbances and feedback control provides the necessary compensation for the effects of model and measurement inaccuracies as well as unmeasured disturbances. 3.3.1. Steady state model feedforward control To illustrate this approach the polymerization process described by Congalidis and Richards (1998) is considered (see Fig. 3). The presence of the recycle stream introduces disturbances in the reactor feed which perturb the polymer properties. The objective of the feedforward control is to compensate for these disturbances by manipulating the fresh feeds in order to maintain constant feed composition and ow to the reactor. Feedforward control of the recycle allows the designer to separate the control of the reactor from the rest of the process. The feedforward control Eqs. (1) and (2) were obtained by writing component material balances around the recycle addition point. For example for monomer A ow this balance is: qna3 = qna1 + ya2 qn2 (1)

Eq. (1) is then solved for the fresh feed of monomer A since it is desired to keep the target ow of monomer A to the reactor (qna3 ) constant: qna1 = qna3 ya2 qn2 (2) The recycle composition (ya2 ) is typically measured by online gas chromatographs, which may have signicant time delays. If a faster response time of the analyzer is required, an Infrared or Raman spectroscopy probe may be used. Recycle ow is typically measured and controlled by manipulating the recycle valve to maintain the desired inventory in the hold tank. Any disturbances in the recycle composition or ow will cause variations in the fresh feed in order to keep the reactor feed constant. Similar feedforward controllers are implemented for monomers B and C. As shown in Ogunnaike and Ray (1994) and Congalidis et al. (1989) the performance of the feedforward control allows for the perfect compensation of disturbances that can for example arise from a step change in the purge ratio so that the reactor polymer characteristics (e.g. composition and molecular weight) are unaffected. Naturally, perfect feedforward control will typically not be possible unless the recycle composition measurement is instantaneous (or the holdup in the recycle tank is signicantly larger than that of the reactor) and all other measurements are perfect. However, the implementation of feedforward control does have the advantage that the reactor is decoupled from the

3.3. Feedforward control The traditional PID feedback controller is very widely used, because it requires minimal process knowledge for its design. In particular, a mathematical model of the process is not required although it can be quite useful for appropriate tuning. Furthermore, if process conditions change, the PID controller can be retuned to maintain satisfactory performance. A properly tuned PID controller can be quite robust in maintaining good steady state operation in the face of unmeasured disturbances. However, since control action can only occur if a deviation occurs between the set point and the measured variable, perfect control is not possible. Therefore, feedback control fails to provide predictive control action to compensate for the effects of known disturbances. A more serious limitation, which is particularly important for polymer reactor control, is that the controlled variable cannot always be measured online.

1454

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

Fig. 5. Conventional temperature control of an adiabatic polymerizer (Richards & Congalidis, 2005). Fig. 4. Ratio control of monomer feeds (Richards & Congalidis, 2005).

recycle stream and thus reduces the control problem to just that of controlling the reactor. In our experience, we have seen continuous polymerization installations similar to the one shown in Figs. 2 and 3 where, in the absence of feedforward control, operators were making adjustments of monomer fresh feed ows based on feedback from laboratory polymer composition measurements resulting in poor performance of the unit. 3.3.2. Ratio control Ratio control is a form of feedforward control, which is widely used in the chemical industry and has proven very useful in reactor control. As is evident from its name, its purpose is to keep the ratio of two process variables at a given value and hence it can be deployed when the objective is to maintain a certain proportion, or stoichiometry, of reactants to the reactor. Although the concept is fairly obvious, in our experience we have seen that it has not been as extensively used in polymer reactor control, especially in the case of reactors producing specialty polymers with complex polymerization recipes. Typically, ow controllers are designed for each of the reactor feed streams (e.g. monomer, initiator, chain transfer agent) and each one of these controllers has a set point, which is dependent on the particular polymer being made. However, when ratio control is implemented as shown in Fig. 4, one of the reactor feed streams (monomer A in this case) is chosen as the reference stream. The measured ow rate of monomer A is then transmitted to the ratio station RC, which multiplies the signal by the desired ratio (typically determined by the polymer chemist) to calculate the set point for the ow controller of monomer B. 3.4. Cascade control Cascade control is also widely used in the chemical process industries and especially in cases where there may be nonlinear behavior in the dynamics of the control loop. It also addresses the main drawback of conventional feedback control namely the

fact that control action only occurs where the controlled variable deviates from the set point. Unlike feedforward control, which requires that disturbances be explicitly measured and a model be available to calculate controller output, cascade control introduces an additional measurement and an additional feedback controller. The secondary measurement is typically located so that it recognizes the upset conditions sooner than the controlled variable. The concept of cascade control has been used extensively for effective reactor temperature control. In many instances polymerization reactors are operated adiabatically. In the case shown in Fig. 5 and in which only traditional feedback control is used, measurement of the reactor temperature is used to manipulate the heat exchanger ow to cool the reactor feed so that the reactor adiabatic temperature rise is adequate to remove the heat of polymerization. This conventional scheme may do a satisfactory job of regulating reactor temperature but disturbances that occur in the feed line may result in a rather sluggish response of the temperature controller. Polymer properties are very sensitive to temperature excursions and in many cases, this sluggish response of the temperature control loop may not be acceptable. Cascade control as shown in Fig. 6 resolves the problem by introducing an additional measurement namely the temperature of the reactor feed and an additional controller. 4. Advanced supervisory control 4.1. Feedforward-feedback control The combination of feedforward and feedback control provides a very powerful practical strategy for the control of polymer properties such as composition and molecular weight. Typically, it is still difcult to have online direct measurements of polymer composition, so the control design has to incorporate the available off line reactor sample composition measurements obtained at the laboratory typically using Infrared or nuclear magnetic resonance techniques. Similarly, despite advances in size exclusion chromatography/gel permeation chromatography technology (SEC/GPC), online SEC/GPC is not routinely avail-

