You are on page 1of 6

Measurement and Control

Step Into
TheVirtual
Plant
The power of this concept will reach out and
grab you. Its ability to warehouse process and
Michael M. Mansy, control-system knowledge offers a vast
Gregory K. McMillan and
Mark S. Sowell III, potential for improved operations by sharing
Solutia, Inc. one common source of information and
avoiding data duplication.

T he decentralization of technical resources


has created the need for a more extensive
and efficient sharing of plant knowledge.
Consider the case where the design origi-
nates in the U.S., the engineering is done in Taiwan,
the distributed control system (DCS) is configured
in Singapore, the displays are built in India and the
At Solutia, we have recently taken the approach
of creating virtual plants that become the warehouse
of process- and control-system knowledge. The vir-
tual plant naturally evolves from the design and be-
comes the common source of information for oper-
ating, maintaining and improving the plant. The
new work needed to create this capability is within
plant is built in Thailand. Also, consider that within the noise level of a project and the potential for
each company, the average turnover rate during the sharing information and avoiding duplication or in-
life of the project is 20%. consistency is tremendous. It is a real-time repre-
Realize as well that companies and engineers sentation of the actual plant running on a standard
have their own individual favorite computer pro- desktop computer. No special hardware is required.
grams for sizing and specifying equipment, piping, The virtual plant starts out as a process flow dia-
instrumentation and valves, as well as for generat- gram (PFD) created from a graphical high-fidelity
ing the process flow and piping and instrumentation steady-state process modeling program. However,
diagrams, and control-system functional diagrams. additional detail is included that is usually associat-
Often, these programs come and go with the indi- ed more with a piping and instrumentation diagram
vidual or the firm. (P&ID) than a PFD. Final elements (e.g., control
Yet, the whole design and construction must valves and variable-speed drives), actuator dynam-
come together seamlessly. Ultimately, local techni- ics, sensor lags and equipment volumes are added.
cians and engineers must smoothly and quickly as- Finally, an interface table to a programmable elec-
sume responsibility for operation, maintenance and tronic system (PES) is completed and communicates
improvement without any assurance of ongoing sup- via object-linking and embedding (OLE) for process
port of the people who designed and built the plant, control (OPC) to the control and interlock system
or understanding of the programs used. configuration and displays. This model can continue
While the globalization of resources cited above to be operated in the steady-state mode or switched
is an extreme example, the proliferation of computer to the dynamic mode to become a virtual plant. The
tools and the diversification of technical resources PES consists of both the basic process-control sys-
have accentuated the need to ensure accessibility of tem (BPCS) and the safety-interlock system (SIS).
all information and to avoid the duplication of effort. There may be some reluctance to invest in a dy-

56 www.cepmagazine.org February 2002 CEP


namic simulation mostly due to the high initial cost and have not been able to build upon the investment in the
maintenance of previous implementations where the dy- steady-state model or its physical properties package.
namic model had to be written from scratch and the These dynamic models required programming the dif-
control system and displays had to be emulated by pro- ferential equations for the unit operations and the con-
grammers. State-of-the-art modeling and PES software trol system and numerically integrating them in special
eliminate these efforts. A properly set up steady-state programs such as MATLAB or Advanced Continuous
model can be readily made dynamic and interfaced to Simulation Language (ACSL). The addition of physical
the actual configuration and displays of the PES. Also, properties, control algorithms and the initial steady
the graphical Windows environment of the modeling state required an intensive effort by highly trained and
program and PES mean that the average engineer can experienced dynamic simulation specialists. These cus-
run or modify the virtual plant. The ease of the ap- tom dynamic simulations are a luxury that most compa-
proach means that the virtual plant is not restricted to nies could not afford. Often, the simulation used for
large or complex processes, but can be employed for control-system studies had to be rewritten in a different
much smaller applications. language, such as FORTRAN or C++, or it had to be
Models are sometimes thought of as artificial or im- configured in a graphical simulator, such as LABVIEW,
practical. This view fails to conceptually understand the to be connected to a PES for control-system checkout
role that models fundamentally play in knowledge and operator training.
building. Advances in engineering and physics use Tieback models came on the scene in the 1980s as an
mathematical models to represent reality. Similarly, inexpensive and efficient method of creating simple dy-
high-fidelity process models use drag-and-drop blocks namics. For discrete control devices, such as pumps and

