Вы находитесь на странице: 1из 25

CH05CH18-Lee ARI 20 May 2014 19:21

REVIEWS Further Progress and Challenges in
Click here for quick links to
Annual Reviews content online,
Control of Chemical Processes
• Other articles in this volume
Jay H. Lee1 and Jong Min Lee2
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

• Top cited articles

• Top downloaded articles
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

• Our comprehensive search 1

Department of Chemical and Biomolecular Engineering, Korea Advanced Institute of Science
and Technology, 305-701 Daejeon, Korea; email: jayhlee@kaist.ac.kr
School of Chemical and Biological Engineering, Institute of Chemical Processes,
Seoul National University, 151-744 Seoul, Korea; email: jongmin@snu.ac.kr

Annu. Rev. Chem. Biomol. Eng. 2014. 5:383–404 Keywords

First published online as a Review in Advance on model-based control, on-line optimization, state estimation, sensors and
March 26, 2014
instrumentation, emerging applications
The Annual Review of Chemical and Biomolecular
Engineering is online at chembioeng.annualreviews.org Abstract
This article’s doi: This review covers key developments and trends in chemical process control
during the past two decades. Control methodologies and related supporting
Copyright  c 2014 by Annual Reviews. technologies are covered, and recent trends in applications are also exam-
All rights reserved
ined. After the widespread adoption of model-based techniques by industry,
control interest has begun to move beyond the traditional realm of readily
measured variables to include chemical compositions and particle features.
However, the shift is being slowed by the shortage of accurate, reliable, and
inexpensive sensing devices. Although the past two decades saw no new ma-
jor theoretical or methodological advances, several important incremental
improvements and extensions have been made to help the ripening of the
technologies developed in the preceding two decades. Control is regaining
its importance owing to society’s renewed focus on energy and the matura-
tion of various emerging technologies, but a community-wide consensus on
what general problems should be solved is lacking.

CH05CH18-Lee ARI 20 May 2014 19:21

Most chemical products are manufactured through a large number of unit operation steps, each
Model predictive of which can involve complex (and oftentimes incompletely understood) phenomena of reactions,
control (MPC): the fluid mechanics, and/or heat and mass transfer. In most processing steps, concerns for safety
de facto standard and stability exist, but the ultimate concern regards the product’s physical and chemical features,
advanced control
including chemical compositions and size and shapes in the case of particles. Controls of such
technique for process
industries that uses product qualities or features are hampered by the customary lack of reliable on-line sensors that
on-line optimization can provide direct, real-time feedback. Typically, more easily measured process variables (e.g.,
to calculate the temperatures and pressures) are regulated as surrogates to provide indirect regulations of the
optimal control moves product-quality variables. Other difficulties for control include the presence of a large number of
interacting variables, hard and soft constraints, uncertainties (owing to poor phenomenological
understandings of the process as well as unknown disturbances and noises), and nonlinearity in
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

the process dynamics. Furthermore, processing steps, which are typically sequential, oftentimes
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

become interrelated owing to reentrances, recycles, and other heat/mass integrations, demanding
careful coordination among the controls of individual units. In summary, control of chemical
manufacturing processes is made challenging by interactions among the steps involved, in addition
to the complexity and tight requirements of individual processing steps.
This review aims to describe recent progress in chemical process control and to suggest remain-
ing and emerging challenges. We cover key developments over the past 20 years or so, starting from
the mid-1990s. Recent review papers giving blanket coverage of chemical process control are rare.
Perhaps the closest are three critiques (of chemical process control theory) written in the 1970s
(1–3), which have had a strong influence on process control research in the following decades.
However, several comprehensive reviews have been written on particular techniques or aspects
of process control, including a large number on model predictive control (MPC) (4–7), nonlinear
process control (8, 9), batch process control (10, 11), state estimation (12, 13), and integration
of design and control (14). There have also been reviews on control of particular applications of
chemical processes, including crystallization processes (15), bioprocesses (16), and semiconductor
manufacturing processes (17). The references listed here are meant to be exemplary; we are sure
many other excellent reviews and surveys exist on the listed topics as well as other topics related
to chemical process control.
Most modern advanced process control techniques (e.g., MPC) are model based and use math-
ematical optimization, either implicitly or explicitly. Key components of such a control system are
(a) a dynamic model, which may be either first-principles based or empirical; (b) a state estimator
that processes sensed information into estimates of model state and parameters; and (c) a control
algorithm that computes optimal control moves based on the model, estimated system state, and
a given set of objective functions and constraints (see Figure 1). It is important that technolo-
gies for all these components are advanced in concert, as each can become a bottleneck for the
performance of the overall control system. The same comment applies to various supporting tech-
nologies, such as those for sensing, communication, computation, and optimization. Henceforth,
this review addresses advances in methodologies for all the key components of a control system,
as well as those in the supporting technologies.
The late 1970s through the mid-1990s were a sort of renaissance period for process control.
There were several significant developments, both in theory and in practice, spurred on by rapid
advances in computing and communication technologies. Some exemplary developments during
this period were the MPC and robust control theories, which had significant and far-reaching
impact on industrial practice. It was also a rare period of time when boundaries between control
theory development and industrial applications became blurred and theory and practice supported

384 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

Control input


Optimizer Model estimator
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

Objective function
and constraints
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

State estimate

Figure 1
Structure of a modern advanced control system for chemical processes. ut represents the manipulated process input, yt is the measured
process output, and x̂t is the system state. Subscript t signifies that the symbols represent their values at a discrete sample time t.

each other nicely (18). It is not surprising, then, that such a vibrant period was followed by a rel-
atively stagnant stretch, filled mostly with incremental advances, extensions, and generalizations.
Instead of new major theoretical or methodological advances, the past two decades saw more
widespread adoption of optimization-based control techniques (e.g., MPC) by a variety of chem-
ical process industry sectors. In addition, several new application areas came into focus, including
nanoscale product manufacturing, biological processes, and energy systems. A new trend emerged
in which academic process control research started with an application area and then developed
(or simply applied) theories and tools demanded by it. In the authors’ opinion, this is a positive
development given that these efforts eventually lead to the definition of new general problems to
be tackled by the community.


Here we review the recent progress on sensors, computation/communication hardware, and op-
timization software.

Optical sensors:
Sensors sensors that use
Traditionally, measurements commonly employed for process control include temperatures, pres- changes in absorbance,
sures, flow rates, and liquid levels. Technologies for measuring these variables are fairly old and reflectance,
standard by now; a review of them here would be redundant, as they can be found in several
chemiluminescence or
undergraduate textbooks and sensor handbooks. Instead, we choose to focus on the development Raman scatter by
of more recent sensors. spectroscopic methods
A next major advance in chemical process control will be driven by the availability of cheap to provide information
and reliable sensors that can measure physicochemical variables directly related to product quality, for chemical
such as chemical compositions and particle morphologies. In this spirit, the recent development of
optical sensors led by low-cost, miniature optoelectronic light sources and detectors is notable (19).

www.annualreviews.org • Control of Chemical Processes 385

CH05CH18-Lee ARI 20 May 2014 19:21

Real-time optical methods, including optical density, near infrared (NIR), mid-infrared (MIR),
ultraviolet-visible (UV-Vis), Raman, and X-ray diffraction, have been widely applied in chemical
processes owing to their high measuring speed and requirement of fewer or no sample prepara-
Smart sensors:
sensors that have some tions. Appropriate process analytical technologies are often used to preprocess and fit the massive
elements of control, spectroscopic data to calibration models for assured process monitoring and closed-loop control
computation, and/or (20).
decision making, such Optical density measurement is commonly used in the bioprocess, especially for measuring
as self-test/self-
the cell growth and biomass concentrations on-line, but it may yield poor estimates in turbid
calibration and
algorithmic media (21). UV-Vis spectroscopy works in the UV-Vis spectral region and has been used mainly
compensation with in environmental applications to monitor pollutants such as heavy metals, hydrocarbons, and
information from volatile organic compounds in air and water (22). NIR and MIR can measure the concentration of
other sensors organic species even in complex media. These IR-based spectroscopy technologies were applied to
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

on-line monitoring of fermentation reactions (23) and monitoring and control of vapor side draw
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

in a solvent recovery column (24). Raman spectroscopy is a useful technique for sensing gaseous or
liquid analytes and has an advantage over IR owing to the absence of water interference. Real-time
monitoring of heterotrophic bioreactor culture conditions via Raman spectroscopy was used to
control the bio-oil productivity (25).
In crystal size and shape control, microscopy tools have been used for imaging, including atten-
uated total reflection Fourier transform infrared (ATR-FTIR) spectroscopy, Raman spectroscopy,
and laser scattering (26). These techniques provide high-accuracy measurements in most solute-
solvent systems. ATR-FTIR spectroscopy has a fast sampling rate and good reproducibility for
various phases (liquid, solid, mixture) of a sample. Furthermore, ATR-FTIR is not affected by
the molecular forces, including the steric and hydrodynamic forces; hence, it is well suited for use
in crystallization processes (27, 28). Focused beam reflectance measurement is a laser-scattering
method that can be used to measure the number concentrations, sizes, and shapes of particles.
Because its accuracy and reliability are not affected by the concentrations of solids present, it is
convenient for monitoring and control of crystallization processes (29).
The next-generation measurement technologies also require new sensing techniques and trans-
ducer elements that can work reliably in harsh environments (30, 31). Materials with improved
thermal and chemical resistance are introduced in existing sensors for those applications requiring
direct immersion into harsh environments. For use in extraordinarily harsh environments that
preclude the use of direct, invasive sensors, noninvasive sensors are currently being developed.
For example, in molten metals, ultrasonic and nuclear sensors are being tested to measure levels
without making any physical contact with the liquid. Ultrasonic flow meters use sound waves and
induce no pressure drop. They can be used in both conductive and nonconductive fluids (32). Py-
rometers focus the energy radiated from a body onto a sensing system to measure its temperature.
It is useful when the temperature is so high that contact with the object is not permitted.
Another important advance in sensing is the emergence of smart field devices (e.g., smart
sensors) with embedded microprocessors (33). Smart valves include proportional-integral-
derivative (PID) control resident in the instrument, thus allowing the central computers to perform
more advanced process control and information-management tasks. Installations of smart instru-
ments may reduce instrumentation costs by up to 30% compared with the conventional approaches
and may also improve the performance, maintenance, and safety of a chemical plant.