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

1455

able for most industrial polymer reactor control applications. Therefore, in many cases the control engineer has to rely on offline measurements of molecular weight and incorporate them in the control strategy. Typically, measures of molecular weight used in control applications are the inherent viscosity and/or the melt index, the latter being very common in polyolen production. Recent advances in capillary rheometry, as reported in the vendor literature, provide the capability of continuous measurement of melt index and viscosity during polymer production using on-line or at-line instrumentation. At-line instrumentation refers to a manual measurement of a sample at the process line by using for example a portable instrument as contrasted with automatic on-line sampling and off-line measurement at the laboratory. It is also very important to establish appropriate set points and specications for the inherent viscosity and melt index by relating them to the underlying molecular weight distribution as shown for example in Seavey, Liu, Khare, Bremner, and Chen (2003). We may return to the polymerization process shown in Fig. 3 and follow the discussion in Congalidis and Richards (1998) as well as the concept of the process control hierarchy shown in Fig. 1. It is very important to use process understanding (typically captured in dynamic simulations using experimentally validated models) in the design of the control system for polymer composition and molecular weight. Polymerization reaction modeling showed for this specic case, which is however typical of a wide class of continuous polymer reactors in practice, that the reactor temperature control is crucial because inherent viscosity is extremely sensitive to temperature variations. As discussed in the preceding section cascade control can be effectively deployed to control reactor temperature within very tight specication limits. Achieving good reactor temperature control can become particularly challenging in multi product semibatch polymerization reactors, because physical properties of the reactor contents vary from run to run and within a run and the standard PID controllers used in a cascade design may not be able to perform satisfactorily over the entire range of

operation required. In these cases more advanced temperature control strategies based on adaptive control (Seborg et al., 2004) or model based control can be effectively used as shown for example by Chylla and Haase (1993) for an industrial reactor and by Defaye, Regnier, Chabanon, Caralp, and Vidal (1993) for a laboratory scale one. Another example of model based control is given by Othman et al. (2004) where a model for the monitoring of both the average polymer molecular weight and the concentration of monomer in the reactor was developed with the partial least-squares calibration technique applied to NIR spectra. On the basis of a process model, a nonlinear input-output linearizing geometric approach was then applied to control the polymer molecular weight by manipulating the inlet ow rate of the monomer. Moreover, adaptive cascade control strategies can provide better temperature control performance without the need for retuning versus a traditional PI cascade control system. Adaptive control strategies were discussed by Tyner, Soroush, and Grady (1999) for jacketed stirred tank reactors in which multiple products are produced and the overall heat-transfer coefcient is unknown and can vary signicantly as a result of fouling. An additional difculty in the control of polymer properties is that in some cases the control problem is multivariable in the sense that there are interactions between the molecular weight and composition loops and therefore when a manipulated variable is chosen to control molecular weight it may affect composition. It is important to use process knowledge to validate the selection of manipulated variables. For example, for the polymerization reactor shown in Fig. 3 process simulations showed that one way to decouple polymer quality control is to take advantage of the fact that polymer composition is naturally very sensitive to changes in reactor feed composition but inherent viscosity is relatively insensitive to reactor feed composition changes. As discussed in Congalidis et al. (1989) there exists a much more formal approach for feedback control system design. It consists of creating an approximate linear multivariable model from the nonlinear polymer reactor model using step test data and then using the techniques of relative gain array (RGA) and singular value decomposition analysis (SVD) (as described for example in Seborg et al. (2004)) to determine the best pairings of controlled and manipulated variables for robust multi loop control. A typical control strategy for polymer composition, which has been successfully implemented for the reactor shown in Fig. 3, is illustrated in Fig. 7. The measurements used are the ows of the fresh monomer feeds, the recycle ow, the monomer composition of the recycle feed and the total monomer reactor feed provided by two on-line gas chromatographs, and the polymer composition provided by laboratory analysis of reactor samples. The supervisory control consists of three levels implemented in a cascade fashion: (1) Feedforward controller, previously discussed, which maintains total monomer feed rate and monomer A feed composition at the appropriate set points for the specic polymer grade being produced. The feedforward controller can be easily extended to many monomers by specifying the total

Fig. 6. Cascade temperature control of an adiabatic polymerizer (Richards & Congalidis, 2005).

1456

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

Fig. 7. Copolymer composition control strategy.

monomer reactor feed rate and the feed composition for the monomers. (2) Gas chromatograph feedback controller, which uses the velocity algorithm for digital PID control, to calculate ow correction factors for the monomers from the gas chromatograph measurement of the actual monomer feed composition. This controller provides the necessary integral action so that the offset between the actual reactor monomer feed composition and its set point, which may be caused by owmeter inaccuracies or other unmeasured disturbances, is minimized. (3) Polymer composition feedback controller, which updates the set points for the reactor monomer feed composition based on the laboratory analysis of a reactor sample. This controller thus provides the necessary integral action so that the offset between the measured composition of the reactor sample and the polymer grade composition is minimized. It is important to note that these feedforward and feedback controllers have been designed hierarchically in the sense that each level in the structure will not activate unless the levels below it are functioning properly (as in Fig. 1). Furthermore, in practice extensive data validation checks must be incorporated so that robust performance can be assured even when the gas chromatograph or laboratory analysis measurements may be unavailable or faulty.

The control of inherent viscosity shown in Fig. 8 uses the same approach as the composition control. Depending on the polymer grade being manufactured the initiator or the chain transfer agent may be used to control emulsion viscosity. An inherent viscosity feedback controller adjusts automatically the set point of the monomer to transfer agent ratio controller based on the measured viscosity value and provides the necessary integral action so that the difference between the reactor inherent viscosity and the polymer grade inherent viscosity specication is minimized. As in the case of composition control, extensive data and controller output checks have been incorporated in the practical implementation to provide robust performance. 4.2. State estimation techniques In the design of the composition and viscosity feedback controllers it is very important to establish whether the polymer reactor dynamics need to be taken explicitly into account. The choice of sampling frequency balances the requirements for good quality control versus the need to minimize analytical costs. Usually, when the reactor residence time is much shorter than the sampling frequency, integral control is appropriate, because the interval of time between measurements is usually sufcient for the effect of an adjustment to a process variable set point to be complete within this interval of time. In other cases the sampling dead time introduced by the periodic

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

1457

Fig. 8. Inherent viscosity control strategy.