... [new] computer tools and the diversification of


technical resources have accentuated the need to ensure
accessibility of all information.
of the mathematical relationships for common unit op- on-off valves, the tieback would read the discrete outputs
erations. While all technical knowledge is based on a and send back the appropriate motor-run contact or
model, not all models are based on technical knowl- valve-limit-switch contact action for the discrete inputs
edge. In particular, tieback models that connect the PES after a suitable delay. For control loops, the tieback
inputs to the PES outputs with a simple delay, gain, would read the analog output and respond back with an
and/or lag make the displays functional for basic check- analog input in the right direction with an adjustable
out and operator training, but do not reflect or convey gain, lag and delay. For levels, ramps were initially de-
process knowledge.

Present practices
A traditional steady-state modeling practice is to
omit utility streams, flow resistances, elevations and
prime movers. Since pressure drops do not determine
flow in these steady-state models, very little attention is
paid to the specifying the proper pressure profile.
Often, there are negative pressure gradients where flow
is shown coming from a source at a lower pressure than
the destination. Heaters and coolers with heat duties
computed from a temperature specification are used
rather than heat exchangers with heat-transfer coeffi-
cients and utility streams. The steady-state model starts
and ends in the office of the simulation specialist.
Figure 1 shows a typical steady-state model for a dis-
tillation column. Note that the sump, reboiler, condenser,
and overhead receiver are built into the column sub-flow-
sheet. It is not possible to insert valves, nozzle locations,
pumps and pressure drops for the reflux- or reboiler-cir- Figure 1.Typical steady-state process flow diagram (PFD)-type model
culation flow or to simulate the utility streams. of a column.
Custom dynamic models for control system studies

CEP February 2002 www.cepmagazine.org 57


Measurement and Control

vised and later integrators were set up to simulate the puter. However, this requires a translation of propri-
change in inventory. Tieback models could automatically etary configuration blocks and displays that is expen-
be generated from the configuration. These tieback mod- sive and vulnerable to errors. Often, there is a different
els made the displays and color conditionals operational, interface since the actual consoles are not used, so that
but did not provide an independent test and did not di- learning the navigation of displays for startup, shut-
rectly show flaws in logic or mistakes in assignments. down or abnormal conditions, which is one of the major
The practice to date for operator training and con- objectives of operator training, is compromised. The
trol-system checkout has been to use a tieback model or cost of a custom dynamic simulation and PES emula-
custom dynamic simulation connected to the inputs and tion can easily exceed $1 million. Typically, only nucle-
outputs (I/O) or data highway of a PES hardware sys- ar power plants or high-capacity petrochemical plants
tem (Figure 2). The cost of this setup varies from can justify such a large expenditure.

The virtual plant naturally evolves from the design


and becomes the common source of information for
operating, maintaining and improving the plant.

