IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 37, NO.

4, JULY 2007 703

Synthesis of Parallel Operation for Enhanced

Chemical Plant Operation
Hossam A. Gabbar

Abstract—There are many ways to enhance the operation of in industrial organizations where tasks should be scheduled in
chemical batch plants. One way is to use a parallel operation where parallel to use available resources efficiently. Among these prob-
two or more operation tasks are executed in parallel or overlap in lems, operation design has been identified as a critical issue to
a time period. This requires providing a computer-aided operation
design environment with the facility to design and validate parallel realize concurrent engineering for plant operation performance
operation. In this research paper, the different aspects of managing (i.e., cost, time, and resources management).
parallel operation are analyzed, and enhancements are proposed This paper shows a practical approach for the design of
and implemented within the automated operation management parallel operation using formal representation and knowledge
solution CAPE-Oper. Implementation and system design consider- engineering. Parallel operation synthesis for chemical batch
ations are discussed using a case study of a chemical batch plant.
plants is presented to enhance operating procedures synthesis.
Index Terms—Batch recipe, CAPE-ModE, CAPE-Oper, control The following section explores the concept and challenges
recipe generation, parallel operation synthesis, standard operating of concurrent engineering and parallel operation synthesis
procedure formal language (SOPFL).
for chemical batch plants. Section III explains the proposed
parallel operation synthesis mechanism. Section IV presents
I. INTRODUCTION
the implementation in the developed and automated operating
NE CRITICAL problem that faces most production plants
O is the time and cost optimization to meet market chal-
lenges and requirements. The traditional sequential operation is
procedures synthesis solution. Examples will be illustrated
using a case study batch plant.

relatively easy to design and manage; however, it is slow and II. CONCURRENT (OPERATION) ENGINEERING
costly in terms of unused facilities, energy consumption, and hu-
man use in longer operation. There are many initiatives world- Concurrent engineering means simultaneously performing
wide to overcome such problems. One approach is scheduling, both actual and simulated processes of designing and devel-
where many mathematical and artificial intelligence techniques oping a product. All phases of a product are considered con-
are adopted to cover such problems. Scheduling is concerned currently, with the design modified as necessary to make sure
with allocation of resources over time so as to execute the pro- the product is useful at all stages of its lifecycle [5]. Many
cessing tasks required to manufacture a given set of products [1]. attempts have focused on performing concurrent engineering
To manage multiproduct scheduling, such as in batch plants, a during product/process/plant design stage where engineering
mixed-integer linear program (MILP) model is used for short- activities for safety design, maintenance design, and operation
term scheduling of single-stage and multiproduct batch plants design are performed in parallel [6]. Other attempts have been
with the objective to reduce total cost and due orders [2]. Mockus were concerned with the different design rationales in parallel
and Reklaitis [3] proposed an alternative model for batch pro- to reduce design and engineering cost and time [7].
cess scheduling based on a nonuniform time discretization ap- In principle, the operation design process goes through three
proach. In addition to MILP use for batch scheduling, a genetic major stages: 1) general recipe or conceptual operation design,
algorithm is also adopted by many researchers to find optimum which is concerned with identifying the material conversion; 2)
scheduling criteria for a given plant [4]. Although parallel opera- master recipe, which is the detailed operation design required
tion is implemented within the scheduling stage, however, there to produce one unit of the product using the underlying plant
is a need to provide a conceptual framework for parallel opera- design; and 3) control recipe, which is the master operation
tion design based on engineering approaches such as concurrent record to operate the underlying plant and produce the desired
engineering. Concurrent engineering is intended to provide en- product/quality/cost/management. Concurrent operation engi-
gineering design practices to do tasks in parallel, which includes: neering can be applied to these three stages where it can lead
process design, operation design, control systems, computer and to improved plant operation. Fig. 1 shows the advantage of
operating systems used in plant operation, and human manage- considering concurrent operation engineering where the overall
ment. Such a problem is also linked to scheduling in all levels operation design time is reduced to plant operation with min-
imum number of changes when starting the actual production,
i.e., commissioning.
Manuscript received April 25, 2005; revised November 18, 2005. This paper
was recommended by Associate Editor R. Brennan.
The author is with the Division of Industrial Innovation Sciences, Depart-
III. PARALLEL OPERATION SYNTHESIS
ment of Systems Engineering, Graduate School of Natural Science and Tech- Parallel operation is one type of concurrent engineering
nology, Okayama University, Okayama 700-8530, Japan (e-mail: gabbar@cc.
okayama-u.ac.jp). practice, which is used when it is required to execute two or
Digital Object Identifier 10.1109/TSMCC.2007.897441 more operation tasks in parallel (either in the same time or

1094-6977/$25.00 © 2007 IEEE

704 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 4, JULY 2007

trigger using different topological areas might be executed in

parallel, provided that the required resources are available. The
control recipe generation process can also decide parallel oper-
ation where it can match and select topology areas which satisfy
two (or more) operation tasks in the master recipe with similar
initiation triggers. During the execution of plant operation, there
are some operation tasks that might be selected to run in parallel
due to any reason such as safety or abnormal situation [12].
This research focuses on the design considerations of paral-
lel operations and the realization as part of a computer-aided
operation environment. The following section highlights some
limitations to handle parallel operation in different stages start-
ing from the master recipe until control recipe generation.