analysis of polymer concentration, polymer composition, and molecular weight may necessitate the incorporation of on-line state estimators of polymer properties. Reaction calorimetry aims to measure heat released from a polymerization in order to infer monomer conversion and polymerization rate (as reviewed for example in Kammona et al. (1999), Moritz (1989) and McKenna, Othman, Fevotte, Santos, and Hammouri (2000)). Careful measurement and balancing of mass and energy ows are necessary for success of this technique. For example, the commercial Mettler-Toledo RC1 jacketed reactor acts as a calorimeter supplying mass balance, polymerization heat generation, and transport data. On-line estimation may also be accomplished using rst principles polymerization kinetic models implemented online in the form of an extended Kalman lter (EKF) as illustrated for example in (Ellis, Taylor, & Jensen, 1994; Kammona et al., 1999; McAuley & MacGregor, 1991; Othman et al., 2004; Park, Hur, & Rhee, 2002). For example, Othman et al. (2004) estimated the reaction rate and molecular weight using a NIR in-line measurement of the monomer concentration. It should be pointed out that the choice of techniques for online estimation of polymer properties is still an active area of research and is very much dependent on the specics of the polymer chemistry and available online and ofine instrumentation. In another example, where the process model (Congalidis et al., 1989) was used, BenAmor et al. (2004) applied an industrial real-time optimization package for the nonlinear model predictive control (NLMPC) of a simu-

lated polymer grade transition. A receding horizon estimation scheme and, separately, a Luenberger observer were designed to reconstruct unmeasured states. In the study by Bindlish and Rawlings (2003), the process model (Congalidis et al., 1989) was again used to develop an extended Kalman lter for state estimation. 4.3. Model predictive control (MPC) The previously discussed single loop and appropriately chosen multi loop feedforward and PID feedback control strategies may not be adequate for the effective control of polymer properties particularly in the case when the polymerization reactor exhibits strongly nonlinear dynamic behavior or when there are strong interactions between the controlled variables or when there are constraints on the manipulated variables. From the advanced process control techniques such as internal model control (IMC), inferential control, and adaptive control that have been developed by the academic process control community for these tough multivariable control problems, model predictive control has reached the stage where it is having a signicant impact on industrial practice. Linear MPC algorithms are rapidly becoming imbedded in the Distributed Control System software libraries, which facilitates their use. As reported in a recent survey (Qin & Badgwell, 2003), by the end of 1999 there were at least 4500 industrial MPC applications worldwide, mainly in oil reneries and petrochemical plants.

1458

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

Fig. 9. MPC block diagram (Seborg et al., 2004). Material used by permission of John Wiley & Sons, Inc.

One of the earlier implementations of MPC is linear dynamic matrix control (DMC). A simulation study was carried out by Meziou, Deshpande, Cozewith, Silverman, and Morrison (1996) to assess the performance of DMC for an ethylene-propylenediene polymerization reactor. Closed-loop simulation results conrmed the potential of the multivariable DMC technique to reduce off-specications during changes in EPDM product quality setpoints, production rate, and catalyst activity. The structure of MPC is shown in the block diagram Fig. 9 (Seborg et al., 2004). A mathematical model of the process is used to predict the current values of the output (controlled) variables. The model is usually implemented in the form of a multivariable linear or nonlinear difference equation. It is typically developed from data collected during special plant tests consisting of changing an input variable or a disturbance variable from one value to another using a series of step changes with different durations or more advanced protocols such as the pseudorandom-binary sequence described in Seborg et al. (2004). The residuals (i.e. the difference between the predicted and actual output variables) serve as a feedback signal to the prediction block and are used in two types of control calculations that are performed at each sampling instant, namely set point calculations and control calculations. A unique feature of MPC is that inequality constraints can be incorporated in both the set point and control calculations. In practice, inequality constraints arise as a result of physical limitations on plant equipment such as pumps, control valves, and heat exchangers. The set points for the control calculations are typically calculated from an economic optimization of the process based on a steady state process. Typical optimization objectives can include maximizing a process function, minimizing a cost function, or maximizing a production rate. The objective of the control calculations in the control block is to determine a sequence of control moves (changes in the manipulated variables) so that the predicted response moves to the set point in an optimal manner for example by following a reference trajectory. The calculated control actions are implemented as set points to regulatory control loops. For a detailed explanation of the different design choices that are necessary for the effective implementation of design of the MPC controller the reader is referred to Seborg et al. (2004) and Ogunnaike and Ray (1994). It is also essential to point out that the quality of MPC is very

strongly dependent on the availability of a reasonably accurate process model that can capture the interactions between input, output, and disturbance variables. Although MPC control has become an established technology for tough control problems in petrochemical plants its application in polymer reactor control is currently transitioning from purely academic studies using simulated examples to applications in industrial reactors. One of the difculties for the widespread use of MPC control in polymer reactors could be that the embedded MPC modules inside the major DCS packages are variants of the linear MPC algorithm. However, the most signicant benets from MPC for polymer reactors would be derived from applying nonlinear MPC due to severe process nonlinearities exhibited for example during product transitions. As an example, BenAmor et al. (2004) described the application of an industrial real-time optimization package to the nonlinear model predictive control of a simulated polymer grade transition. The NLMPC algorithm was formulated by using orthogonal collocation to integrate the model equations and sequential quadratic programming to solve the resulting nonlinear programming problem. As expected, academic researchers have been prominent in the application of nonlinear MPC in polymer reactors. For example, referring to the polymerization process previously described in Fig. 2 and discussed in Congalidis et al. (1989), several academic researchers as for example in Maner and Doyle (1997), Ozkan et al. (2003), and Bindlish and Rawlings (2003) have designed and implemented nonlinear MPC controllers. These controllers were used successfully for plant startup, minimization of off grade product during grade transitions and regulation around a set point. For example, Maner and Doyle (1997) implemented linear MPC using the same transfer functions as were identied by step tests by Congalidis et al. (1989). However, they were not able to get a signicantly better performance than that obtained from a multiloop PI control strategy. They then proposed to close the dominant loop with a PI controller and identied Autoregressive plus Volterra models for the other loops. A multivariable nonlinear MPC scheme was implemented and performance improvement was observed compared to a multiloop PI control strategy and a multivariable linear MPC structure. As an additional example in the paper by Ozkan et al. (2003) the authors studied the control of the solution copolymerization reactor proposed by Congalidis et al. (1989) using an MPC algorithm based on multiple piecewise linear models. The control algorithm was a receding horizon scheme with an objective function, which has nite and innite horizon cost components and uses multiple linear models in its predictions. Simulation results on the model (Congalidis et al., 1989) demonstrated the ability of the algorithm to rapidly transition the process between different operating points. It was observed that the transition period decreased as the number of linear models increased, which resulted in a reduction of off-specication product produced during the grade transition. Furthermore, in the study by Bindlish and Rawlings (2003), the process model proposed by Congalidis et al. (1989) was used to develop a target linearization and model predictive controller