Computation and Communication Hardware

During the past decade, one important trend was integration of personal computer (PC) and
control systems. Toward enterprise integration, the PC has become a new solution framework for

386 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

process control in client-server architectures. Since its advent in the mid-1970s, the distributed
control system (DCS) has promoted an open-application environment that makes accessible the
wide variety of PC object-oriented software tools (34).
Distributed control
With the continuing expansion of computing power, processor speed has now exceeded system (DCS):
what existing process control algorithms generally require. This has enabled creation of a employs a hierarchy of
virtual simulator in which the actual DCS application software can run on a PC. One PC computers, with a
can emulate up to 20 DCS controllers. The benefits of virtualization are increased server single computer
controlling 8 to 16
utilization, reduced cost/downtime, and easy implementation of operator training systems
individual control
(35, 36). loops
DCS data highways had used vendor-specific designs without any standard architecture. Under
Moving horizon
such designs, communications outside the network were costly. The chemical process industry estimation (MHE):
then adopted two established communication standards, fiber-distributed data interface and fast an optimization
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

Ethernet, from the computer industry. Both operated at 100 megabits per second, which is 10- approach to state
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

to 50-times faster than the speed of typical in-house DCS networks. Their improved speed and estimation that uses a
model and a series of
high level of reliability made them attractive for industrial control applications. They also offered
the benefit of easy access to third-party devices and corporate information networks. During the observed in a moving
past decade, the architecture has been joined by the gigabyte Ethernet. time window
NP-complete: refers
to a class of decision
Optimization Software problems that can
A wide variety of numerical optimization techniques and software have been applied to solving become unsolvable
with any amount of
optimization problems that occur within control system design or implementation. Because many
computing power
of the modern control techniques, such as MPC, require that an optimization be solved on- available today
line within a specified time interval, development of an optimization solver became an integral
part of an advanced control system design. For linear problems, linear programming or quadratic
programming problems typically result, but they are convex problems, and solvers for them are well
established by now. Hence, most recent research in this area has been devoted to the development
of reliable solvers for nonconvex, mixed-integer, differential-algebraic, bilevel, and nonfactorable
problems. Though research efforts for such optimization problems have been driven mainly by
those working in the process design and scheduling areas, several research teams have addressed
certain problem features and requirements unique to control applications. Here we summarize
the developments by such teams.
For nonlinear MPC, the convention now is to discretize the nonlinear differential equation
as algebraic equations (e.g., via the collocation method) and then solve the resulting large-scale,
sparse nonlinear program (37). Direct single shooting methods discretize control inputs to obtain
state as a function of the inputs and initial state via an integrator. They need only a small number
of optimization variables even for large systems, but they do not work well with poor initial guesses
or unstable systems. Multiple shooting methods have addressed these drawbacks and have been
used for optimal control applications (38, 39). Recent advances in nonlinear dynamic optimization
techniques have allowed solution of a large-scale dynamic optimization problem with millions of
variables (40). Nonlinear programming solvers are needed in other control-related applications,
such as in moving horizon estimation (MHE) and parameter estimation.
In certain types of applications, such as hybrid system control or supply-chain opti-
mization/control, decision variables include integer variables, thus yielding mixed-integer
optimization problems. Depending on the nature of the problem, they can be mixed-integer
linear (or quadratic) programs [MIL(Q)Ps] or mixed-integer nonlinear programs. Though these
problems are NP-complete and therefore inherently difficult to solve, several efficient and
well-tested commercial solvers, like CPLEX, should provide good solutions within a reasonable

www.annualreviews.org • Control of Chemical Processes 387

CH05CH18-Lee ARI 20 May 2014 19:21

amount of time for most interesting problems. However, because control applications require
solvers that can work in real time, the computational complexity and lack of a hard guarantee for
a converged solution still act as major barriers to practical use.


Model Predictive Control. MPC is a model-based control technique that calculates optimal
control moves by solving an on-line dynamic optimization problem at each sample time, based
on a model’s prediction of the system behavior within some future time window and an objective
function and constraints specified by the user. Though the idea had been discussed in the control
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

literature (with names like receding horizon control and open-loop optimal feedback control) and
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

even experimented with by some companies during the 1950s and 1960s, MPC did not become
a mainstream control technique until the late 1970s. This change was brought about mainly by
the advent of cheap and fast computing hardware and the push by some influential industrialists
(18, 41). MPC has had a major impact on industrial control practice, and nowadays there is
not a major refinery or a chemical plant where MPC has not made it into the control room
(4, 42).
A general form of the optimization solved by MPC at each sample time (for a linear system)
can be represented as

min T
zk+1 Qzk+1 + vkT Rvk 1.
{vk ∈V}k=0,..., p−1
with the model equation of
zk+1 = Azk + Bvk , k = 0, . . . , p − 1. 2.
V represents the allowable set for the control input defined by the various constraints that typ-
ically exist (e.g., valve limits). At a discrete sample time step t, either a direct measurement of
the model state or, more commonly, its estimate based on the measurements via state estimation
[denoted as x̂(k)] is obtained and used as the starting state z0 of the above optimization. Hence,
though the optimization solved at each time is an open-loop optimal control problem, the overall
scheme is a feedback control (possibly combined with feedforward control). In addition, although
optimal values for a series of control moves (v0 , . . . , v p−1 ) are calculated, only the first move v0
is implemented on the actual process (i.e., u t = v0 ), as the next move u t+1 is decided through a
new optimization occurring at the next sample time step. Its main attractive feature for industrial
process control is its ability to account for constraints (and possibly nonlinearities) explicitly in the
context of large-scale interactive multivariable systems commonly found in refineries and petro-
chemical plants. The widespread industrial adoption during the 1970s and 1980s was followed by
vigorous research efforts by the academic community, which contributed theoretical foundations
for constrained system control and various enhancements to the initial industrial version of MPC.
Stability theory for constrained receding horizon control was developed during the 1980s and
became fully established by the early 1990s (6, 43).
After the establishment of the stability theory and the generalized state-space formulation,
attention during the past two decades has shifted to the use of different or more general system
descriptions. For example, in nonlinear MPC, the model form assumed is generalized to
ż = f (z, v), 3.

388 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

where f is a general nonlinear operator (with properties such as being continuous and differen-
tiable). Whereas linear MPC mostly uses empirical input-output models developed based on data
from plant tests, nonlinear MPC allows for direct use of first-principles-based models, which are
important in problems like grade changeovers in polymer plants. The nonlinearity of the model dynamic
does not pose any difficulty in concept. In actual implementation, however, the nonconvexity programming (ADP):
of the resulting mathematical optimization poses numerical difficulties owing to the presence of uses simulations to
multiple local minima. Therefore, research in this area has been focused on developing a reliable focus learning to
relevant regions and
nonlinear optimization solver suitable for real-time implementation (44, 45) and devising a for-
mulation such that stability and/or near-optimality may be achieved even when a global optimal approximations to
solution is not found (38, 46, 47). parameterize the value
Another extended model form studied is the mixed logical dynamical system, which can rep- function
resent a variety of systems described by interdependent physical laws, logic rules, and operating
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

constraints, including linear hybrid systems, finite-state machines, discrete-event systems, con-
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

strained linear systems, and nonlinear systems represented by piecewise linear functions (48). An
MPC algorithm based on such a model yields a mixed-integer quadratic programming problem,
which must be solved on-line.
Finally, given the recognized importance of uncertainty in control, robust or stochastic MPC
has attracted much attention, where the model form assumed is
zk+1 = A(θk )zk + B(θk )vk + F θk , k = 0, . . . , p − 1. 4.
θ is an uncertain parameter or disturbance vector modeled either by a set membership description
(e.g., θ ∈  where  is some compact, convex set) or by some linear stochastic process driven by
a white noise vector. The optimization solved at each time step in this case is