Fig. 1. Concurrent versus traditional operation engineering.

B. Limitations in Current Control Recipe Generation Practices
to Handle Parallel Operation
A computer-aided operation management system (CAPE-
Oper) [8] is proposed to synthesize master recipe and gen-
erate/synthesize the corresponding control recipe of chemical
batch plants. However, there are some limitations and chal-
lenges that face the proposed solution when dealing with parallel
operation.
1) Lack of Parallel Operation Abstraction Mechanism in
Master Recipe: Currently, there is no standard mechanism in
Fig. 2. Parallel operation design. the master recipe that can be used to define a parallel operation.
User needs to specifically define two or more operation tasks
overlapped in portion of time) in a predefined topological area. for parallel operation. For example, in order to define the above
The use of parallel operation (instead of sequential) will reduce parallel operation, two operations are used: 1) feed water from
production time and cost by processing independent operation water source; and 2) feed emulsifier from emulsifier source.
tasks in parallel. Each operation task should have an initiation trigger and termi-
nation trigger that should ensure the parallel execution of two
A. Decision of Parallel Operation During Operation Design
or more parallel operation. There is no guarantee that the two
The design of parallel operation for chemical plants can be conditions will match.
veiwed in different hierarchical levels of abstraction, namely: 1) 2) Conflict During Control Recipe Generation: For the case
macro level, which is the production planning and management study specified earlier, when generating phase-level operation
level; 2) micro level, which is the chemical process level; and instructions for feeding water, a phase level close instruction is
3) nano level, which is the material design level as a preparation generated for control valve in the emulsifier feeding line. And
for chemical process engineering and production, as shown in when generating phase-level operation instructions for feeding
Fig. 2. The decision of parallel operation during material design emulsifier, a phase-level open instruction is generated for the
level can be inherited into the general recipe, which is in turn same control valve. Such conflict might cause wrong operation
inherited into the master recipe and control recipe. For exam- with possible hazard. This is confusing for both the control layer
ple, during material design, it could be decided to prepare two management and the operator.
input materials simultaneously to obtain the required material. 3) Confusion in the Definition of Control Rules: Each op-
This is similar to a general recipe preparation stage where pro- eration might be directly or indirectly associated with control
cess is designed conceptually (i.e., laboratory level) to produce rules. For example, when feeding water, the destination tank
the required material from the input raw materials. During the should not be full, or the topology node should be available be-
master recipe design stage, an engineer can decide to perform tween source and destination. Another example, when feeding
two operations in parallel such as feeding raw material and water, the emulsifier should be fed for the reaction process of
cooling or heating the jacket. Also, during the design of plant type “A.” It is essential to define a structured way to describe
topology, it might be decided to have two independent topology and synthesize control rules with respect to parallel operation.
lines to support parallel operation. In addition, parallel operation 4) Conflict Between Scheduling and Master Recipe Defini-
might cause changes to plant topology such as providing two tion: During the preparation of the master recipe, an operation
independent lines for parallel operation. Moreover, during the designer or engineer might need to define parallel operation
scheduling process, parallel operations could be used to reduce where it will be independent from plant structure (i.e., topol-
the overall operation time and increase the utilization of plant ogy). During the scheduling stage, parallel operations might
structures and resources. Operation tasks with a similar initiation be defined where their initiation conditions are satisfied. The
GABBAR: SYNTHESIS OF PARALLEL OPERATION FOR ENHANCED CHEMICAL PLANT OPERATION 705

TABLE I
PARALLEL OPERATION REPRESENTATIONS

Fig. 3. Parallel operation levels for operating procedures.

sequencing of operation tasks (including parallel operation)