$100,000 to $1 million depending upon the size and Proposed practices


vintage of the PES. For large systems, it is too expen- The expected pressure drops for valves, exchangers,
sive to have all of the controllers and consoles, so only filters and piping should be added to a steady-state
a portion of the plant is simulated at any one time. A model and built-in sizing routines used to compute the
special room with particular air-conditioning and flow coefficients based on all of the physical property
power-conditioning requirements is also needed to han- data that are in the fluid basis to create the beginning of
dle the actual controllers, I/O card racks, network, and a virtual plant. Nozzle locations should be specified or
operator consoles. Also, with hardware processing and pseudo-pumps inserted, to show the effect of liquid
memory capabilities changing yearly for a state-of-the- head. The individual split-ranged and on-off (isolation)
art PES, the hardware system is obsolete by the time it valves should be depicted to enable the detailing of sep-
is running. Finally, there is no opportunity to either arate characteristics and responses. Finally, the efficien-
speed up or slow down the PES execution to run scenar- cies and characteristic curves of pumps and compressors
ios faster or to analyze confusing situations. should be added to provide the proper pressure-flow re-
An alternative is the emulation of the PES in a com- lationships. Individual series or parallel pumps or com-
pressors should also be inserted to
show the interaction or transition
from one unit to another.
Delta V Excel Actual heat exchangers should be
Add In used with utility streams, rather
than heaters and coolers with heat
OPC duties, to simulate the effect of
Mimic I/O Hysys pressure and temperature changes,
fouling, frosting, piping configura-
Delta V Delta V Delta V tions and control valves in brine,
Hub Controllers I/O steam, Therminol, Dowtherm and
Simulates Field Simulates Field water systems. The omission of this
Discrete Devices Analog Devices extra effort has historically created
a fertile ground for performance
Delta V problems. The unexpected periodic
upsets from these utility streams
due to interactions and control-
valve stick and slip are a major
source of variability. It is usually
Figure 2. Hardware-based system for configuration check out and operator training. too expensive to fix these utility

58 www.cepmagazine.org February 2002 CEP


systems after the fact. Consequently, various control
techniques, such as power-spectrum analysis, are used
to track down the source of the oscillations. Special
tuning of responsive control valves and techniques,
such as feed-forward control, are used to reduce the
variability. The deficiency of current modeling practice
is good news if you are hoping to be contracted for
these automation services.
Figure 3 shows the same column as Figure 1, but
with the sump and overhead receiver modeled as sepa-
rators, and the reboiler and condenser modeled as heat
exchangers. Pumps, valves, measurements, volumes,
steam, and cooling-tower-water streams have been
Figure 3. Column model with piping and instrument design added. A transfer function has been used to simulate the
(P&ID)-type details added. thermowell dynamics for the temperature measurement
and a separator has been placed on the condensate
stream from the reboiler to simulate a steam trap.
Valves are inserted and sized to provide the expected
piping pressure drops. This structure also facilitates the
modeling of a thermosyphon reboiler by a spreadsheet
calculation where the bottoms pump-pressure rise, and
hence, the reboiler circulation flow is a function of the
sump level. The pump pressure rise is used because
nozzle locations do not affect pressure in the steady-
state mode. The increase in time for the steady-state
model to converge is just a few seconds. This more de-
tailed model is saved as a template that can be brought
into other models as a starting point.
Figure 4 shows the interface table that has been
added to connect the virtual process to a virtual PES via
OLE. Any operating condition in the model (e.g., com-
position, density, level, pressure and temperature) can
be inserted into the process variable (PV) export table
Figure 4. Interface table for the object-linking and -embedding (OLE) for
process control (OPC)-connection of the virtual process to the pro- to be sent from the process to the control system. Field
grammable electronic system (PES). switches, such as level switches, can be readily set to
activate via spreadsheets in the process model and
placed into the same PV Export table. Control-valve
analog signals, on-off valve and pump discrete com-
mand signals are read from the PES via the PV Import
table. They are also automatically initialized in the PES
by the model, putting the analog output block or dis-
crete control block in automatic, and writing a setpoint
to match the position or status in the process model.
When the PES link button is enabled, the model ver-
ifies each of the OPC connections and pinpoints any
missing ones. When the integrator start button is select-
ed, each controller output, on-off valve and pump in the
PES is automatically initialized to match the current
valve position and pump status in the process. No pro-
grams or scripts are needed. This initialization is criti-
cal to provide a fast and smooth transition to the de-
sired operating conditions. For slow processes, such as
columns, this is essential, since it can take hours to re-
Figure 5. Interface for monitoring and controlling the speed of the virtual cover from a valve that is in the wrong position or a
programmable electronic system (PES). pump that is not running.

