Industrial Plant Optimization and Advanced Control Application

Nenad Bolf, Ivan Mohler

University of Zagreb, Faculty of Chemical Engineering and Technology
Savska c. 16/5a, 10000 Zagreb, Croatia
bolf@fkit.hr, imohler@fkit.hr

Abstract
In today’s globally competitive marketplace, industrial plants are looking at new ways to
increase plant efficiency, production rates, safety and reliability. Engineer education and training,
monitoring, diagnosis and plant optimization play a key role in satisfying technological, economical
and environmental constraints. Furthermore, control system optimization is the basis for system
improvement and advanced process control (APC) implementation.
Very few plants use modern software for control quality monitoring, controller tuning, APC or
optimization. The reasons are absence of engineering knowledge and unavailability of practical and
robust process control software tools for system identification, parameter optimization and control
quality monitoring, running plants conservatively due to fear of causing shutdowns and plant
problems.
Process control software tools for quick and easy system identification using available data from
the plant’s historian can help tremendously improve the control quality and the plant’s profit
margin. It is possible to analyze multivariable systems, complex, nonlinear and slow processes with
long dead times and long time constants commonly encountered in process industry.
Optimization of primary and advanced control schemes stabilizes the process and allows the
plant to run closer to process, equipment and economic constraints. This increases production rates,
minimizes operating costs and improves product quality. The overall control system performance is
significantly improved which ultimately has a positive effect on product quality and energy
consumption thus proving application of control system diagnostics and optimization usefulness.
 closed-loop mode without conducting any
1 Introduction additional new step-tests in the plant. Current
Application of advanced control methods has system identification and optimization
rapidly increased since the 1990s in the algorithms such as: autoregressive–moving-
chemical, petrochemical, and oil refining average model with exogenous inputs
industry. The terms DCS, PLC and PID can be (ARMAX), step response coefficient models,
found in many articles like: Henriquez et al. [1], Box and Jenkins, etc., are rather sensitive to the
Van Schuppen et al. [2], Cauffriez et al. [3], presence of noise, disturbances and drifts in the
Campelo et al. [4], Rullán [5], Valencia-Palomo data.
et al. [6], Bolton [7], Reznik et al. [8], Panda [9] In the PID and APC parameter optimization
and Escobar et al. [10]. area, the application of internal model control
Also, the terms APC, model predictive (IMC), Lambda tuning, Ziegler-Nichols, Cohen
control (MPC) and MBC are mentioned many Coon, and other methods, still produces
times in literature as in: Lababidi et al. [11], conservative and not optimal control action. The
Dobos et al. [12], Zhi Gao et al. [13], Willersrud 3G algorithm can accurately calculate PID/APC
et al. [14], Al-Gherwi et al. [15], Peng et al. [16] parameters for the processes where currently
and Malchow et al. [17]. known heuristic-based methods have failed.

As can be seen, these terms are ubiquitous in

chemical process control literature. A 2 Process control application hierarchy
prerequisite for APC/MPC success is a well- Fig. 1 shows the common industrial process
designed primary PID control platform with control hierarchy split into three major
optimized parameters. Increasing application of categories. At the lowest level are the primary
AP schemes places higher demands on the skills control loops – mostly PID controllers for
and experience level of process control controlling flow-rates, pressures, levels,
engineers and technicians in the control rooms. temperatures and other variables in the
industrial plant. To handle slow process
This paper nicely explains the application of
dynamics, multivariable interactions, long dead
the powerful 3G optimization method [18–20]
times and complex control loops, pure PIDs
which helps the control engineer and technician
alone cannot effectively provide the best control
to design and implement control schemes inside
quality, and APC applications are necessary.
the DCS, optimize the controller performance
Further, to incorporate market, economic
and increase the plant’s profit because of
conditions, process and equipment constraints
improved plant operation.
and nonlinearities, a third application level –
Application of process control software tools real time optimization can further provide
for system identification, PID tuning monetary benefits.
optimization and APC calculations is still not
Primary control and DCS-based APC, if
too common in the control room environment in
correctly implemented, can significantly
manufacturing plants. The reasons for this is
increase the plant’s profit margins. Optimized
that most current software tools and dedicated
primary and advanced control stabilizes process
optimization algorithms are too complex, rather
operation and pushes the operation closer to
expensive, and neither robust nor practical for
process, equipment and economic constraints.
the control room environment.
This increases production rates, minimizes
Further, applications of the newly developed operating costs and improves product quality
3G algorithm are illustrated. It can accurately [21].
identify process models admist the presence of
large unmeasured disturbances or oscillations
and high noise from the data, all in complete
 Using modern DCS and/or PLCs, various
powerful, robust, money-making control
schemes can be implemented. This paper also
describes the following powerful techniques for
designing and implementing DCS or PLC-based
APC schemes, which are optimized using the
3G algorithm:
1. Process dynamics identification,
2. Primary PID and APC optimization,
3. Online adaptive control,
4. Model-based control for product quality
and production rate maximization.