min max T
zk+1 Qzk+1 + vkT Rvk 5.
{vk ∈V}k=0,..., p−1 {θk ∈ }k=0,..., p−1

or, in the stochastic model’s case,


min E T
zk+1 Qzk+1 + vkT Rvk . 6.
{vk ∈V}k=0,..., p−1

The idea here is to include a quantitative description of uncertainty in the model to account for it
directly in the optimization proactively rather than to respond reactively through feedbacks. As it
turns out, MPC’s open-loop control assumption is very limiting in this generalized model’s context
(49). Though some interesting partial solutions have come along (50–52), a general solution seems
to be out of reach. This recognition has motivated researchers to explore new approaches, such
as approximate dynamic programming (ADP), discussed later (53).
The past two decades also saw the more widespread adoption of MPC not only within the
chemical process industries but also by industries like automotive and other mechanical parts,
wherein applications typically require very fast sampling rates. The latter demanded that compu-
tational efficiency of the on-line optimization solver of MPC be improved by orders of magnitude.
This was achieved through several approaches, including the development of customized solvers
taking advantage of the special structure of the optimization involved (54, 55), or the so-called
explicit MPC, which used the multiparametric programming approach to solve the optimization
off-line for the entire state space. In the latter, explicit state feedback laws, typically linear, are
defined for different polyhedral regions of the state space. This way, the on-line optimization re-
quirement could be replaced by a simpler table lookup (56). One problem with the explicit MPC
approach is that the number of polyhedral regions grows quickly as the horizon size, the number

www.annualreviews.org • Control of Chemical Processes 389

CH05CH18-Lee ARI 20 May 2014 19:21

of states, and the number of constraints grow, complicating the implementation of table lookup.
Various researchers have addressed this problem, suggesting efficient search algorithms that speed
up the search with some minor sacrifice in optimality. A detailed review on all these aspects of
programming (DP): MPC can be found in a recent survey by J.H. Lee (7).
a method to solve
nested multistage Dynamic Programming–based Control. The robust control problem (defined by Equations 4
decision problems by and 5) and the stochastic optimal control problem (defined by Equations 4–6) posed earlier render
decomposing them
themselves naturally to the approach of dynamic programming (DP), which is the only general
into a successive series
of single-stage way known for constructing optimal feedback control laws for stochastic systems (57). The main
problems idea of DP is to solve for the value function (also called the cost-to-go function), representing the
finite-horizon or infinite-horizon cost achievable under optimal control starting from a particular
state, which is the argument of the function defined over the whole state space. Obtaining a value
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

function requires the solution to an optimality equation referred to as the Bellman equation, which
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

can seldom be solved analytically and poses a set of numerical difficulties known as the curse of
dimensionality (COD).
In spite of the COD, if a general solution strategy could be found that yields even good
approximate solutions, it would represent a major advance in optimal control. Research efforts
to find such an approximate solution strategy have been made during the past two decades or so,
under the collective name of ADP (58). The basic ideas of ADP are (a) to use simulations (e.g.,
Monte Carlo runs) to focus the learning of the value function on those regions of the state space
with high tendency or likelihood of being visited during control and (b) to use some function
approximator to parameterize the value function and interpolate among the learned sampled
points. The research efforts were driven initially by the artificial intelligence community, whose
main interest was reinforcement-based learning of optimal decision strategies. The operations
research community later joined the effort, given a wide array of multistage stochastic decision
problems found in scheduling and resource allocation areas. ADP has been shown to be useful
in several applications, including elevator dispatch problems, robot pathfinding problems, and
resource-constrained project scheduling problems. The basic problem formulation studied by the
ADP community is the Markov decision process, which involves discrete state and action spaces and
probabilistic state-transition functions. Though this is natural for the decision problems studied
by the artificial intelligence and operations research communities, it is not suitable for process
control problems, which are typically cast in continuous-state and action-space settings. The
COD gets even more severe for these cases as the size of discretized space explodes exponentially
with the dimension of the continuous space. Recent works by the authors of this review have
attempted to bring awareness of the potentials of the approach to the process control community
and to modify the ADP methods so that they are better suited for process control problems (53,
59, 60). Though much more development and extensive testing are needed, the approach holds
much promise for improved control of chemical processes, which are inherently nonlinear and

Iterative Learning Control for Batch Processes. Batch processing has become ubiquitous with
the rise of industries like biotechnology, fine chemicals, and microelectronics. Batch processes
pose some additional barriers to deployment of advanced control methods like MPC, including
the lack of a steady state (thus accentuating the nonlinearities of process dynamics) and emphasis
on end-product quality, which can seldom be measured directly on-line. Most batch processes
are controlled indirectly by assigning predefined trajectories for certain measured variables, such
as processing temperature, and then trying to follow them as closely as possible. In addition, lab
analysis results of end-product qualities from previous batch runs, when they become available,

390 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

are used in the run-to-run (R2R) control layer to correct for slow drifts or bias in the process
equipment and in the feedstock.
Following an assigned trajectory during processing is not a straightforward task in itself. The
Iterative learning
task is often made difficult by nonlinearities and non-minimum-phase dynamics (e.g., delays) control (ILC):
inherent to these processes. Significant tracking errors typically result, and these errors tend a method of tracking
to repeat themselves as multiple identical or similar runs are performed in sequence. Iterative control used for
learning control (ILC), which was originally developed for training robot-arm manipulators, can systems operating in a
repetitive mode
be used to overcome this problem, but typically it had been studied in the frequency domain
in the context of the single-input-single-output (SISO) system and therefore was not considered Kalman filter (KF):
a recursive state
suitable for chemical process control problems where effects of nonlinearities and constraints tend
estimator developed
to dominate the performance. However, recent efforts by several researchers have given an MPC primarily by Rudy
formulation equipped with the attractive feature found in ILC (10, 61). Convergence to perfect Kalman; an optimal
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

trajectory tracking was proved under certain assumptions for such controllers. The controllers estimator for linear
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

were also demonstrated experimentally on several batch-control applications, including chemical stochastic systems with
Gaussian white noises
reactors and rapid thermal processing (RTP) systems, in which good temperature tracking is
regarded as both difficult and critical (62, 63).

State Estimation
A state estimator is a key component of a modern model-based control system. Most modern
control techniques (e.g., MPC) assume an availability of exact state information at each sample
time, which is seldom the case for a chemical process. In this case, a state estimator is used to
convert the measurement information into an estimate of the whole state. As such, accurate state
estimation is critical to achieving good overall closed-loop performance. A state estimator can
also be used for the purpose of soft-sensing, monitoring, fault detection, and diagnosis. The stan-
dard techniques for state estimation have been the deterministic Luenberger observer and the
Kalman filter (KF). In particular, the KF has been the workhorse because it is simple to imple-
ment and gives optimal estimates for linear Gaussian systems. Adoption of KF by the chemical
process industries has been slow, however, having been hampered by a couple of obstacles. First
is the difficulty in constructing a linear stochastic system model (driven by external white noise
signals) that is appropriate for a given chemical process. The role of white noise has often been
misunderstood by engineers without adequate training in the subject of stochastic processes, and
the result often has been disastrous. Second is the strong nonlinearity exhibited by most chemical
processes. For a strongly nonlinear process, one typically uses the extended Kalman filter (EKF),
which is a modified version of the KF based on relinearization of the nonlinear model at each
time step. Though popular, EKF has shown poor performances at times, especially during initial
phases when the estimates and thus the linearized models are highly inaccurate (64). The Gaussian
distribution assumption inherent to Kalman filtering can also be limiting at times. Progress has
been made on both fronts. Many papers have described some of the potential pitfalls in apply-
ing Kalman filtering to a chemical process. In addition, several alternatives to EKF have been
proposed, which we describe briefly here.

Moving Horizon Estimation. The idea of formulating and solving an optimization on a moving
time window, a cornerstone of MPC, has also been contemplated for use in state estimation and
has led to a technique known as MHE. MHE finds optimal estimates of the unknowns, either
the state vector sequence or external noise sequence, that are most consistent with measurements
obtained in a specified time window. The difference is that, in the case of MHE, the time window
of optimization covers a past time period (with respect to the present time) as opposed to a future

www.annualreviews.org • Control of Chemical Processes 391

CH05CH18-Lee ARI 20 May 2014 19:21


Error Predicted state trajectory

{zk+i }i=1,…,p

Control input trajectory

{vk+i }i=0,1,…, p–1

Decision variable

Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

t t+1 t+2 t+3 t+p

Access provided by University of Saskatchewan on 03/05/17. For personal use only.

Past Future time window

Current time


Predicted output trajectory

{ yt=i }i=0,1,…,N

… Error

Smoothed state trajectory

{xˆk–i }i=0,1,…, N

Decision variable

t–N t–3 t–2 t–1 t

Past time window Future

Current time

Figure 2
Comparison of the optimizations formulated for model predictive control (MPC) and moving horizon estimation (MHE).

one in MPC (see Figure 2). For linear systems, the resulting optimization is a quadratic program.
Without any extra constraint applied other than the model, the state estimate built from the
optimization solution can be shown to be equivalent to that from the KF given that the weight
matrix for the initialization error is chosen in a specific way (65). An advantage of MHE is that
additional constraints can be enforced to ensure physical consistencies or even to relax the Gaussian
distribution assumption (66). For nonlinear systems, a nonlinear model can be applied directly
without linearization within the time window (67). MHE has been found to outperform the KF in
several studies (68). The larger the estimation window of MHE, the higher the improvement over
the EKF, in general. However, a larger window means a larger optimization problem to solve, and
the optimization to be solved is a nonlinear program for a nonlinear system. Hence, the method
presents the user with a classical trade-off between accuracy and computational complexity.