CEP February 2002 www.cepmagazine.org 59


Measurement and Control

A key concept is that there is no duplication, transla- state modeling techniques have been around a long time
tion or modification of the actual control system. This compared to those for dynamic flowsheet modeling.
is of great significance in terms of reduced cost and in- The use of templates for common unit operations, and
creased confidence. The duplication of a complex or sub-flowsheets will eventually make the transition from
advanced control strategy in the process modeling pro- a steady-state PFD to a dynamic P&ID model faster
gram can easily add 50% more time to the modeling ef- than the custom approach presently used, by providing
fort, especially, since the model software lacks a com- a common starting point and method. The result is a
plete set of function blocks used in a PES. The invest- much more accurate representation of the plant plus a
ment in these blocks is incredible and cannot be expect- standardization of techniques that expedite the effort
ed to be a feature of a process simulation program. and expand the potential number of people who can
There is no assurance that the simulated control system create or interrogate the model. An analogous benefit to
will behave the same as the real control system. Conse- the 25% or more effort savings in developing and main-
quently, changes would need to be made in both places. taining control-system configurations gained from the
The process model should be used to represent the field standardization on the control-system type and configu-
equipment, measurements and final elements, and be ration-module library is expected. Of greater signifi-
connected to the actual PES configuration and operator cance is the wider audience of contributors and users of
displays to create a virtual plant. the knowledge base in the virtual plant.
The virtual plant resides in a single desktop comput-
er, or for large processes, in a network of desktops. Benefits
These desktop computers can be at any location and The virtual plant facilitates the simulation of a wide
connected to another desktop or laptop anywhere on the range of operating conditions, process upsets, control
local area network (LAN). You dont need to wait for schemes and equipment conditions, such as condenser

A virtual plant can go where you dont want the real plant
to go or think the real plant can go. Thus, it can show
how to recover from undesirable situations and reveal
unforeseen or doubted capabilities.
satellites for Internet connection. Access and navigation and reboiler fouling, control valve stick-slip and pump
via a modem, analog telephone connection is slower, failures (1). Pressures, temperatures, and concentrations
but sometimes is sufficient for demonstration purposes. can be viewed and changed by simply double clicking
Thus, the virtual plant can be accessed while at home, on the stream and either the condition or composition
traveling or visiting another site. The virtual plant can option. The user can also easily navigate process and
be enhanced with other Windows-based software. For ambient heat-transfer coefficients, volumes and flow
example, software for data archiving, performance coefficients. A pump can be started or stopped and a
monitoring, model predictive control, neural networks valve opened or closed from the operator display.
and real-time optimization can be loaded and interfaced Trials of improvements in batch sequences and regu-
via OPC. The only limits are your imagination and the latory controls can be conducted and the actual imple-
number of desktops in the control network. mentation in the configuration checked out. Expert sys-
A tool built into the virtual PES changes the neces- tems can be developed for advisory control and neural
sary parameters of the desired functional blocks to en- networks can be trained as intelligent sensors to predict
able simulation without making any permanent changes compositions. Model predictive control and optimiza-
to the actual configuration. The interface for this tool as tion strategies can be quickly prototyped. The virtual
shown in Figure 5 displays the status of each functional plant can be used to uncover a common flaw in these
block and allows the user to run the PES system either advanced control techniques, the ability to handle a
30 times slower or faster than real time to match the ac- wide range of normal and abnormal operating condi-
celeration factor set in the process model. tions and plant dynamics. A virtual plant can go where
There is, of course, some reluctance to add these de- you dont want the real plant to go or think the real
tails. The size and convergence time for a steady-state plant can go. Thus, it can show how to recover from un-
model will increase. However, as the processing speed desirable situations and reveal unforeseen or doubted
of PCs increase, this becomes less of an issue. Steady- capabilities. This last advantage of being able to

60 www.cepmagazine.org February 2002 CEP


Table. A Cost and Benefit Comparison of Simulation Methods.