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

1459

(TLMPC) that separates the target calculations from dynamic regulation. The proposed controller targets are calculated from the nonlinear model, and then used to get the correct linearized model. The linearized model is utilized to calculate the actual control moves for dynamic regulation and develop an extended Kalman lter for state estimation. The authors claim that the resulting TLMPC controller does not have the computational burdens associated with a full nonlinear model predictive controller (NMPC) but shows a performance similar to a NMPC. The TLMPC shows optimal plant startup, product grade transitions, and regulation around a set point for the polymerization process in comparison to a xed linear model predictive controller (LMPC). They found that TLMPC performs better than a xed linear model predictive controller for different operating conditions. Recent reported industrial application of MPC in polymer reactors as for example by Seki et al. (2001) and Bartusiak, Fontaine, Schwanke, and Gomatam (2003) have focused on the transition control requirements, which typically consist in completing the transition in either a minimum amount of time or with minimum amount off-specication product. As pointed out by Takeda and Ray (1999), MPC is well suited to the grade transition problem, which often translates into aggressive manipulated variable moves often against manipulated variable limits or plant constraints. Depending on the specic process, reported benets have also been achieved in product consistency or increased production rate. A review of the application of linear MPC tools to a prototype continuous polymerization (CP) process was given by Schnelle and Rollins (1998) where the authors discuss in general why the MPC technology may be a good t for CP control problems (e.g., minimize settling time after rate changes or process upsets, compensate for signicant multivariable interactions and unusual process dynamics). A related control technique, internal model control (IMC), was tested on the Congalidis et al. (1989) problem by Harris and Palazo lu (1998) and Harris and Palazo lu (2003). A nong g linear control design strategy was proposed based on functional expansion (FEx) models. The controller was formulated as a linear controller with a series of correction terms to account for nonlinearities in the system. IMC was also used by Mahadevan, Doyle, and Allcock (2002) on the problem of scheduling grade transitions in an isothermal methyl methacrylate polymerization reactor. 4.4. Batch and semibatch control Previous sections focused on continuous reactor control. Now we will discuss an even more common class of reactors, batch and semibatch, which are always operated dynamically. 4.4.1. Operation and variability Batch reactors are the most common reactor used in polymerization engineering. They may vary in size from a ve-gallon pilot unit to a 30,000 gallon (or greater) production size (Schork, 1993). Removal of the heat of polymerization is accomplished by circulating coolant through a jacket or by reuxing monomer and solvent. The main advantage of batch reactors is the exi-

bility to accommodate multiple products. They are well suited for low-volume products and for products for which there are numerous grades (as in specialty polymers), because each batch can be made according to its own recipe and operating conditions without the costs incurred when a continuous reactor is shut down and restarted. Process control of batch reactors must address the main disadvantage of batch reactors versus continuous ones, namely variability within a batch and/or variability from batch to batch. This variability is particularly important in batch free radical polymerization, where the time of formation of a single chain is only a very small fraction of the batch time and therefore inhomogeneity results from the fact that polymer chains can be formed under very different conditions during the course of the batch. This is especially signicant for composition control in a free radical batch copolymerization reactor where, unless special control strategies are deployed, polymer chains formed early in the reaction may contain a higher fraction of the more reactive monomer than the chains formed later in the reaction (i.e. compositional drift). On the contrary, in step growth polymerization (e.g. polyamides and polyesters), where the growth time of an individual chain is approximately the batch time, the effects of the changing reaction environment and hence within batch inhomogeneities are much less of an issue, since all chains will see the same changing environment (Schork, 1993). Operation in semibatch reactor mode is very common in polymer reaction engineering practice. Typically, one way to address the issue of compositional drift in free radical batch copolymerization is to operate the reactor under the so-called starved feed policy (e.g. Arzamendi & Asua, 1989). In this case the monomer feed rate is automatically adjusted to maintain a constant rate of reaction as inferred for example by reactor pressure. In this starved feed operation the reaction environment is maintained constant during the batch and therefore the monomer composition in the reactor feed is equal to the desired polymer composition. However, a cost penalty may occur because of unacceptably long batch times for slow reactions. Therefore, it is also possible to implement more sophisticated control strategies during the batch by establishing operating trajectories for initiator addition, monomer addition, and/or reactor temperature to achieve desired polymer properties in minimum time, maximize productivity, or tailor the polymer molecular weight distribution. This is typically accomplished by solving off line an optimization problem using a kinetic model of the process as shown for example in Maschio, Bello, and Scali (1992, 1994), Saenz de Buruaga, Armitage, Leiza, and Asua (1997), Thomas and Kiparissides (1984), and Tyner, Soroush, Grady, Richards, and Congalidis (2001). These essentially open loop trajectories constitute a form of feedforward control and are then implemented as part of the batch sequential logic and recipe management system using ladder logic and binary logic diagrams as shown in Seborg et al. (2004). Leffew et al. (2005) recently demonstrated a different approach to maintain polymer compositional uniformity in a pilot scale semibatch reactor used to produce specialty copolymers containing up to four monomers for electronic applications. In this application, a Raman spectrometer was used on line to