392 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

A major theoretical complication in MHE is that the requirement of a finite-size time window
forces the estimator to discard the measurement information outside the window or to summarize
it in the form of a probability density function for the initial state (or, equivalently, the so-called cost
Particle filters:
to arrive) (69). This can be thought of as an analog to the cost-to-go function in control. Because sequential
it is generally impossible to obtain an exact form of the arrival cost, it must be approximated. This sampling-based
may after all require linearization of the model, as in the EKF. However, if the chosen window methods that use point
size is sufficiently large, the initialization error for a state far into the past is likely to have a much mass representations
of probability densities
smaller effect on the accuracy of the estimate for the current time. This is the reason for improved
(called particles) to
accuracy over the EKF, (67). propagate the
probability density
Unscented Kalman Filter. Unscented Kalman filter (UKF) has been suggested as an alternative
to EKF (70). The main idea of UKF is to select 2n + 1 (where n is the state dimension), the so-
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

called sigma points in the state space based on the current mean and covariance; to integrate them
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

forward for one time step; and to use the results to calculate the weighted mean and covariance
for the new time step. After that, similar sample-based calculations are performed to update the
mean and the covariance based on the new measurements. Hence, unlike the EKF, which requires
the calculation of the Jacobian matrices at each time step, the UKF is a derivative-free method.
However, it requires 2n + 1 integrations at each time step, as opposed to the single integration
required by the EKF. A significant number of cases are reported in which the UKF showed
improved performance over the EKF, although the UKF is developed and tested mostly for
aerospace navigation and tracking problems. For chemical processes, Romanenko and coworkers
(71, 72) tested the UKF on a nonisothermal continuous stirred tank reactor and also on a pH
system; they reported significant performance enhancements over the EKF.

Estimation of Non-Gaussian Distributions. An assumption underlying Kalman filtering is

that the distribution is Gaussian. Thus, the KF and its extensions provide only a point estimate
and covariance at each time. This is sufficient for a Gaussian system, but such information may
be limiting for highly non-Gaussian systems (e.g., multimodel distributions). For such cases, an
entire probability distribution in the state space, if available, can be highly informative. One
technique that can provide such information is particle filtering (PF) (73). The sampling is done
according to the so-called importance function, which puts higher emphasis on those regions
of high-probability densities. Particles are moved to the next time step via model integration,
and weights for the propagated particles are then adjusted based on the obtained measurements
according to Bayes’ rule. It was originally developed for signal processing applications but has
been considered recently for process control applications (74).
One problem with PF is that as the probability density tends to become more and more con-
centrated in a specific region as measurements are added, variances among the weights increase
with time. This phenomenon, known as degeneracy, can be dealt with by removing those sam-
ples with small weights and equalizing weights by adding more samples (73). Another problem
is the exponentially increasing sample-number requirement with state dimension to maintain
accuracy; this makes it all but impossible to apply PF to a high-dimensional system. For this,
some hybrid approach that can use both parametric and nonparametric representations of the
probability density may be needed. A bigger question perhaps is how to best use an entire prob-
ability density function in the control calculation, if available. This is not a simple issue, as the
control moves affect the temporal and spatial propagation of the density function, and hence con-
trol and estimation become intertwined once the probability distributions in the state space are

www.annualreviews.org • Control of Chemical Processes 393

CH05CH18-Lee ARI 20 May 2014 19:21


Refineries and Petrochemicals

Refineries and petrochemical plants involve complex networks of processes that convert oil into
petroleum products (75). A moderately sized refinery has on the order of hundreds of thousands
of sensors and tens of thousands of actuators. Most refineries use the PID controllers to control
basic variables, such as flows, pressures, levels, temperatures, and some quality variables. Advanced
SISO control structures, including dead-time compensators, feedforward, and overrides, had been
used for over half a century to enhance the basic feedback control (76), but since the 1980s, MPC
has become the method of choice for difficult, multivariable control problems with significant
interactions between inputs and outputs (77). Though small-sized model predictive controllers are
implemented on a DCS directly, they are more often implemented on a separate workstation that
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

can communicate with the DCS, given their sometimes extensive computational requirements.
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

However, newly introduced DCSs with improved computational power are starting to include
MPC at the function block level (77).
When the feed or process changes that excite nonlinear process behaviors occur frequently,
a real-time optimizer (RTO) is often implemented to determine the optimal operating point
under changing conditions (75). The RTO is typically a nonlinear program based on steady-state,
plant-wide models, often with several hours between implemented solutions. For example, in
the fluidized catalytic cracker unit, the cracking reaction rates involve the exponential Arrhenius
expression (75). A typical implementation of RTO involves the periodic execution of an algorithm
that gathers data from the steady operation, reconciles the gathered data, adapts some model
parameters so that the model calculations match the reconciled data, calculates the optimum
targets for the process, and modifies the process set points as necessary (78). It is common to
find RTO applications with fewer than 100 decision variables, more than 10,000 total variables,
more than 100,000 constraints, and more than several hundred thousand nonzero Jacobian-matrix
entries (75). The complexity of large-scale nonlinear models also motivated some practitioners to
adopt linear dynamic models used in the predictive controller and collapse the MPC and RTO
levels into a single application (75).
Refinery-wide planning and scheduling have also been studied to determine optimal targets
for the individual refinery processing units with the goal of meeting given product demand vol-
umes and specifications. This optimization is often based on a linear program (75). Because the
scheduling often integrates continuous and batch operations, mixed-integer optimization tech-
niques are used. The combinatorial aspect of the formulation has thus limited the applications
to subsystems of a plant with considerable simplifications (79). Despite many developments of
nonlinear planning models, most of them were eventually reverted to linear models owing to the
lack of skills and staff for data validation and tool support (79). Another challenge to be addressed
for exploiting rigorous planning models is how the uncertainties, such as market demands and
price, can be represented effectively within the optimization framework.

Polymers and Chemicals

In the polymer and chemical industry, the measurement and control of quality variables have
been considered to be difficult owing to the complex kinetics and reaction mechanisms involved.
The main challenge is that the exact reaction mechanisms are difficult to identify and extremely
sensitive to the operating conditions. Also, many physical variables related to product properties
(e.g., viscosity, composition, molecular weight, and particle size) are difficult to measure on-line.

394 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

PID control enhanced with steady-state model-based feedforward control, ratio control, and
cascade control has been used mostly owing to its simplicity (80). A feedforward-feedback super-
visory control can be added for the control of polymer composition and molecular weight (81).
Soft sensor designs for implementation of quality control schemes have been one of the vibrant
research areas during the past decade. Data-driven approaches, such as multivariable statistical
methods like the partial least squares and artificial neural networks, have been popular (82–84).
State estimation techniques based on first-principles models, such as the KF, EKF, and MHE,
have also been considered for this purpose (85, 86). Grade changeover operations are common
but challenging tasks found in polymer production plants. Grade changes typically accompany
large settling times and/or overshoots and consequently leave significant amounts of off-spec
products. It is necessary to change the operating conditions as safely and quickly as possible while
maintaining the quality within an acceptable range. Model-based optimal control schemes, often
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

involving nonlinear mixed-integer programs, have been investigated as a promising approach for
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

this difficult problem (87, 88).

Owing to the uncertainty associated with the chain termination mechanism, the radical poly-
merization process, which has been the standard for synthesizing general-purpose polymers,
presents innate challenges in terms of controlling quality variables, such as composition and
molecular weight. To overcome the limitation, new processes based on anionic polymerization
and controlled radical polymerization have been developed (89). Though quality controls in these
new processes are significantly easier, there still is a need for soft sensor design and control methods
suited to them.

Process control in semiconductor manufacturing processes developed relatively late compared
with other processes for the following reasons: In situ measurements of important process variables
are rarely available; product specifications are extremely tight, pushing the limits of measurement
technology; and the processes drift with process runs (90, 91). The R2R controller combines
the advantages of both statistical process control and feedback control to reduce the variability
of the process output and compensate for drift (92). For R2R control techniques, exponentially
weighted moving average (EWMA) is the most popular choice in the semiconductor industries.
EWMA was proposed for quality-monitoring purposes in the 1960s (93), but it has been used
as a basis for R2R control since the 1990s (94, 95). For processes with severe drifts, the EWMA
controller is insufficient even when large weights are used. This problem becomes more severe
when there is a measurement delay, which is almost inevitable in semiconductor manufacturing.
To better control drifting processes, the predictor-corrector controller (PCC) was developed (94).
The PCC uses a second exponential filter to forecast the trend in the estimated intercept in an
attempt to predict how the intercept will change in the future. Double EWMA (d-EWMA), a
modification from the PCC, was also suggested (96). The advantages of d-EWMA controller over
the PCC are that d-EWMA controller is a direct form of integral-double-integral controller and
that the asymptotic behavior of the d-EWMA parameters have more straightforward meanings.
The d-EWMA-based R2R control was later generalized in a stochastic optimal control’s setting
and made more robust (97).
For those processes that repeat an operation multiple times, such as the RTP system, ILC can
be used to improve the temperature-tracking performance from one run to next. ILC is similar to
R2R control in concept, except that the former addresses trajectory-tracking problems, whereas
the latter addresses end-quality-control problems (10).

www.annualreviews.org • Control of Chemical Processes 395

CH05CH18-Lee ARI 20 May 2014 19:21

Another big issue in semiconductor manufacturing control is the lack of in situ measurements
to provide real-time information on the wafer status. Moreover, in most semiconductor processes,
the product quality data from the previous run are not available before the start of a new run. To
counteract these problems, researchers developed an adaptive R2R control approach employing
the sliding window technique for recursive estimation of the probability density function of the
output tracking error (98) and a virtual metrology–embedded R2R control approach (99).

