Instrumentacion y control de procesos

Full text

Present and Future Trends in Process Control
C. de Prada, G. Gutierrez
Dpt. of Systems Engineering and Automatic Control, University of Valladolid, Spain
prada@autom.uva.es, gloria@autom.uva.es
Abstract.- This paper presents a personal view about the current state of process
control and the challenges it faces, and describes some research paths that aim to help
solving them, giving the field a central role in the operation of modern process
factories.
1. Introduction
Process control has evolved to become a mature field that plays an important role in the
daily operation of process factories. From the early days of pneumatic instruments and
Ziegler-Nichols tuning rules, it has developed besides technologies. Today, most plants
are operated from control rooms equipped with distributed control systems, PLCs and
SCADAs for supervisory and regulatory tasks, safety or alarm management. Field
instrumentation has also followed this path, and quite frequently we can see transmitters
equipped with the HART protocol and more and more smart instrumentation linked
with fieldbuses or wireless communications. Control methods have evolved also from
the single PID to complex control structures, and more recently to linear MPC
multivariable controllers linked to economic optimizers of process units.
In the meantime, the context in which process industries operate has changed a lot. Due
to the globalization of the economy, transport facilities, availability of information
about products, prices, etc. we live in a world of global competition. The raw materials,
energy, etc. are no longer an infinite and cheap source and the impact of our industrial
activities about the environment is in the center of our worries about a sustainable
future. In addition, accidents and inadequate locations have raised concerns about safety
and impact on populations. The result is that industries are now subjected to increased
competition and pressures to decrease costs, improve quality, be more energy efficient
and respect environmental and safety norms.
The response to these problems has come from different perspectives including markets,
prices, processes and products. In particular, European process industry has small
chances to compete on wages or avoiding implementing environmental or safety norms,
so that it has to concentrate on other aspects. Besides new and intensified processes and
innovative products, the field of process operations and process control can help in a
significant way improving the way factories are run, fostering the position of process
plants in the market. Moreover, it is difficult to think in operating a plant optimally with
respect to energy and costs savings, quality assurance, flexibility of adaptation to
market changes, bottlenecks avoidance, safety assurance, etc. without the concourse and
ample use of better control and operation methods.

This paper reviews the trends and developments that appear in the horizon, regarding
both, technologies and research activities linked to process control and optimization. It
is written with the purpose of helping reflecting about its role and the focus they need in
order to respond to the previously mentioned challenges the process industry faces.
2. Technology trends
In this perspective, what is offered today to the process industry by the other actors
involved in process control? If you attend a presentation of one of the leading suppliers
of control and instrumentation companies about the future trends, likely, what you will
hear there is related to better and more sophisticated instrumentation including wireless,
fancier human interfaces and monitoring systems, better and integrated communication
system at all levels in the enterprise, remote support for data analysis, malfunctions,
etc., cyber security measures, implementation of standards and norms for data exchange
or knowledge representation,… that is, a lot of software and communications
technology and tools generating more data and best ways to organize them.
Similarly, engineering companies focus their activities more and more in the
development of software tools for the management of this information in the upper
activity levels, in particular in the so called Manufacturing Execution Systems (MES) or
Manufacturing Operations Management (MOM). Figure.1 displays the “control
pyramid”, a classical view of the different activities in a process plants from the point of
view of process control and operations, organized in layers, left and right hand sides
reflecting the academic and industrial views of these levels. The MES - MOM is
composed of information systems that support the things a production department must
do in order to prepare and manage work instructions, schedule production activities,
monitor correct execution of the production process, gather and analyze information
about the production process and products and feed this back to other departments such
as accounting and logistic, take care of maintenance and key performance indicators
(asset management), the laboratory information systems, etc. In summary, information
systems that allow feed elaborated results back to the personnel and take earlier
corrective actions.
Figure 1. Academic and industrial views of the “control pyramid”