1460

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

measure the monomer composition in the reactor, which was then used by an advanced feedback controller to manipulate monomer feeds and maintain the monomer reactor composition within tight limits. This closed loop policy resulted in better polymer compositional uniformity versus a simpler open loop starved feed approach. Improved polymer compositional uniformity resulted in substantially improved performance of the polymers in the end-use application. If monomer conversion and molecular weight information is available during the batch (for example through online densitometry, energy balance estimation, or online gel permeation chromatography) it can be incorporated as part of a feedback adaptive predictive control strategy. For example, in the work of Houston and Schork (1987), the objective was to maintain the molecular weight at a desired value, while bringing the reaction to a specied monomer conversion in minimum time by manipulating initiator feed rate and coolant jacket temperature. In addition, if a fundamental model of the process is available including reaction kinetics and an energy balance, control moves during the batch can be calculated by solving a nonlinear dynamic optimization problem within the context of the previously discussed model predictive control to account for the wide variety of constraints typically encountered in batch systems. This approach was also described by Peterson, Hernandez, Arkun, and Schork (1992) and applied recently by Wassick, Coffey, and Callihan (2003) to a set of commercial polymerization reactors exhibiting challenging dynamic behavior that prevented conventional control from delivering optimum manufacturing performance. During start-up or grade-change operation of the reactor described by Congalidis et al. (1989), increasing the transition time required to reach the desired steady state from the initial state, increases the amount of off-specication polymer product. Lee et al. (1997) calculated the minimum transition time and subsequently determined optimal control input trajectories using genetic algorithms combined with heuristic constraints. The proposed methodology was shown to have better performance than iterative dynamic programming (IDP) or sequential quadratic programming (SQP) in terms of accuracy and efciency. In a subsequent paper (Lee et al., 1999), they proposed a modied differential evolution (MDE) algorithm for the Congalidis et al. (1989) problem. The authors claim that the proposed MDE algorithm offers faster speed, exible implementation, and higher robustness to nd the global optimum than differential evolution algorithms. 4.4.2. Statistical process control In many cases of batch and semi batch polymerization control there are no online measurements of polymer quality (e.g. polymer composition and molecular weight) during the batch and these measures of end use properties are only available at the end of the batch. In this case recipe modications from one run to the next are common. The minimal information needed to carry out this type of batch-to-batch control is a static model relating the manipulated variable to the quality variables at the end of the batch. As pointed out in Seborg et al. (2004), this model can be as simple as a steady state (constant) gain relationship or a

nonlinear model that includes the effects of different initial conditions and the batch time. The philosophy of statistical process control can be very useful in this case, since the polymer quality variable (for example the Mooney viscosity in elastomers manufacture) can be plotted for each successive batch on a Shewhart (x-bar) chart with the upper and lower control limits placed at three standard deviations above and below the target. The likelihood of a point outside the control limits means that the batch is out-of-control and the batch recipe and possibly the sequence logic must be adjusted for the next batch. If the quality variable for the batch is within the control limits, no control action is taken to prevent manipulations of the batch process based on stochastic variations within it. Very often in DCS operated batch polymer reactors the primary process variables such as pressure, temperature, level and ow are recorded during the batch as well as the quality variables at the end of the batch. However, it may be very difcult to obtain a kinetic model of the polymerization process due to the complexity of the reaction mechanism, which is frequently encountered in the batch manufacture of specialty polymers. In this case, it is possible to use advanced statistical techniques such as multi-way principal component analysis (PCA) and multiway partial least squares (PLS) along with an historical database of past successful batches to construct an empirical model of the batch (Clarke-Pringle & MacGregor, 1998; Flores-Cerrillo & MacGregor, 2003; Kourti, Nomikos, & MacGregor, 1995; Nomikos & MacGregor, 1995). This empirical model is used to monitor the evolution of future batch runs. Subsequent future unusual events can be detected during the course of the batch by referencing the measured process behavior against this incontrol model and its statistical properties. It may be therefore possible to detect a potential bad batch before the run is over and take corrective action during the batch in order to bring it on aim. An alternate technique to the purely statistical estimation and control methodologies has been used by Fotopoulos, Stenger, and Georgakis (1998) for the modeling and optimization of batch reactor processes when a detailed understanding based on a fundamental principles and detailed kinetic studies are not available. This technique known as tendency modeling, originally proposed by Filippi et al. (1986), consists in the development of a low order nonlinear dynamic model, which approximates the stoichiometric and kinetic models of a process using available plant data along with fundamental knowledge of the process characteristics. The tendency model is evolutionary since it is updated as new process data become available. Fotopoulos et al. (1998) explored the effect of process-model mismatch associated with tendency models on the state estimation of a batch processes consisting of eight individual reaction steps and showed how these models could be used in an extended Kalman lter and to tune the model covariance matrix. Although the use of tendency models in polymer reactors, where the number of individual reactions is much higher, has been limited, Wang, Pla, and Corriou (1995) constructed a tendency model and used it as an open loop predictor of product quality (molecular weight and polydispersity) for the control of a batch styrene polymerization reactor.

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463

1461

5. Future trends In discussing future trends it may be useful to reect why the plethora of advanced process control techniques developed by academic researchers is not as widely practiced in polymer reactor control as in other parts of the chemical process industries for example distillation or more broadly petrochemical renery control. We agree with the authors of previous reviews (Embirucu et al., 1996; Richards & Schnelle, 1988) who concluded that there is a gap between industry and research practices. In our assessment the reasons for this lag, which therefore become motivators for new research are: (1) Whereas off line or on line measurements of polymer architecture such as composition, molecular weight and degree of branching are available for commodity polymers, they are not always available for specialized polymers, which are increasingly becoming important in high value applications such as electronics. It is almost a given that deployment of more accurate robust measurements will increase the performance of the control system. Embirucu et al. (1996) found that actual implementation of advanced control theory in the polymerization area requires the improvement of measurement and state estimation techniques. (2) The development of process models appropriate for control design is challenging, because of the complexity of the polymerization kinetics and the reaction environment which may be homogeneous or heterogeneous or characterized by zones of poor mixing. It may not always be obvious what is the appropriate level of model complexity or how best to identify model parameters or how to integrate available measurements in these models in order to estimate unmeasured states related to nal product properties. For example Li, Corripio, Henson, and Kurtz (2004) have recently proposed a hierarchical extended Kalman lter (EKF) to estimate unreacted state variables and key kinetic parameters in a rst principles model of a continuous ethylene-propylene-diene polymer reactor and used simulation tests to show that the hierarchical EKF generates satisfactory estimates even in the presence of measurement noise and plant/model mismatch. Part of the challenge, and at the same time the opportunity, would be that some of these process modeling and control solutions tend to be polymer platform specic and must be validated in a laboratory or at least a pilot plant before their deployment in a large scale plant. (3) Some of the newer multivariable control techniques such as linear MPC as illustrated by Schnelle and Rollins (1998) or nonlinear MPC as illustrated by Ozkan et al. (2003) require specialized knowledge in the eld that may not always be available. It is encouraging that the level of process control education of the engineers working in the polymer industries has been steadily increasing thus facilitating the technology transfer from academia. At the same time it is a challenge to develop sophisticated algorithms in a way that their implementation and maintenance would be relatively transparent to a process control engineer familiar only with the basics