Pharmaceuticals and Biotechnology

Most bio- or pharmaceutical processes are operated in a fed-batch or batch mode, involving a
large number of manual operations. The lack of on-line sensors for concentration measurements
has been the main bottleneck for implementing advanced control techniques to improve the yield
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

of bio- or pharmaceutical processes. As such, much research in this area has been devoted to the
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

development of process analytical technology (16, 100). For example, spectroscopic techniques,
such as NIR spectroscopy and Raman spectroscopy, have emerged as valuable tools for quality
control, potentially replacing expensive and time-consuming methods such as chromatography
and mass spectroscopy (101). On-line estimation of unmeasurable variables based on a process
model, via techniques like Kalman filtering, can also complement the spectroscopy data and has
been used for effective control strategies (102). These estimation techniques were extended even
further to estimate intracellular flux distributions (103).

Renewable energy (e.g., energy derived from biomass, sunlight, wind, waves, and geothermal
heat) has continually gained public and scientific attention since the oil crisis in the late 1970s;
lately, that interest has surged owing to various environmental concerns posed by increased fossil
fuel consumption, such as global warming. Future energy sustainability depends heavily on how
renewable energy can be harvested on a large scale. Among several technical issues industry
must address to realize this goal, control is regarded as one of the key enabling technologies
for the deployment of renewable energy systems (104). For example, despite an estimated 200
companies working on algal biofuels in 2010, a scalable and financially viable industry is yet to
emerge (105). One of the imminent issues to resolve is the reduction of capital and operation
costs. Toward this objective, several studies have focused on modeling microalgae growth and
optimizing biomass productivity (106). Effective controls of processes for cultivation, harvesting,
and further processing will be key. A novel approach to using intracellular metabolism for deriving
an optimal control strategy is also under way (107).
Solar and wind powers also require effective use of advanced control techniques for reliable
operation. Wind turbines work under turbulent and unpredictable environmental conditions and
are connected to constantly varying electrical grids. Regulation of the amount of power capture
by properly varying rotor speed and maximization of power production are among the control
problems for the wind turbine (108, 109). Solar-sourced electricity can be generated either directly
using photovoltaic cells or indirectly by collecting and concentrating the solar power to produce
steam. The latter bears more relevance to control, as it yields many interesting problems, such
as sun tracking and control of thermal variables. The solar plants exhibit time-varying dynamics,
nonlinearities, and uncertainties. Currently, classical PID control is employed and often detuned
for regulations (110). Uncertainty and intermittency are major difficulties for both solar and wind
generation systems. Although the smart grid is a solution from the perspective of system design,

396 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

advanced control strategies are also needed to cope with these issues and exploit the full potential
of the renewable resources (104).


Optimal control of nonlinear stochastic systems, or more generally, nonlinear uncertain systems,
remains the Holy Grail of control research. The problem is complicated by several issues, not only
the difficulties associated with state estimation and control calculations (with full state information)
on their own but also those arising from the intricate coupling between the two. Most current
control techniques use the so-called certainty equivalence principle, which assumes a point distri-
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

bution and therefore obscures this important issue. Some of the difficulties were revealed earlier
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

during the discussions of robust MPC. ADP and PF offer a viable direction toward a solution,
but several theoretical and computational challenges remain for them to constitute a complete
solution. Each incremental advance and new insight can bring us closer to the goal and also allow
better solutions to practical control problems.
Another outstanding problem involves multiscale systems. Multiscale modeling and analysis
have been popularized by those working in the computational materials development arena who
are trying to link quantum and density function theory calculations to molecular dynamics and
Monte Carlo simulations. The multiscale issue for control arises in decision problems that involve
decisions at multiple timescales. An immediate issue with such a problem is that the number of
decision variables can become intractable quickly as fine-scale decisions demand the use of a short
time interval between stages, whereas coarse-scale decisions and dynamics demand the use of a very
long decision horizon. Recently, some researchers (111, 112) pointed out a similar issue common
among many energy supply-chain optimization problems. Once again, ADP was indicated as a
viable direction to explore for such problems.
On the state estimation side, emergence of nanosystems and nanosensors may present us with
an interesting theoretical challenge. At nanoscales, behavior of a sensor may be highly stochastic,
as it may be governed by a probabilistic description like the chemical master equation (CME).
In such a case, the usual description of additive white noise may be inadequate, and the CME
may be needed directly within state estimation. Some interesting results have begun to emerge in
the context of single-walled carbon nanotube–based sensors developed for detection of important
molecules like NO at nanoscale.

Emerging Applications
Several chemical companies have changed or expanded their product portfolios to cover functional
materials and products with a high level of complexity; examples include nanomaterials, lab-on-a
chip for medical diagnosis, electronic books, fuel cells, and battery systems (113). Therefore, it is
necessary to provide a viable design and control method for those processes making such novel ma-
terials or incorporating them into devices. Such a novel process may require a multiscale-modeling
point of view, starting from the molecular or electron level and spanning the entire range of process
operations, because control of molecular structure is the only means to achieve a desired nanoscale
pattern the product will take. A major challenge in this regard is that average bulk properties can-
not be used for the characterization and manipulation of the supramolecular unit operation (114).
Given the lack of sensors and actuators at such fine scales, it becomes very difficult to ensure good
control of molecular-level structure or behavior throughout an entire spatial domain. This perhaps

www.annualreviews.org • Control of Chemical Processes 397

CH05CH18-Lee ARI 20 May 2014 19:21

explains why so few of these novel materials, despite the abundance in the research arena, have
been commercialized. A systems approach, referred to as the energy-gap-maximization method,
was successfully applied to design a statistically stable nanostructure exploiting molecular config-
uration. It searches for the structure that results from a self-assembly process with a high-energy
barrier sufficient to keep the structure from collapsing (115). For the control technology to be-
come an enabling tool for systems involving multiscale, multiphase phenomena, much progress is
still required in several areas, including computational power, compatible simulation platforms,
signal processing, sensor/actuator arrays, and control-oriented modeling (116, 117).
At the other end of the scale spectrum, supply-chain management may be an area that can benefit
from advanced control techniques. Supply-chain management is an important aspect in the process
industries involving multiproduct, multipurpose, and multisite facilities operating in different
regions. Early studies mainly focused on developing mathematical programming formulations,
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

such as MILPs, and solving operational issues, such as facility selection, equipment use, and
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

manufacture and distribution of products. Development of steady-state, dynamic, multiperiod,

or multisite production supply-chain and planning models has been the main theme of the field
during the past decades. As more complex problems were considered, the resulting optimization
problems were often described as mixed-integer nonlinear programs (118).
Deterministic models have limitations in providing expected returns and evaluating associated
risks for systems with significant uncertainties in, for example, demand, prices, and raw material
availability. Stochastic programming involving two-stage formulation with a selected number of
scenarios has been the popular choice (119). However, many interesting planning and scheduling
problems require a large number of decision stages and cannot be captured adequately by a
two-stage decision problem. The stochastic programming approach tends to come up short for
multistage problems, as the number of scenarios and decision variables explodes in an exponential
manner with the number of stages. A potentially advantageous approach in this regard is the
ADP approach, which was mentioned earlier for stochastic optimal control problems. Because the
ADP approach can seamlessly incorporate Monte Carlo simulations into the learning of optimal
decision policies, it is not limited by the number of scenarios as the stochastic programming
approach is. The handling of uncertainty still presents future research challenges for large-scale,
multistage, multiperiod supply-chain problems. Other interesting applications related to planning
and scheduling are found in the areas of sustainability and resource efficiency, such as energy
resource planning (111) and health care (120).

1. Major progress in process control for approximately the past four decades has been
driven by advances in computational and communication hardware and optimization soft-
ware. Use of complex control algorithms that require on-line solutions to optimization
problems has spread fast and is considered routine today, even in large-scale industrial
2. Control interest is starting to move beyond the traditional realm of readily measured
physical variables (e.g., temperature, pressure, and flow rates) to include, for example,
chemical compositions and particle size and shapes. The main limiting factor is the
availability of accurate, reliable, and inexpensive sensing devices for them. In addition,
models, and therefore control algorithms for controlling these variables, are necessarily
more complex, though this is less a factor than the sensing technology.

398 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

3. In contrast to the period from the mid-1970s to the mid-1990s, the past two decades have
not seen any new major theoretical or methodological advances. However, the major
developments from the previous era, such as MPC and MHE, have been extended and
matured through several incremental improvements and generalizations. These tech-
nologies have been advanced to the point that they are readily accepted by industry.
4. Main theoretical challenges remain in the area of robust control of large-scale nonlinear
systems and/or highly stochastic systems driven by the needs in emerging applications.
Such challenges extend beyond control to the supporting technological areas, such as
modeling and state estimation.
5. Control is regaining its importance, as it is expected to play a major role in commercial
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

scale-ups of new discoveries in emerging areas such as renewable energy, materials, and
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

bio-related fields, which are finally beginning to mature. Opportunities abound as they
present new challenges for process control research. However, no consensus has been
formed on what new general problems should be tackled. It may be timely for the process
control community to engage in systematic and across-the-board discussion of this.