4 Process dynamics identification

Process measurement ranges can be from as
fast as milliseconds on compressor surge control
and motor control to as low as many hours in
tall super-fractionator distillation columns. In
Figure 1: Process control hierarchy. modern control rooms, there are plenty of data
sets available containing the controller output
(OP), process variable (PV) and set point (SP).
3 Challenges and opportunities in modern Data may contain OP step changes with the
control rooms controller in manual mode, or may contain SP
The increasing use of primary and advanced changes in auto mode. There are many
control poses the following challenges in the opportunities in the plant where the operator
control rooms: may have made changes to the SP or OP. All
these data sets are abundantly available from the
1. New control and process engineers and
plants data historians that continuously archive
DCS or PLC technicians come into the plant on
the data.
a regular basis. They need to be trained in
practical primary and advanced process control. From collected data it is possible to identify
process dynamics i.e. the dynamic relationship
2. Many DCS-based APC concepts cannot be
between the controlled variable (CV) and
taught practically at schools and colleges.
manipulated variable (MV) for each control
Learning practical process control skills quickly
loop. Most chemical processes can be
is not easy and simple.
characterized by one of the common industrial
3. Changes in process or operating process models (zero, first or second order) [22,
conditions, complexity, nonlinearities, external 23].
unmeasured disturbances, high signal noises can
Pitops [24] identification and optimization
impact closed-loop control quality resulting in
software tool was used to identify the model
inefficient operation including lost production,
parameters using existing data of distillation
and could even cause equipment shutdowns and
column pressure controller from an olefins
safety/reliability incidents.
plant, which were stored in the plant’s data
4. Constant software and hardware upgrades historian. Fig. 2 shows pressure control (PC)
add to the maintenance challenges in the control data when the loop was in manual mode. The
room. PC’s output (OP), i.e. the valve position, was
moved a few times which caused the pressure Using temperature PV and steam flow-rate
PV to respond. Identified first order process SP data in a closed-loop control, the first order
model parameters are shown on the right top model was identified. The identified process
side of the Figure: Time delay = 1.3 min; parameters are: Delay time = 1 min; Process
Process gain = –1.017 bar ⁄ %; Time constant = gain = 1.234 °C ⁄ (kg h−1); Time constant =
5.4 min. 23.7 min.
The blue trend in the top window shows the The last example shows a multivariable
model prediction which follows the red trend of model-predictive controller from a chemical
the actual column pressure data in the top plant manufacturing catalyst in closed-loop
window and the bottom trend is the control mode, simultaneously manipulating three MVs:
valve position. distillation column feed, side product flow-rate
and reflux flow-rate, as shown in Fig. 4. The
product impurity is impacted by all three MVs
(red trend in the top window). All three second-
order models are identified simultaneously
using the data from the closed-loop mode. This
identification can be used to improve the step
response coefficient or any other kind of models
used in the commercially multivariable model-
predictive controllers in order to improve the
controller performance.
Figure 2: Pressure control loop model
identification [24].
The next example shows model parameter
identification using closed-loop data (controller
in auto mode), as shown in Fig. 3. This example
involves a temperature controller which is
manipulated by a steam flow controller in a
distillation batch process. The TC output is
often zero (there is no steam flow). When the
process is ready for increasing the temperature, Figure 4: Multivariable control loop model
a batch sequence logic tag changes the TC’s set identification – closed loop mode.
point. Optimal tuning of the TC was demanding
because the TC is not always in control during
the day shift. 5 PID Controller Tuning and Optimization
Knowing the process model helps to
optimally tune base-level and cascade/advanced
controllers. Fig. 5 shows an industrial pressure
control (PC) example, which is the same as in
Fig. 2. The bottom window shows the PC’s
output. The top window shows the SP (blue
trend) and PV (red trend).
The PC’s objective is to not only provide
crisp SP control but also to respond
Figure 3: Temperature control loop model aggressively when hit by a disturbance.
identification – closed loop mode. Disturbances can come and go anytime and it is
important for the PC to respond quickly by
closing or opening the valve immediately. The PID parameters. Fig. 6 shows an example of a
key is that such aggressive control action master AC and its slave TC from the simulator,
needed during disturbance rejection should not which mimics a distillation column cascade
result in sustained oscillations at steady state. example.
The 3G optimization algorithm generates
such tuning parameters that give crisp, non-
oscillatory SP control while responding quickly
during fast and large disturbances. This resulted
in increasing the controller proportional gain
from 2 to 11 in one step and the integral from 8
to 3 minutes.