of regulatory control. This is an area where the major DCS vendors have a signicant role to play in the development of hardware/software solutions. (4) Trends in the polymer industry, which reinforce the importance of measurement and reactor process control, include the increasing emphasis by customers on receiving a uniform product with desired property specications from their polymer suppliers. This is a powerful economic incentive for reactor control especially when it is coupled with the need to operate the unit safely without in-line rework of material and with the objective to minimize yield losses during product transitions. As polymer reactors are tasked to produce several grades of a polymer recipe management, production scheduling, and production optimization (Flores-Tlacuahuac, Biegler, & Saldvar-Guerra, 2005) will become increasingly important in practice. In addition, the increasing computing power provided by DCS manufacturers and the availability of powerful supervisory computers interfaced with the DCS can make the implementation of model based control and especially nonlinear MPC much easier and faster. (5) As new polymer reactors are designed (e.g., to use supercritical carbon dioxide as the reaction medium as recently reviewed in Kemmere and Meyer (2005)) or as new specialty polymers are designed for specic end-use applications where time to market is of essence, it would be benecial to incorporate process control considerations in the process design phase, so that the controllability of the polymer reactor can be established before it is actually constructed and costly process control problems after startup can be avoided. The formulation and practical implementation of a control strategy is also an important area of research for some of the newer polymer processes such as the manufacture of polymers with tailor-made properties using novel metallocene catalysts or living radical polymerizations systems by means of radical addition and fragmentation transfer (RAFT) agents or degenerative chain transfer agents as described in Matyjaszewski and Davis (2002), Mueller, Zhuang, Yan, and Litvinenko (1995a), and Mueller, Yan, Litvinenko, Zhuang, and Dong (1995b). (6) Furthermore, it is expected that the on-line control not only of average polymer properties but also of polymer distributions such as the particle size distribution and the branching distribution will become important. As shown in recent work by Doyle, Harrison, and Crowley (2003) as well as Edouard, Sheibat-Othman, and Hammouri (2005), on-line sensors used to measure the PSD, such as capillary hydrodynamic fractionation, usually require taking samples and the analysis time is still too long for control. Fruitful areas of research would include the combination of validated mathematical models with existing on-line sensors to obtain a continuous estimation of the PSD and to detect particle nucleation. It is certain that the instrumentation and control methodologies that will be needed to be deployed to meet the evolving needs of polymer producers are a challenging and vibrant area

1462

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463 Flores-Cerrillo, J., & MacGregor, J. F. (2003). Within-batch and batch-to-batch inferential-adaptive control of semibatch reactors: A partial least squares approach. Industrial and Engineering Chemistry Research, 43, 3334. Flores-Tlacuahuac, A., Biegler, L. T., & Saldvar-Guerra, E. (2005). Dynamic optimization of HIPS open-loop unstable polymerization reactors. Industrial and Engineering Chemistry Research, 44, 2659. Fotopoulos, J., Stenger Jr., H. G., & Georgakis, C. (1998). Use of tendency models and their uncertainty in the design of state estimators for batch reactors. Chemical Engineering and Processing, 37, 545. Harris, K. R., & Palazo lu, A. (1998). Studies on the analysis of nonlinear g processes via functional expansions-III: Controller design. Chemical Engineering Science, 53, 4005. Harris, K. R., & Palazo lu, A. (2003). Control of nonlinear processes using g functional expansion models. Computers and Chemical Engineering, 27, 1061. Hipp, A. K., Storti, G., & Morbidelli, M. (2002a). Acoustic characterizations of concentrated suspensions and emulsions. 1. Model analysis. Langmuir, 18, 391. Hipp, A. K., Storti, G., & Morbidelli, M. (2002b). Acoustic characterizations of concentrated suspensions and emulsions. 2. Experimental validation. Langmuir, 18, 405. Houston, W. E., & Schork, F. J. (1987). Adaptive predictive control of a semi batch polymerization reactor. Polymer Process Engineering, 5, 119. Kammona, O., Chatzi, E. G., & Kiparissides, C. (1999). Recent developments in hardware sensors for the on-line monitoring of polymerization reactions. Journal of Macromolecular ScienceReviews in Macromolecular Chemistry, C39, 57. Kemmere, M. F., & Meyer, T. (2005). Supercritical carbon dioxide in polymer reaction engineering. Weinheim, Germany: Wiley-VCH. Kourti, T., Nomikos, P., & MacGregor, J. F. (1995). Analysis, monitoring, and fault diagnosis of batch processes using multiblock and multiway PCA. Journal of Process Control, 5, 277. Lee, M. H., Han, C. H., & Chang, K. S. (1997). Hierarchical time-optimal control of a continuous copolymerization reactor during start-up or grade change operation using genetic algorithms. Computers and Chemical Engineering, 21, S1037S1042. Lee, M. H., Han, C. H., & Chang, K. S. (1999). Dynamic optimization of a continuous polymer reactor using a modied differential evolution algorithm. Industrial and Engineering Chemistry Research, 38, 4825. Leffew, K. W., Zumsteg, F. C., Feiring, A. E., Crawford, M. K., Farnham, W. B., Petrov, V. A., Schadt, F. L., & Tran, H. V. (2005). The impact of uoropolymers on Line Edge Roughness in 193 nm imaging. In Proceedings of the 22nd Conference on Photopolymer Science and Engineering. Lewin, D. R., & Bogle, D. (1996). Controllability analysis of an industrial polymerization reactor. Computers and Chemical Engineering, 20 (Sup[B]), S871. Li, R., Corripio, A. B., Henson, M. A., & Kurtz, K. J. (2004). On-line state and parameter estimation of EPDM polymerization reactors. Journal of Process Control, 14, 837. Ling, W. -M., & Rivera, D. E. (1998). Control relevant model reduction of Volterra series models. Journal of Process Control, 8, 79. Lipt k, B. G. (2003). Instrument engineers handbook, 4th edition: Process a measurement and analysis, I. New York: CRC Press. Mahadevan, R., Doyle III, F. J., & Allcock, A. C. (2002). Control-relevant scheduling of polymer gradetransitions. AIChE Journal, 48, 1754. Maner, B. R., Doyle, F. J., Ogunnaike, B. A., & Pearson, R. K. (1996). Nonlinear model predictive control of a simulated multivariable polymerization reactor using second-order Volterra models. Automatica, 32, 1285. Maner, B. R., & Doyle, F. J. (1997). Polymerization reactor control using autoregressive-plus Voltera-based MPC. AIChE Journal, 43, 1763. Maschio, G., Bello, T., & Scali, C. (1992). Optimization of batch polymerization reactors: Modelling and experimental results for suspension polymerization of methyl methacrylate. Chemical Engineering Science, 47, 2609. Maschio, G., Bello, T., & Scali, C. (1994). Optimal operation strategies to control the molecular weight distribution of polymer products. Chemical Engineering Science, 49, 5071. Matyjaszewski, K., & Davis, T. (2002). Handbook of radical polymerizations. New York: Wiley Interscience.