The authors are not aware of any affiliations, memberships, funding, or financial holdings that
might be perceived as affecting the objectivity of this review.

This research was supported by the Advanced Biomass R&D Center of the Global Frontier
Project funded by the Ministry of Education, Science and Technology (ABC-0031354) and the
Basic Science Research Program through the National Research Foundation of Korea (NRF)
funded by the Ministry of Education, Science and Technology (20110006839).

1. Foss AS. 1973. Critique of chemical process control theory. AIChE J. 19:209–14 4. Presents a classic
2. Lee W, Weekman VW. 1976. Advanced control practice in the chemical industry: a view from industry. review on model
AIChE J. 22:27–38 predictive control
theory and practice up
3. Kestenbaum A, Shinnar R, Thau FE. 1976. Design concepts for process control. Ind. Eng. Chem. Process
to the year 2000.
Des. Dev. 15:2–13
4. Morari M, Lee JH. 1999. Model predictive control: past, present, and future. Comput. Chem.
Eng. 23:667–82 6. Points out the
limitation posed by the
5. Qin SJ, Badgwell TA. 2003. A survey of industrial model predictive control technology. Control Eng.
open-loop control
Pract. 11:733–64
assumption of model
6. Mayne DQ, Rawlings JB, Rao CV, Scokaert POM. 2000. Constrained model predictive control: predictive control.
stability and optimality. Automatica 36:789–814
7. Lee JH. 2011. Model predictive control: review of the three decades of development. Int. J.
7. Provides one of the
Control Autom. Syst. 9:415–24
latest reviews on model
8. Bequette WB. 1991. Nonlinear control of chemical processes: a review. Ind. Eng. Chem. Res. 30(7):1391–
predictive control
413 discussing the recent
9. Kravaris C, Kantor JC. 1990. Geometric methods for nonlinear process control: 1. Background. Ind. developments.
Eng. Chem. Res. 29:2295–310

www.annualreviews.org • Control of Chemical Processes 399

CH05CH18-Lee ARI 20 May 2014 19:21

10. Lee JH, Lee KS. 2007. Iterative learning control applied to batch processes: an overview. Control Eng.
Pract. 15:1306–18
11. Wang Y, Gao F, Doyle FJ III. 2009. Survey on iterative learning control, repetitive control, and run-to-
run control. J. Process Control 19:1589–600
12. Dochain D. 2003. State and parameter estimation in chemical and biochemical processes: a tutorial.
J. Process Control 13:801–18
13. Soroush M. 1998. State and parameter estimations and their applications in process control. Comput.
Chem. Eng. 23:229–45
14. Ricardez-Sandoval LA, Budman HM, Douglas PL. 2009. Integration of design and control for chemical
processes: a review of the literature and some recent results. Annu. Rev. Control 33:158–71
15. Braatz RD. 2002. Advanced control of crystallization processes. Comput. Chem. Eng. 26:87–99
16. Alford JS. 2006. Bioprocess control: advances and challenges. Comput. Chem. Eng. 30:1464–75
17. Badgwell TA, Breedijk T, Bushman SG, Butler SW, Chatterjee S, et al. 1995. Modeling and control of
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

microelectronics materials processing. Comput. Chem. Eng. 19:1–41

Access provided by University of Saskatchewan on 03/05/17. For personal use only.

18. Richalet J. 1993. Industrial applications of model based predictive control. Automatica 29:1251–74
19. Wolfbeis OS. 1986. Analytical chemistry with optical sensors. Fresenius Z. Anal. Chem. 325(4):387–92
20. Chen Z, Lovett D, Morris J. 2011. Process analytical technologies and real time process control a review
of some spectroscopic issues and challenges. J. Process Control 21:1467–82
21. Vojinović V, Cabral JMS, Fonseca LP. 2006. Real-time bioprocess monitoring: part I: in situ sensors.
Sens. Actuators B Chem. 114(2):1083–91
22. Provides an 22. McDonagh C, Burke CS, MacCraith BD. 2008. Optical chemical sensors. Chem. Rev. 108(2):400–
overview of 22
developments in the
23. Nordon A, Littlejohn D, Dann AS, Jeffkins PA, Richardson MD, Stimpson SL. 2008. In situ monitoring
field of optical chemical
of a seed stage of a fermentation process using non-invasive NIR spectrometry. Analyst 133:660–66
sensors from 2000 to
24. Chen ZP, Morris J. 2008. Improving the linearity of spectroscopic data subjected to fluctuations in
external variables by the extended loading space standardization. Analyst 133:914–22
25. Nadadoor VR, De la Hoz Siegler H, Shah SL, McCaffrey WC, Ben-Zvi A. 2012. Online sensor for
monitoring a microalgal bioreactor system using support vector regression. Chemom. Intell. Lab. Syst.
26. Provides an 26. Nagy ZK, Braatz RD. 2012. Advances and new directions in crystallization control. Annu. Rev.
overview of
Chem. Biomol. Eng. 3:55–75
crystallization process
27. Grön H, Borissova A, Roberts KJ. 2003. In-process ATR-FTIR spectroscopy for closed-loop supersatu-
control up to the year
ration control of a batch crystallizer producing monosodium glutamate crystals of defined size. Ind. Eng.
Chem. Res. 42(1):198–206
28. Schöll J, Bonalumi D, Vicum L, Mazzotti M, Müller M. 2006. In situ monitoring and modeling of the
solvent-mediated polymorphic transformation of L-glutamic acid. Cryst. Growth Des. 6(4):881–91
29. Ruf A, Worlitschek J, Mazzotti M. 2000. Modeling and experimental analysis of PSD measurements
through FBRM. Part. Part. Syst. Charact. 17(4):167–79
30. Bogue R. 2012. Sensors for extreme environments. Sens. Rev. 32(4):267–72
31. Ghoshal A, Le D, Kim H. 2012. Technological assessment of high temperature sensing systems under
extreme environment. Sens. Rev. 32(1):66–71
32. Smith J. 1995. Flowmeter review. Sens. Rev. 15(4):11–14
33. Prosser SJ, Schmidt EDD. 1999. Smart sensors for industrial applications. Microelectron. Int. 16(2):20–23
34. Yeager B. 2008. Distributed control technology: from a decade of progress to an era of new possibilities. Presented
at Power Ind. Div. (POWID) Conf., Scottsdale, AZ
35. Hodge P. 2012. Virtualizing your process control computers. Chem. Eng. Prog. 108(1):22–26
36. Conover J. 2013. Virtualization on the plant floor. Control Eng. 2:34–36
37. Kameswaran S, Biegler LT. 2006. Simultaneous dynamic optimization strategies: recent advances and
challenges. Comput. Chem. Eng. 30(10–12):1560–75
38. Diehl M, Bock HG, Schlöder JP, Findeisen R, Nagy Z, Allgöwer F. 2002. Real-time optimization and
nonlinear model predictive control of processes governed by differential-algebraic equations. J. Process
Control 12(4):577–85

400 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

39. Leineweber DB, Bauer I, Bock HG, Schlöder JP. 2003. An efficient multiple shooting based reduced
SQP strategy for large-scale dynamic process optimization. Part I: theoretical aspects. Comput. Chem.
Eng. 27(2):157–66
40. Biegler LT, Zavala VM. 2009. Large-scale nonlinear programming using IPOPT: an integrating 40. Demonstrated an
framework for enterprise-wide dynamic optimization. Comput. Chem. Eng. 33(3):575–82 on-line dynamic
41. Cutler CR, Ramaker BL. 1980. Dynamic matrix control—a computer control algorithm. Proc. optimization strategy
for a large-scale
Joint Autom. Control Conf., San Francisco. New York: Am. Inst. Chem. Eng.
nonlinear model
42. Qin SJ, Badgwell TA. 2003. A survey of industrial model predictive control technology. Control
predictive control with
Eng. Pract. 11:733–64 moving horizon
43. Rawlings JB, Muske KR. 1993. The stability of constrained receding horizon control. IEEE Trans. Autom. estimation.
Control 38:1512–16
44. Martinsen F, Biegler LT, Foss BA. 2004. A new optimization algorithm with application to nonlinear
MPC. J. Process Control 14:853–65 41. The original
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