.
All those applications that are now been developed and will be increasingly available in
the near future, will provide useful tools for a better management of the plants.
Nevertheless, being important, they lack a significant presence of methods and
techniques oriented to a flexible and optimal operation of the process plants that are
often mentioned in the literature as key elements of the future process control systems:
dynamic simulation, model predictive control, real-time optimization, fault detection,
on-line scheduling, software sensors….being more oriented to provide support for
human decisions, than automated real-time optimal ones.
.
3. Control and optimization of process plants
Facing the challenges posed by the current and future economic context requires
incorporating new visions of the role of process control. For many years the focus has
been placed on the control loop and, certainly, we are able today of maintaining in a
sensible way the values of controlled variables around specified values in spite of
disturbances. But operating a plant optimally according to an economic criterion,
adapting production to the changes in market demands and respecting constraints
imposed by quality, safety, etc. requires moving the focus from the individual control
loops to the whole factory.
3.1 Modeling and Simulation
In order to make the right decisions about the operation of a plant, it is necessary to take
into account many variables and their mutual influences. The quantitative relations
among these variables are given by a dynamic model, which is one of the best ways of
storing knowledge about the plant. Hence, mathematical modeling is in the basis of any
sound approach to optimal process operation. Developing models and in particular
dynamic ones, is considered to be a difficult and time consuming task with not perfect
results, what stop many from using models in the decision making process. This is true,
nevertheless, today there are many tools and methodologies that facilitate a lot the task
of model development. In particular, modern simulation languages offer graphical
libraries in different fields, with many pre-defined models of typical process units, like
the one in Figure 2.
Figure 2. Typical graphical interface of a simulation language (EcosimPro)

Building a model of a plant is mainly a task of connecting together graphically the
process units as they are connected in the plant, and providing values for the different
model parameters, such as sizes, compositions, heat transfer coefficients, etc., the whole
model being generated automatically. For those parameters which value is unknown,
these modeling environments offer parameter estimation tools based on optimization,
that compute the values of the parameters that best fit a set of experimental data.
Alternatively, simplified models can be obtained using well established identification
methods. They require performing experiments on the process and then the parameters
of a (linear) model are computed by optimization in order to fit the model to the data.
Models can be used for many different purposes. We have introduced them as the basis
for decision making, but there are many other uses that justify their development, such
as training operators, testing new control strategies, testing strategies or equipment
before commissioning, etc. According to the intended use, different types of models
with different complexity can be used.
3.2 Model based Control and Optimization
Models are one the pillars of a modern approach to decision making and they can be
used explicitly to control a process unit as a whole. The approach is known as Model
Predictive Control (MPC) and uses the model to predict the future behavior of the
process as a function of the future control actions. Then, these control actions are
computed to minimize the differences of the predictions and the desired values of the
controlled variables taking into account the interactions among the model variables,
delays, etc. and the process constraints on the variables given by the physical ranges,
quality margins, safety, etc. The fact that all variables and constraints are considered
simultaneously and an optimization procedure is applied to compute the control actions,
leads to a clear improvement of the quality of control. As a result, the margin for
choosing a better operating point is widen as can be seen in Figure 3, where the
decrement in the variability of a signal due to better control allows to move the SP to a
more favorable value respecting the constraint. In this way, improving control provides
more opportunities for economic optimization. In fact, very often, MPC is linked to a
local optimizer that chooses the SP of the MPC in order to maximize or minimize a
certain economic cost function.
Figure 3. Reducing the variance due to better control allows to move the SP to a different and better value
respecting constraints.
MPC is used in many industries providing consistent benefits, most of the times with a
linear internal model and attached to a local optimizer, as in the left hand side of Figure

4. Here RTO stands for Real Time Optimization and uses a static process model to
compute the best operating point according to an economic target. The results of the
optimization are sent as set points to the MPC layer. When used with a linear MPC,
RTO is a Linear Programming problem (LP) with the same model as the MPC limited
to a process unit (distillation column, FCC, etc.). But other times is a non-linear model
based on first principles covering a wide section of the plant.
Figure 4. Traditional RTO + MPC layers (left) and Optimizing Control merging both (right)
Both layers, MPC and RTO can be merged in a single one, called Optimizing Control, if
the MPC modifies its target and instead to follow a set point it tries directly to optimize
the economic index of the RTO, taking into account dynamics and constraints. In this
case, the models are normally non-linear and based on physical principles, which are
more difficult to solve, but widen the range of operation and integrates dynamics and
economics. Quite often the process (and the corresponding models) involves not only
continuous, but also discrete variables or some logic of operation that carries
discontinuities in the operation. In these cases, the associated dynamic optimization
problem to be solved on-line is a mix-integer one, increasing the difficulty of the
problem, Research on these topics continues and, in spite of the challenges it represents,
there are very promising results in processes of different characteristics and targets:
optimal start-ups, combined operation of continuous and batch processes, energy
integrated columns, etc.
3.3 Large scale systems
Optimization is the second pillar of the way factories of the future will be operated. The
success of the optimization approaches above mentioned is closely linked to the ability
to solve large scale non-linear problems and, at the same time, these problems become
bigger and bigger as they cover more complex and large processes and plants.
Fortunately, research in optimization algorithms and software has produced very
efficient methods so that, today, is possible to solve in a short time problems with
thousands of variables that were unthinkable some time ago. Also, methods for reducing
the size of the models, substituting them by simplified ones that can be considered god
approximations for the intended used, are under development, contributing to speed up
the calculations.