of investigation for academic researchers and industrial practitioners alike.

Acknowledgements The authors would like to acknowledge helpful discussions with K. W. Leffew, P. D. Schnelle, Jr., and Y. Dimitratos of DuPont.

References
Adebekun, A. K., Kwalik, K. M., & Schork, F. J. (1989). Steady-state multiplicity during solution polymerization of methyl methacrylate in a CSTR. Chemical Engineering Science, 44, 2269. Arzamendi, G., & Asua, J. M. (1989). Monomer addition policies for copolymer composition control in semicontinuous emulsion polymerization. Journal of Applied Polymer Science, 38, 2019. Bartusiak, D., Fontaine, R. W., Schwanke, C. O., & Gomatam, A. R. (2003). Nonlinear model predictive control of polymerization reactors. AIChE Annual Meeting. BenAmor, S., Doyle, F. J., & McFarlane, R. (2004). Polymer grade transition control using advanced real-time optimization software. Journal of Process Control, 14, 349. Bindlish, R., & Rawlings, J. B. (2003). Target linearization and model predictive control of polymerization processes. AIChE Journal, 49, 2885. Chylla, R. W., & Haase, D. R. (1993). Temperature control of semibatch polymerization reactors. Computers and Chemical Engineering, 17, 257. Clarke-Pringle, T., & MacGregor, J. F. (1998). Optimization of molecular weight distribution using batch-to-batch adjustments. Industrial and Engineering Chemistry Research, 37, 3660. Congalidis, J. P., & Richards, J. R. (1998). Process control of polymerization reactors: An industrial perspective. Polymer Reaction Engineering, 6, 71. Congalidis, J. P., Richards, J. R., & Ray, W. H. (1989). Feedforward and feedback control of a solution copolymerization reactor. AIChE Journal, 35, 907. Defaye, G., Regnier, N., Chabanon, J., Caralp, L., & Vidal, C. (1993). Adaptivepredictive temperature control of semi-batch reactors. Chemical Engineering Science, 48, 3373. Dimitratos, J., Elicabe, G., & Georgakis, C. (1994). Control of emulsion poly merization reactors. AIChE Journal, 40, 1993. Doyle, F. J., Ogunnaike, B. A., & Pearson, R. K. (1995). Nonlinear model-based control using 2nd-order Volterra models. Automatica, 31, 697. Doyle, F. J., Harrison, C. A., & Crowley, T. J. (2003). Hybrid model based approach to batch-to-batch control of particle size distribution in emulsion polymerization. Computers and Chemical Engineering, 27, 1153. Dub , M. A., Soares, J. B., Penlidis, A., & Hamielec, A. E. (1997). Mathemate ical modeling of multicomponent chain growth polymerizations in batch, semibatch, and continuous reactors. A review. Industrial and Engineering Chemistry Research, 36, 966. Edouard, D., Sheibat-Othman, N., & Hammouri, H. (2005). Observer design for particle size distribution in emulsion polymerization. Industrial and Engineering Chemistry Research, 51, 3167. Elizalde, O., Leiza, J. R., & Asua, J. M. (2004). On-line monitoring of all-acrylic emulsion polymerization reactors by Raman spectroscopy. Macromolecular Symposium, 206, 135. Ellis, M. F., Taylor, T. W., & Jensen, K. F. (1994). On-line molecular weight distribution estimation and control in batch polymerization. AIChE Journal, 40, 445. Embirucu, M., Lima, E. L., & Pinto, J. C. (1996). A survey of advanced control of polymerization reactors. Polymer Engineering and Science, 36, 433. Filippi, C., Greffe, J. L., Bordet, J., Villermaux, J., Barnay, J. L., Bonte, P., & Georgakis, C. (1986). Tendency modeling of semibatch reactors for optimization and control. Chemical Engineering Science, 41, 913.