Figure 6: Optimized PID controller in presence

of SP change.
One of the most powerful schemes to
maximize plant profits is implementing
production rate maximizer schemes in the DCS
or PLC, as illustrated in Fig. 7. As many as ten
or more constraints can be implemented as part
of the constraint pusher schemes. The optimizer
can identify the process dynamics for all the
constraints and then precisely optimize all the
tuning parameters.

Figure 5: Optimized pressure controller in

presence of disturbance.
Without modern optimization tools that use
the 3G optimization algorithm, control
engineers confronted with tuning such a PC
would not have the confidence to increase the
controller gain drastically from 2 to 11 in one
step. They would have crept up the gain from 2
to 2.5 and 3 etc., over a much longer time
period. And since the disturbance does not come
all the time, it is hard to manually tune the loop
Figure 7: Production rate maximizer
for optimal control without the help of modern
controllers.
tools.
The following text explains the optimization 6 Feedforward control and model-based
of the cascade control which is common in all control
chemical processes. Cascade controllers can be
fast as in PC to FC chains or slow as in AC to In all chemical processes, control quality can
TC or TC to TC cascades controlling product be significantly improved on various important
stream quality measured by on-line analysis or control schemes using feedforward control.
temperature inferential controllers. The Unfortunately, almost all feedforward tuning
optimizer can identify both slave and cascade parameters are estimated today in the control
process dynamics and then optimizes cascade
room by trial-and-error or “best-guessed”
estimates.
The 3G algorithm provides powerful
functionality to mathematically identify the
feedforward parameters: lead, lag, gain and
dead time. Understanding how feedforwards
work allows building custom new applications
all inside in the DCS or PLC for numerous other
innovative purposes. Figure 9: Model-based control scheme for
dead-time compensation.
By mastering the quantitative details of how
feedforwards work, an engineer, operator or
7 Conclusion
technician can easily build powerful
feedforward control schemes inside the DCS or Application of the newly developed
PLC with numerous benefits. Fig. 8 shows a identification and optimization 3G algorithm is
sample calculation overview of a feedforward presented. It can be implemented in the DCS or
control scheme using the 3G algorithm. PLC with increased recurring average annual
profits ranging from 50 000 to 3 million dollars,
depending on the size and type of the plant.
Using the 3G algorithm, it is possible to identify
process dynamics in the presence of
disturbances and noise, design primary and
advanced process control schemes, and optimize
their PID/APC parameters. The ability to
identify system dynamics using this 3G
approach allows successful identification using
ultra-short duration data amidst disturbances
Figure 8: Feedforward control simulation and
and allows optimization of PID tuning and APC
optimization scheme
implementation inside an existing DCS or PLC
Another great opportunity is the in a remarkably short duration, at a lower cost
implementation of model-based controllers in and higher accuracy.
the DCS or PLC. Any model based on rigorous,
empirical, semi-empirical or regressed models 8 Literature
can be implemented in the DCS using once
[1] H. R. Henríquez, C. Cuevas, M. Rabelo, A. Caicedo,
through or iterative calculations. Such models
Stabilization of Distributed Control Systems with
can be used to implement closed-loop Delay, Syst. Control. Lett. 60 (2011) 675–682
controllers in the DCS or PLC. Furthermore, [2] J. H. van Schuppen, O. Boutin, P. L. Kempker, J.
measurement feedback such as from online gas Komenda, T. Masopust, N. Pambakian, A. C. M.
chromatographs or laboratory analysis can be Ran, Control of Distributed Systems: Tutorial and