At the same time, in order to be able to deal with large scale problems, several
approaches have been developed with the common factor of distributing or parallelizing
the computations among several computers. In addition to parallelizing the
computations of a big problem formulated in a centralized way, other approaches try to
formulate the problem as a set of local and smaller controllers/optimizers each in charge
of a sub-process with some type of coordination among them, so that the solution of the
individual optimizers coincides with the one of the centralized and large scale problem.
Some of these approaches can be seen in Figure 5. The one named as Hierarchical
consists of two layers: the lower one is formed by the local optimizers that receive
targets from the upper one in charge to coordinate them. In the distributed case, there is
a single layer of controllers/optimizers, and these interchange information about its
future actions, so that the others can take them into account when computing their own
decisions. Finally, in the Price Coordination approach the cost functions of the local
optimizers are changed by the upper layer (the market) modifying prices so that the
local decisions converge to the optimal global one.
Figure 5. Hierarchical, Distributed and Price coordination strategies for solving large scale optimization
problems in a decentralized way.
3.4 Uncertainty
In addition to the concerns about the size of the optimization problems, the other aspect
that poses doubts about the real chances to operate optimally a process is the lack of
knowledge that we have many times about it, due to the presence of unmeasured
variables or unreliable measurements, as well as models that are never perfect.
Methods for estimating unmeasured variables from other measurements are well known
in the literature. The Kalman fiiler, Extended or unscent Kalman filter (EKF, UKF),
non-linear observers, or more recently the Moving Horizon Estimators (MHE) all of
them use a model to estimate the unknown variables. In the same way, software sensors

use correlations and physical relations to perform similar inferences. All of them,
together with new instrumentation, provide a set of tools that can help in improving the
information obtained from the process. Moreover, data reconciliation techniques that
solve an optimization problem looking for the values of variables of a model that best fit
a set of measurements, provide a coherent and complete set of variables and can be used
for other purposes such as instruments fault detection. The use of these techniques in a
factory, largely can improve the knowledge about the state of the process.
Regarding the other problem, the necessary imperfectness of the models used to make
decisions, two main options to deal with it appears: Either measurements are used to
update the model from time to time, or the uncertainty is considered explicitly in the
optimization problem. The first one is linked to data reconciliation or adaptive
approaches. The second one presents also many different lines, from robust and
stochastic optimization, that takes decisions to satisfy constraints even in the worst case
of uncertainty, to others where measurements are used to correct the wrong decisions
made with the imperfect knowledge, such as modifier adaptation methods, or to
minimize the effect of the uncertainty, such as optimizing control schemes. Many are
still in the research phase, but increasingly, new tools are been developed that will
contribute to eliminate this problem.
4 Conclusions
This paper reviews some trends in the process control field, as well as challenges linked
to the need of re-focusing the targets from the control loop to the whole factory
operation. Modelling, optimization, large scale and uncertainty are mentioned as the key
elements to be considered in this evolution.
Many of the techniques mentioned are in operation in industry, such as MPC or RTO,
inferences, dynamic simulation, etc. but their used is still rather limited. Why? Partly,
because of lack of education and training, both in the Universities and industries. Partly
because develop and maintain the systems running properly require organizational
changes and resources, integrating management and production. Convergence of
systems (Suppliers), software (engineering companies) and methods and education
(academia) with process industries is a must to advance in the development and
implementation of modern optimization based systems able to cope with the problems
of global competition.
References
C. de Prada, El futuro del control de procesos, RIAI, CEA-IFAC, Vol.1,n.1, ISSN:
1697-7912, pg. 5-20, April, 2004
S. Engell, S. Lohmann, T. Moor, C. de Prada, J. Raisch, D. Sarabia, S. Sonntag,
Industrial Controls, HYCON Handbook, Cambridge University Press, ISBN 978-0-521-
76505-3 pp.405-437, 2009
Proceedings of Foundations of Computer-Aided Process Operations FOCAPO 2012 and
Chemical Process Control VIII, CPC 2012, CACHE, Savannah, Georgia, USA
February/2012