J.R. Richards, J.P. Congalidis / Computers and Chemical Engineering 30 (2006) 14471463 McAuley, K. B., & MacGregor, J. F. (1991). On-line inference of polymer properties in an industrial polyethylene reactor. AIChE Journal, 37, 825. McKenna, T. F., Othman, S., Fevotte, G., Santos, A. M., & Hammouri, H. (2000). An integrated approach to polymer reaction engineering: A review of calorimetry and state estimation. Polymer Reaction Engineering, 8, 1. Meira, G. R. (1981). Forced oscillations in continuous polymerization reactors and molecular weight distribution control. A survey. Journal of Macromolecular ScienceReviews in Macromolecular Chemistry, C20, 207. Meziou, A. M., Deshpande, D., Cozewith, C., Silverman, N. I., & Morrison, W. G. (1996). Dynamic matrix control of an ethylene-propylene-diene polymerization reactor. Industrial and Engineering Chemistry Research, 35, 164. Moritz, H. U. (1989). Polymerization calorimetry: A powerful tool for reactor control. Polymer reaction engineering, 248. New York: VCH Publishers. Mueller, A. H. E., Zhuang, R., Yan, D., & Litvinenko, G. (1995a). Kinetic analysis of Living polymerization processes exhibiting slow equilibria. 1. Degenerative transfer. Macromolecules, 28, 4326. Mueller, A. H. E., Yan, D., Litvinenko, G., Zhuang, R., & Dong, H. (1995b). Kinetic analysis of Living polymerization processes exhibiting slow equilibria. 2. Molecular weight distribution for degenerative transfer. Macromolecules, 28, 7335. Nomikos, P., & MacGregor, J. F. (1995). Multi-way partial least squares in monitoring batch processes. Chemometrics and Intelligent Laboratry Systems, 30, 97. Ogunnaike, B. A., & Ray, W. H. (1994). Process dynamics, modeling, and control. New York: Oxford University Press. Othman, N. S., Fevotte, G., Peycelon, D., Egraz, J. -B., & Suau, J. -M. (2004). Control of polymer molecular weight using near infrared spectroscopy. AIChE Journal, 50, 654. Ozkan, L., Kothare, M. V., & Georgakis, C. (2003). Control of a solution copolymerization reactor using multi-model predictive control. Chemical Engineering Science, 58, 1207. Park, M. -J., Hur, S. -M., & Rhee, H. -K. (2002). Online estimation and control of polymer quality in a copolymerization reactor. AIChE Journal, 48, 10. Parker, R. S., Heemstra, D., Doyle III, F. J., Pearson, R. K., & Ogunnaike, B. A. (2001). The identication of nonlinear models for process control using tailored plant-friendly input sequences. Journal of Process Control, 11, 237. Peterson, T., Hernandez, E., Arkun, Y., & Schork, F. J. (1992). A nonlinear DMC algorithm and its application to a semi-batch polymerization reactor. Chemical Engineering Science, 47, 737. Qin, S. J., & Badgwell, T. A. (2003). A survey of industrial model predictive control technology. Control Engineering Practice, 11, 733. Ray, W. H., & Villa, C. M. (2000). Nonlinear dynamics found in polymerization processesA review. Chemical Engineering Science, 55, 275. Ray, W. H., Soares, J. B., & Hutchinson, R. A. (2004). Polymerization reaction engineering, past, present, and future. Macromolecular Symposium, 206, 1. Ray, W. H. (2003). Process modeling for polymerization process operation and control. AIChE Annual Meeting. Regnier, N., Defaye, G., Caralp, L., & Vidal, C. (1996). Software sensor based control of exothermic batch reactors. Chemical Engineering Science, 51, 5125.

1463

Reis, M. M., Ara jo, H. H., Sayer, C., & Giudici, R. (2004). Comparing near u infrared and Raman spectroscopy for on-line monitoring of emulsion copolymerization reactions. Macromolecular Symposium, 206, 165. Richards, J. R., & Congalidis, J. P. (2005). Measurement and control of polymerization reactors. Handbook of polymer reaction engineering (pp. 595678). Weinheim, Germany: Wiley-VCH. Richards, J. R., & Schnelle Jr., P. D. (1988). Perspectives on industrial reactor control. Chemical Engineering and Processing, 84, 32. Rodriguez, F., Cohen, C., Ober, C., & Archer, L. A. (2003). Principles of polymer systems (5th ed). New York: Tayor and Francis. Saenz de Buruaga, I., Armitage, D., Leiza, J. R., & Asua, J. M. (1997). Nonlinear control for maximum production rate of latexes of well-dened polymer composition. Industrial and Engineering Chemistry Research, 36, 4243. Schnelle, P. D., & Rollins, D. L. (1998). Industrial model predictive control technology as applied to continuous polymerization processes. ISA Transaction, 36, 281. Schork, F. J., Deshpande, B., & Leffew, K. W. (1993). Control of polymerization reactors. New York: Marcel Dekker. Schork, F. J. (1993). Reactor operation and control. Polymer reaction engineering. Dordrecht: Kluwer Academic Publishers. Seavey, K. C., Liu, Y. A., Khare, N., Bremner, T., & Chen, C. -C. (2003). Quantifying relationships among the molecular weight distribution, non Newtonian shear viscosity, and melt index for linear polymers. Industrial and Engineering Chemistry Research, 42, 5354. Seborg, D. E., Edgar, T. F., & Mellichamp, D. A. (2004). Process dynamics and control. New York: Wiley & Sons. Seki, H., Ogawa, M., Ooyama, S., Ahamatsu, K., Ohshima, M., & Yang, W. (2001). Industrial application of nonlinear model predictive control to polymerization reactors. Control Engineering Practice, 9, 819. Takeda, M., & Ray, W. H. (1999). Optimal-Grade transition strategies for multistage polyolen reactors. AIChE Journal, 45, 1776. Thomas, I. M., & Kiparissides, C. (1984). Computation of the near optimal temperature and initiator policies for a batch polymerization reactor. Canadian Journal of Chemical Engineering, 62, 284. Tyner, D., Soroush, M., & Grady, M. C. (1999). Adaptive temperature control of multi product jacketed reactors. Industrial and Engineering Chemistry Research, 38, 4337. Tyner, D., Soroush, M., Grady, M. C., Richards, J. R., & Congalidis, J. P. (2001). Mathematical modeling and optimization of a semi-batch polymerization reactor. Proceedings of the IFAC Symposium on Advanced Control of Chemical Processes 2000 (ADCHEM 2000), 3, 983, New York: Elsevier. Vega, M., Lima, E. L., & Pinto, J. C. (2001). In-line monitoring of weight average molecular weight in solution polymerization using intrinsic viscosity measurements. Polymer, 42, 3909. Villa, C. M., Dihora, J. O., & Ray, W. H. (1998). Effects of imperfect mixing on low-density polyethylene reactor dynamics. AIChE Journal, 44, 1646. Wang, Z. L., Pla, F., & Corriou, J. P. (1995). Nonlinear adaptive control of batch styrene polymerization. Chemical Engineering Science, 50 13, 2081. Wassick, J. W., Coffey, D., & Callihan, B. (2003). Nonlinear model predictive control of a commercial polymerization semi-batch reactor. AIChE Annual Meeting. Wyzgoski, F. J., Rinaldi, L., McCord, E. F., Stewart, M. A., & Marshall, D. R. (2004). Poly(n-butyl acrylate-co-carbon monoxide-co-ethylene) characterization by high-temperature two-dimensional NMR at 750 MHz. Macromolecules, 37, 846.

You might also like