45. Diehl M, Ferreau HJ, Haverbeke N. 2009. Efficient numerical methods for nonlinear MPC and mov- publication that sparked
the interest in model
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

ing horizon estimation. In Nonlinear Model Predictive Control: Towards New Challenging Applications,
predictive control.
Volume 384, ed. L Magni, DM Raimondo, F Allgöwer, pp. 391–417. Berlin: Springer
46. Biegler LT. 2007. An overview of simultaneous strategies for dynamic optimization. Chem. Eng. Proc.
Process Intensif. 46:1045–53 42. Offers a survey on
47. Zavala VM, Biegler LT. 2009. The advanced-step NMPC controller: optimality, stability and robustness. industrial use of model
Automatica 45:86–93 predictive control
48. Bemporad A, Morari M. 1999. Control of systems integrating logic, dynamics, and constraints. Automatica conducted in the year
49. Lee JH, Yu Z. 1997. Worst-case formulation of model predictive control for systems with bounded
parameters. Automatica 33(5):763–81
50. Kothare MV, Balakrishnan V, Morari M. 1996. Robust constrained model predictive control using linear
matrix inequalities. Automatica 32(10):1361–79
51. Genceli H, Nikolau M. 1993. Robust stability analysis of constrained l1 norm model predictive control.
AIChE J. 39(12):1954–65
52. Lee JH, Cooley BL. 2000. Min-max predictive control technique for a linear state-space system with a
bounded set of input matrices. Automatica 36:463–73
53. Lee JH, Lee JM. 2006. Approximate dynamic programming based approach to process control and
scheduling. Comput. Chem. Eng. 30(10–12):1603–18
56. Represents a
54. Pannocchia G, Rawlings JB, Wright SJ. 2006. Fast, large-scale model predictive control by partial
definitive paper on
enumeration. Automatica 43:852–60 explicit model
55. Wang Y, Boyd S. 2010. Fast model predictive control using on-line optimization. IEEE Trans. Control predictive control for
Syst. Technol. 18:267–78 linear systems.
56. Bemporad A, Morari M, Dua V, Pistikopoulos EN. 2002. The explicit linear quadratic regulator
for constrained systems. Automatica 38(1):3–20
57. Bellman RE. 1957. Dynamic Programming. Princeton, NJ: Princeton Univ. Press 58. Provides a recent
58. Powell WB. 2011. Approximate Dynamic Programming: Solving the Curses of Dimensionality, 2nd comprehensive
Edition. Hoboken, NJ: Wiley-Intersci. treatment of
approximate dynamic
59. Lee JM, Lee JH. 2004. Approximate dynamic programming strategies and their applicability for process
programming written
control: a review and future directions. Int. J. Control Autom. Syst. 2(3):263–78
from the operations
60. Lee JH, Wong WC. 2010. Approximate dynamic programming approach for process control. research perspective.
J. Process Control 20:1038–48
61. Lee KS, Lee JH. 2003. Iterative learning control-based batch process control technique for integrated
control of end product properties and transient profiles of process variables. J. Process Control 13:607–21 60. Offers the latest
62. Yang DR, Lee KS, Ahn HJ, Lee JH. 2003. Experimental application of a quadratic optimal iterative review on how the
approximate dynamic
learning control method for control of wafer temperature uniformity in rapid thermal processing. IEEE
programming may be
Trans. Semicond. Manuf. 16:36–44
used in process control.
63. Lee KS, Chin IS, Lee HJ, Lee JH. 1999. Model predictive control technique combined with iterative
learning for batch processes. AIChE J. 45:2175–87

www.annualreviews.org • Control of Chemical Processes 401

CH05CH18-Lee ARI 20 May 2014 19:21

64. Julier SJ, Uhlmannn JK. 2004. Unscented filtering and nonlinear estimation. Proc. IEEE 92:401–22
65. Rao CV, Rawlings JB, Lee JH. 2001. Constrained linear state estimation—a moving horizon approach.
Automatica 37:1619–28
66. Robertson DG, Lee JH. 2002. On the use of constraints in least squares estimation and control. Automatica
67. Represents one of 67. Robertson D, Lee JH, Rawlings J. 1994. A moving horizon based approach for least squares state
the earliest papers
estimation. AIChE J. 42:2209–24
pointing out the
68. Haseltine EL, Rawlings JB. 2005. Critical evaluation of extended Kalman filtering and moving-horizon
potential of moving
horizon estimation for estimation. Ind. Eng. Chem. Res. 44:2451–60
nonlinear state 69. Rao CV, Rawlings JB, Mayne DQ. 2002. Constrained state estimation for nonlinear discrete-time sys-
estimation. tems: stability and moving horizon approximations. IEEE Trans. Autom. Control 48:246–58
70. Kandepu R, Foss B, Imsland L. 2008. Applying the unscented Kalman filter for nonlinear state estimation.
J. Process Control 18:753–68
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

71. Romanenko A, Castro JAAM. 2004. The unscented filter as an alternative to the EKF for nonlinear state
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

estimation: a simulation case study. Comput. Chem. Eng. 28:347–55

72. Romanenko A, Santos LO, Afonso PAFNA. 2004. Unscented Kalman filtering of a simulated pH system.
Ind. Eng. Chem. Res. 43:7531–38
73. Arulampalam MS, Maskell S, Gordon N, Clapp T. 2002. A tutorial on particle filters for online
73. Offers a classic nonlinear/non-Gaussian Bayesian tracking. IEEE Trans. Signal Process. 50:174–88
tutorial on particle
74. Lang L, Chen W, Bakshi BR, Goel PK, Ungarala S. 2007. Bayesian estimation via sequential Monte
filtering, with over
Carlo sampling—constrained dynamic systems. Automatica 43:1615–22
6,000 citations.
75. Young RE. 2006. Petroleum refining process control and real-time optimization. IEEE Control Syst. Mag.
76. Ogden-Swift A. 1996. An industrial review of advanced process control of continuous processes on
refineries and petrochemical plants. Trans. Inst. Meas. Control 18(1):9–14
77. Kozak S. 2012. Advanced control engineering methods in modern technological applications. Proc. 13th
Int. Carpathian Control Conf., High Tratas, Slovakia, pp. 392–97. Washington, DC: IEEE
78. Bodington CE. 1995. Planning, Scheduling, and Control Integration in the Process Industries. New York:
79. Lautenschlager Moro LF. 2003. Process technology in the petroleum refining industry—current situa-
tion and future trends. Comput. Chem. Eng. 27(8–9):1303–5
80. Congalidis JP, Richards JR. 2005. Measurement and control of polymerization reactors. Comput. Chem.
Eng. 30:1447–63
81. Seavey KC, Liu YA, Khare NP, Bremner T, Chen CC. 2003. Quantifying relationships among the
molecular weight distribution, non-Newtonian shear viscosity, and melt index for linear polymers.
Ind. Eng. Chem. Res. 42(21):5354–62
82. Sharmin R, Sundararaj U, Shah S, Griend LV, Sun YJ. 2006. Inferential sensors for estimation of polymer
quality parameters: industrial application of a PLS-based soft sensor for a LDPE plant. Chem. Eng. Sci.
83. Facco P, Doplicher F, Bezzo F, Barolo M. 2009. Moving average PLS soft sensor for online product
quality estimation in an industrial batch polymerization process. J. Process Control 19(3):520–29
84. Gonzaga JCB, Meleiro LAC, Kiang C, Filho RM. 2009. ANN-based soft-sensor for real-time process
monitoring and control of an industrial polymerization process. Comput. Chem. Eng. 33(1):43–49
85. Prasad V, Schley M, Russo LP, Bequette BW. 2002. Product property and production rate control of
styrene polymerization. J. Process Control 12(3):353–72
86. Zavala VM, Biegler LT. 2009. Optimization-based strategies for the operation of low-density polyethy-
lene tubular reactors: nonlinear model predictive control. Comput. Chem. Eng. 33(10):1735–46
87. Ohshima M, Tanigaki M. 2000. Quality control of polymer production processes. J. Process Control
88. Chatzidoukas C, Perkins JD, Pistikopoulos EN, Kiparissides C. 2003. Optimal grade transition
and selection of closed-loop controllers in a gas-phase olefin polymerization fluidized bed reactor.
Chem. Eng. Sci. 58(16):3643–58

402 Lee · Lee

CH05CH18-Lee ARI 20 May 2014 19:21

89. Braunecker WA, Matyjaszewski K. 2007. Controlled/living radical polymerization: features, develop-
ments, and perspectives. Prog. Polym. Sci. 32(1):93–146
90. Edgar TF, Butler SW, Campbell WJ, Pfeiffer C, Bode C, et al. 2000. Automatic control in microelec-
tronics manufacturing: practices, challenges, and possibilities. Automatica 36(11):1567–603
91. Lee HJ. 2008. Advanced process control and optimal sampling in semiconductor manufacturing. PhD Thesis,
Univ. Tex., Austin
92. Sachs E, Hu A, Ingolfsson A. 1995. Run by run process control: combining SPC and feedback control.
IEEE Trans. Semicond. Manuf. 8(1):26–43
93. Box GEP, Jenkins GM. 1963. Further contributions to adaptive quality control: simultaneous estimation
of dynamics: non-zero costs. Bull. Int. Stat. Inst. 34:943–74
94. Butler SW, Stefani JA. 1994. Supervisory run-to-run control of polysilicon gate etch using in situ ellip-
sometry. IEEE Trans. Semicond. Manuf. 7(2):193–201
95. Qin SJ, Cherry G, Good R, Wang J, Harrison CA. 2006. Semiconductor manufacturing process control
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

and monitoring: a fab-wide framework. J. Process Control 16(3):179–91

Access provided by University of Saskatchewan on 03/05/17. For personal use only.