Initial Life Model Model Model Control Ease of Idea Operator


Cost Cost Scope Fidelity Access Studies Prototyping Training

Custom Tieback ! ! " " ! " ! "

Tieback ! ! " "

Virtual "

demonstrate setpoints closer to constraints builds oper- ibrate, troubleshoot and adjust setpoints, equipment, in-
ator confidence. strumentation and control schemes for various scenar-
The portability of the virtual plant means that train- ios and improvements. The time to prototype an idea
ing is not relegated to a training room or a particular can be a matter of a few hours or days. The effort and
audience or purpose. Operator training systems have cost to explore and demonstrate a new idea is no longer
traditionally been used to gain familiarity with the in- an obstacle. Investments in more advanced control tools
terface, process, and to learn how to start up, shut can be analyzed, optimized and justified. PES configu-
down, and to deal with failures (2). It is particularly im- rations with new operating points, equipment or control
portant to provide continual access to these systems to capabilities can be checked out, tuned, exported and
refresh skills in units with frequent transitions, trips, downloaded without modification.
campaigns and high personnel turnover rates. Less ob- The table summarizes the cost and benefits of the
vious is the need to replace operating skills lost from present practices that use hardware or emulated systems
the lack of interaction with the process and loops with tieback or custom models and the proposed prac-
caused by extensive automation. tice of creating a virtual plant. The virtual plant offers
There are huge additional benefits if you consider that the opportunity to warehouse plant understanding,
the operators are potentially the biggest constraint to where each person develops a stake in using, updating
pushing plant capacity, especially when it is beyond the and upgrading the knowledge base. CEP

original nameplate. The gap between the operating point


and the constraint is often set based on a perceived rela-
tionship that may have only occurred one time to one op-
MICHAEL M. MANSY is in the advanced process control group for
erator. The gap based on a war story is typically an order Solutia, Inc. (P.O. Box 66760, St Louis, MO 63141; Phone: (314) 674-
of magnitude larger than the gap dictated by process 8733; Fax (314) 674-7098; E-mail: mmmans@solutia.com). Prior to
variability. Consider how much plant production has in- his current position, he worked as a technology, manufacturing and
design engineer within Solutia. He holds a BS degree in chemical
creased when an experienced process or control system
engineering from the Univ. of Texas at Austin.
engineer is in the control room or an advanced control
system stays online for all of the shifts. High-fidelity GREGORY K. McMILLAN is a retired senior fellow from Solutia Inc.
process simulations and actual control-system configura- During his 33-year career with Monsanto and its spin off Solutia, he
specialized in improving loop performance, controller tuning, valve
tions can be used to increase the knowledge base and
dynamics, opportunity assessments, dynamic simulation, fermentor
skill level of the operator to push the limits safely and re- control, pH control and reactor control. McMillan is the author of 30
liably to achieve this incremental capacity. This can be articles and 12 books, his most recent being the ISA best seller for
accomplished by the operators just choosing better set- 2001, Good Tuning A Pocket Guide. He has contributed to five
handbooks and is the editor of the Process/Industrial
points or by keeping control systems in their highest
Instrumentation and Controls Handbook. He received a BS from
mode, despite rough or confusing conditions. Kansas Univ. in engineering physics and a MS from the Univ. of
Access to the virtual plant should be opened to non- Missouri Rolla in control theory. McMillan is an ISA Fellow and is
traditional audiences, such as maintenance, mechanical presently an adjunct professor at Washington Univ. in St. Louis, MO
and is a consultant through EDP Contract Services in Austin, TX. He
and control-system engineers, so that they can test, cal-
can be reached by phone at (314) 703-9981 or e-mail,
gkmcmi@msn.com.

Literature Cited MARK S. SOWELL III is a process control fellow for Solutia, Inc. at the
Pensacola, FL Nylon Plant (E-mail: mssowe@solutia.com). He has
1. Lo, P., et al., Distillation Tower Pump Failure, Control, pp. 7182 over 20 years of experience implementing process control systems
(Oct. 2000). and advanced control applications. He holds a BS degree from
2. Chin, K., Learning in a Virtual World, Chem. Eng., pp. 107110 Auburn Univ. in chemical engineering and a MS in systems analysis
(Dec. 2000). from the Univ. of West Florida.

CEP February 2002 www.cepmagazine.org 61

You might also like