incorporated into the predictive models. This Overview, Eur. J. Control. 17 (5-6) (2011) 579–602
[3] L. Cauffriez, J. Ciccotelli, B. Conrard, M. Bayart,
model-based controller design with predictive, Design of Intelligent Distributed Control Systems: A
corrective and feedback closed-loop control Dependability Point of View, Reliab. Eng. Syst.
action can also be nicely built using the new 3G Safe. 84 (1) (2004) 19–32
optimization algorithm. Model-based structure [4] J. C. Campelo, P. Yuste, P. J. Gil, J. J. Serrano, A
is shown in Fig. 9. Real-Time Distributed Industrial Control System for
Embedded Applications, Control Eng. Pract. 9
(2001) 439–447
[5] A. Rullán, Programmable Logic Controllers versus
Personal Computers for Process Control, Comput.
Ind. Eng. 33 (1–2) (1997) 421–424
[6] G. Valencia-Palomo, J. A. Rossiter, Programmable [17] H. Peng, T. Ozaki, Y. Toyoda, H. Shioya, K.
Logic Controller Implementation of an Auto-Tuned Nakano, V. Haggan-Ozaki, M. Mori, RBF-ARX
Predictive Control Based on Minimal Plant Model-Based Nonlinear System Modeling and
Information, ISA T. 50 (1) (2011) 92–100 Predictive Control with Application to a NOx
[7] W. Bolton, Programmable Logic Controllers 5 (2009) Decomposition Process, Control Eng. Pract. 12 (4)
1–19 (2004) 191–203
[8] L. Reznik, O. Ghanayem, A. Bourmistrov, PID Plus [18] H. Badreddinea, S. Vandewalleb, J. Meyersa,
Fuzzy Controller Structures as a Design Base for Sequential Quadratic Programming (SQP) for
Industrial Applications, Eng. Appl. Artif. Intel. 13 optimal control in direct numerical simulation of
(4) (2000) 419–430 turbulent flow, J. of Comp. Phy. 256 (2014) 1–16
[9] R. C. Panda, Synthesis of PID Controller for [19] M. Fesangharya, M. Mahdavi, M. Minary-Jolandanc,
Unstable and Integrating Processes, Chem. Eng. Sci. Y. Alizadeha, Hybridizing harmony search
64 (12) (2009) 2807–2816 algorithm with sequential quadratic programming
[10] M. Escobar, J. O. Trierweiler, Multivariable PID for engineering optimization problems, Comp. Meth.
Controller Design for Chemical Processes by in App. Mech. and Eng. 197 (33-40) (2008) 308-
Frequency Response Approximation, Chem. Eng. 3091
Sci. 88 (2013) 1–15 [20] R. Sharmaa, B. Glemmestadb, On Generalized
[11] H. M. S. Lababidi, S. Kotob, B. Yousuf, Refinery Reduced Gradient method with multi-start and self-
Advanced Process Control Planning System, optimizing control structure for gas lift allocation
Comput. Chem. Eng. 26 (9) (2002) 1303–1319 optimization, J. Process Contr. 23 (8) (2013) 1129–
[12] L. Dobos, J. Abonyi, Application of Non-linear 1140
Dynamic Optimization in Advanced Process Control [21] S. Howes, J. LePore, I. Mohler, N. Bolf,
of Product Grade Transitions of Polymerization Implementing Advanced Process Control for
Processes, Comput. Aided Chem. Eng. 28 (2010) Refineries and Chemical Plants, Goriva i maziva 53
559–564 (2) (2014) 297–119
[13] S. Z. Gao, J. S. Wang, X. W. Gao, Modeling and [22] J. Buzása, R. Kicsinyb, Transfer functions of solar
Advanced Control Method of PVC Polymerization collectors for dynamical analysis and control design,
Process, J. Process Contr. 23 (2013) 664–681 Renewable Eng. 68 (2014) 146–155
[14] W. Al-Gherwi, H. Budman, A. Elkamel, A Robust [23] N. D. Ramirez-Beltran, Transfer function models to
Distributed Model Predictive Control Algorithm, J. control a chemical process, Comp. & Ind. Eng. 33
Process Contr. 21 (8) (2011) 1127–1137 (1-2) (1997) 417
[15] A. Willersrud, L. Imsland, S. O. Hauger, P. Kittilsen, [24] URL: http://www.PiControlSolutions.com (3. 3.
Short-term Production Optimization of Offshore Oil 2014.).R. Isermann, Identifikacija dinamičnih
and Gas Production Using Nonlinear Model sistemov 1. del (Times New Roman 10pt italic),
Predictive Control, J. Process Contr. 23 (2) (2013) Fakulteta za elektrotehniko v Ljubljani, Ljubjana
215–223 1987. (Times New Roman 10pt normal).
[16] F. Malchow, O. Sawodny, Model Based
Feedforward Control of an Industrial Glass Feeder,
Control Eng. Pract. 20 (1) (2012) 62–68