96. Chen A, Guo RS. 2001. Age-based double EWMA controller and its application to CMP processes.
IEEE Trans. Semicond. Manuf. 14(1):11–19
97. Lee JH, Kiew CM. 2009. Robust forecasts and run-to-run control for processes with linear drifts.
J. Process Control 19:636–43
98. Zhang JH, Chu CC, Munoz J, Chen JH. 2009. Minimum entropy based run-to-run control for semi-
conductor processes with uncertain metrology delay. J. Process Control 19(10):1688–97
99. Kang P, Kim D, Lee HJ, Doh S, Cho S. 2011. Virtual metrology for run-to-run control in semiconductor
manufacturing. Expert Syst. Appl. 38(3):2508–22
100. Preuß K, Le Lann MV, Cabassud M, Anne-Archard G. 2003. Implementation procedure of an advanced
supervisory and control strategy in the pharmaceutical industry. Control Eng. Pract. 11(12):1449–58
101. Oh S-K, Yoo SJ, Jeong DH, Lee JM. 2013. Real-time estimation of glucose concentration in algae
cultivation system using Raman spectroscopy. Bioresour. Technol. 142:131–37
102. Schügerl K. 2001. Progress in monitoring, modeling and control of bioprocesses during the last 20 years.
J. Biotechnol. 85(2):149–73
103. Henry O, Kamen A, Perrier M. 2007. Monitoring the physiological state of mammalian cell perfusion
processes by on-line estimation of intracellular fluxes. J. Process Control 17(3):241–51
104. Camacho EF, Samad T, Garcı́a-Sanz M, Hiskens I. 2011. Control for renewable energy and smart grids.
In Impact of Control Technology, ed. T Samad, A Annaswamy, pp. 69–88. New York: IEEE Control Syst.
105. Philp JC, Guy K, Ritchie RJ. 2012. Biofuels development and the policy regime. Trends Biotechnol. 31:4–6
106. Bernard O. 2011. Hurdles and challenges for modelling and control of microalgae for CO2 mitigation
and biofuel production. J. Process Control 21(10):1378–89
107. Hoffner K, Harwood SM, Barton PI. 2013. A reliable simulator for dynamic flux balance analysis.
Biotechnol. Bioeng. 110(3):792–802
108. Johnson KE, Pao LY, Balas MJ, Fingersh LJ. 2006. Control of variable-speed wind turbines: standard
and adaptive techniques for maximizing energy capture. IEEE Control Syst. Mag. 26(3):70–81
109. Garcı́a-Sanz M, Houpis CH. 2012. Wind Energy Systems: Control Engineering Design. Boca Raton, FL:
CRC Press
110. Camacho EF, Rubio FR, Berenguel M, Valenzuela L. 2007. A survey on control schemes for distributed
solar collector fields. Part II: advanced control approaches. Solar Energy 81(10):1252–72
111. Powell WB, George A, Simão H, Scott W, Lamont A, Stewart J. 2012. SMART: a stochastic multiscale
114. Provides insights
model for the analysis of energy resources, technology, and policy. INFORMS J. Comput. 24(4):665–82
into the research
112. Lee JH. 2014. Energy supply planning and supply chain optimization under uncertainty. J. Process Control
opportunities for
24(2):323–31 nanoscale processing
113. Klatt K-U, Marquardt W. 2009. Perspectives for process systems engineering—personal views from systems from the
academia and industry. Comput. Chem. Eng. 33(3):536–50 systems engineering
114. Stephanopoulos N, Solis EOP, Stephanopoulos G. 2005. Nanoscale process systems engineering: perspective.
toward molecular factories, synthetic cells, and adaptive devices. AIChE J. 51(7):1858–69

www.annualreviews.org • Control of Chemical Processes 403

CH05CH18-Lee ARI 20 May 2014 19:21

115. Solis EOP, Barton PI, Stephanopoulos G. 2010. Controlled formation of nanostructures with desired
geometries. 1. Robust static structures. Ind. Eng. Chem. Res. 49(17):7728–45
116. Charpentier JC. 2005. Four main objectives for the future of chemical and process engineering mainly
concerned by the science and technologies of new materials production. Chem. Eng. J. 107(1–3):3–17
117. Ulissi ZW, Strano M, Braatz RD. 2013. Control of nano and microchemical systems. Comput. Chem.
Eng. 51:149–56
119. Demonstrates the 118. Papageorgiou LG. 2009. Supply chain optimisation for the process industries: advances and opportuni-
application of stochastic ties. Comput. Chem. Eng. 33(12):1931–38
programming to the 119. Gupta A, Maranas CD. 2003. Managing demand uncertainty in supply chain planning. Comput.
planning problem under Chem. Eng. 27(8–9):1219–27
uncertainty. 120. Kraiselburd S, Yadav P. 2013. Supply chains and global health: an imperative for bringing operations
management scholarship into action. Prod. Oper. Manag. 22(2):377–81
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

404 Lee · Lee

CH05-FrontMatter ARI 5 May 2014 12:43

Annual Review of
Chemical and

Volume 5, 2014 Contents

Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

Plans and Detours

Access provided by University of Saskatchewan on 03/05/17. For personal use only.

James Wei p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 1
Simulating the Flow of Entangled Polymers
Yuichi Masubuchi p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p11
Modeling Chemoresponsive Polymer Gels
Olga Kuksenok, Debabrata Deb, Pratyush Dayal, and Anna C. Balazs p p p p p p p p p p p p p p p p p p p35
Atmospheric Emissions and Air Quality Impacts from Natural Gas
Production and Use
David T. Allen p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p55
Manipulating Crystallization with Molecular Additives
Alexander G. Shtukenberg, Stephanie S. Lee, Bart Kahr,
and Michael D. Ward p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p77
Advances in Mixed-Integer Programming Methods for Chemical
Production Scheduling
Sara Velez and Christos T. Maravelias p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p97
Population Balance Modeling: Current Status and Future Prospects
Doraiswami Ramkrishna and Meenesh R. Singh p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 123
Energy Supply Chain Optimization of Hybrid Feedstock Processes:
A Review
Josephine A. Elia and Christodoulos A. Floudas p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 147
Dynamics of Colloidal Glasses and Gels
Yogesh M. Joshi p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 181
Rheology of Non-Brownian Suspensions
Morton M. Denn and Jeffrey F. Morris p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 203
Factors Affecting the Rheology and Processability of Highly
Filled Suspensions
Dilhan M. Kalyon and Seda Aktaş p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 229

CH05-FrontMatter ARI 5 May 2014 12:43

Continuous-Flow Differential Mobility Analysis of Nanoparticles

and Biomolecules
Richard C. Flagan p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 255
From Stealthy Polymersomes and Filomicelles to “Self ”
Peptide-Nanoparticles for Cancer Therapy
Núria Sancho Oltra, Praful Nair, and Dennis E. Discher p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 281
Carbon Capture Simulation Initiative: A Case Study in Multiscale
Modeling and New Challenges
David C. Miller, Madhava Syamlal, David S. Mebane, Curt Storlie,
Debangsu Bhattacharyya, Nikolaos V. Sahinidis, Deb Agarwal, Charles Tong,
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org

Stephen E. Zitney, Avik Sarkar, Xin Sun, Sankaran Sundaresan, Emily Ryan,
Dave Engel, and Crystal Dale p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 301
Access provided by University of Saskatchewan on 03/05/17. For personal use only.

Downhole Fluid Analysis and Asphaltene Science for Petroleum

Reservoir Evaluation
Oliver C. Mullins, Andrew E. Pomerantz, Julian Y. Zuo, and Chengli Dong p p p p p p p p p p 325
Biocatalysts for Natural Product Biosynthesis
Nidhi Tibrewal and Yi Tang p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 347
Entangled Polymer Dynamics in Equilibrium and Flow Modeled
Through Slip Links
Jay D. Schieber and Marat Andreev p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 367
Progress and Challenges in Control of Chemical Processes
Jay H. Lee and Jong Min Lee p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 383
Force-Field Parameters from the SAFT-γ Equation of State for Use in
Coarse-Grained Molecular Simulations
Erich A. Müller and George Jackson p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 405
Electrochemical Energy Engineering: A New Frontier of Chemical
Engineering Innovation
Shuang Gu, Bingjun Xu, and Yushan Yan p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 429
A New Toolbox for Assessing Single Cells
Konstantinos Tsioris, Alexis J. Torres, Thomas B. Douce,
and J. Christopher Love p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 455
Advancing Adsorption and Membrane Separation Processes for the
Gigaton Carbon Capture Challenge
Jennifer Wilcox, Reza Haghpanah, Erik C. Rupp, Jiajun He, and Kyoungjin Lee p p p p p 479
Toward the Directed Self-Assembly of Engineered Tissues
Victor D. Varner and Celeste M. Nelson p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 507
Ionic Liquids in Pharmaceutical Applications
I.M. Marrucho, L.C. Branco, and L.P.N. Rebelo p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 527

Contents vii
CH05-FrontMatter ARI 5 May 2014 12:43

Perspectives on Sustainable Waste Management

Marco J. Castaldi p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 547
Experimental and Theoretical Methods in Kinetic Studies of
Heterogeneously Catalyzed Reactions
Marie-Françoise Reyniers and Guy B. Marin p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 563


Cumulative Index of Contributing Authors, Volumes 1–5 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 595

Cumulative Index of Article Titles, Volumes 1–5 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 598
Annu. Rev. Chem. Biomol. Eng. 2014.5:383-404. Downloaded from www.annualreviews.org
Access provided by University of Saskatchewan on 03/05/17. For personal use only.


An online log of corrections to Annual Review of Chemical and Biomolecular Engineering

articles may be found at http://www.annualreviews.org/errata/chembioeng

viii Contents