could be defined during the master recipe stage, while the deter-
mination of plant-specific structures could be organized during
the scheduling process.
Based on the above review of some limitations and difficulties devices such as temperature controllers, etc. Similarly, the job
to handle parallel operation, Section III-C explains the proposed isolation area (JIA) and equipment isolation area (EIA) are de-
design framework to handle parallel operation. fined. It is possible to develop an intelligent algorithm to au-
tomatically partition plant topology based on these hierarchical
C. Mapping Operation and Topology Hierarchies levels, i.e., JIA, unit, cell, OIA, and EIA. The proposed partition-
ing mechanism is independent from automated control devices
Parallel operations can be defined for any recipe level in the because some operations are executed by manual control de-
general, master, or control recipe, as shown in Fig. 3. In this fig- vices/valves, while other operations are executed by automated
ure, there are four layers of operation hierarchical abstraction: control devices/valves. Limiting the partitioning to automated
job, procedure, task, and phase. Job is used to deliver required control devices will cause confusion during operation design and
product(s) as per order specifications. Procedure is the process execution. The concept of OIA is defined as the topology area
executed in cell topological areas to provide required product(s). required to execute a given operation and surrounding control
It usually consists of a set of tasks. Each task is composed of a devices. Similarly, EIA is defined as the dynamically identified
set of phases. Parallel operation can be considered at any level topology area surrounding the underlying equipment (such as
of recipe abstraction: job, procedure, task, or phase. For exam- a reactor), which is needed to isolate that equipment within a
ple, the parallel procedure is usually linked with production line given operation. Again this is dynamically specified based on
scheduling where more than two cells are considered to run in the given operation (i.e., considering the current condition of
parallel. Similarly, the decision to use parallel unit procedures the surrounding control devices around the given equipment).
can be considered during production line or process schedul- In this proposed operation–structure mapping, job is mapped to
ing. Parallel tasks and phases might be considered during the JIA, a procedure is mapped to a cell, a unit procedure is mapped
design of the master recipe or as inherited from general recipe to unit, a job is mapped to JIA, operation is mapped to OIA, and
design (i.e., material design), where it might be required to in- phase is mapped to EIA. In the case of parallel operation, two or
put two raw materials simultaneously in a topology area such more jobs, unit procedures, procedures, tasks, and/or phases can
as the reactor. The focus in this research will be on the parallel be executed in parallel. In such cases, the corresponding topol-
task/phase, which is specified during the master recipe stage. ogy area might be overlapped to use shared topological areas or
The proposed batch control by ANSI/ISA-S88 showed a use- resources as per the topology and operation/production control
ful framework for control recipe structuring [13], [14]. However, constraints and rules. This means two or more jobs might share
there were some gaps to clearly specify the mapping between a topological area in their JIAs. Similarly, two or more unit pro-
structural levels and operation levels. In this research, a new cedures might share topological area in their cells; and two or
concept for mapping structure and operation hierarchies is pro- more tasks might share topological area in their OIAs.
posed. The proposed idea is based on isolating the topology area
required for a given operation using flow control devices such
as control valves. This can be thought at any operation level, D. Parallel Operation Representation
i.e., job, unit procedure, procedure, task, and phase. Operation As per the operation representation proposed by [9], opera-
isolation area (OIA) is the topological area required to carry tion tasks are defined for each operating procedure level where
out the task, and is surrounded by flow control devices (based they are composed of initiation trigger, action, and termination
on the condition of the plant and current position/situation of trigger. During the master recipe stage, parallel operation can
the surrounding flow control devices). This concept is relatively be defined and represented in different forms as described in
easy to understand and does not require any link to control Table I.
706 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 4, JULY 2007

TABLE II
CONTROL RULES/CONSTRAINTS CLASSIFICATION

Fig. 4. CAPE-Oper system architecture.

In case (A), a set of operation tasks in any level can be guage called a standard operating procedure formal language
grouped in an operation subroutine so that they will be executed (SOPFL) [8].
in parallel. This means that they should have the same initiation The proposed computer-aided operation design environment
trigger. In case (B), a set of operation tasks can be defined in enables an operation designer to define plant design using
parallel by simply using the same initiation trigger, which will CAPE-ModE and define the associated domain knowledge such
force them to run in parallel. In case of (C), meta-operation as material, products, structure classes, ports, and subassem-
is defined for two or more parallel operations in the generic blies. The topology analyzer is included where it automatically
topology area for generic operation such as material movement defines the hierarchical partitions of the underlying plant topol-
and/or isolation. In case (D), generic control rules are defined to ogy such as cell, OIA, and EIA. The topology analysis algorithm
force operation tasks to run in parallel. Detailed classification is based on the definition of control devices, which are associ-
of control rules is described in Table II. The intelligent control ated with control flags. In such an algorithm, EIA is defined
rule editor and control layer are required to synthesize control as the smallest topological area that is surrounded by control
rules and validate them during master and control recipe stages. devices. It is usually equipment such as a reactor, tank, furnace,
For the identified parallel operations, initiation triggers will be which is surrounded by control valves. Cell is the topological
synchronized and validated. area required to carry out an unit procedure such as reaction or
Table II shows the detailed classification of control rules separation. Cell area is also surrounded by control valves and
as defined within the proposed solution. Other control includes one or more EIA. OIA is the topological area that is
rules/constraints classification such as quality-, management-, required to carry out an operation such as material movement
safety-, and/or environmental-related control rules/constraints or isolation. Also, OIA is surrounded by flow control devices,
could be discussed similarly. such as control valves, and includes one or more smaller EIAs.
Currently, the identification of EIAs and OIAs is done automat-
IV. PARALLEL OPERATION IMPLEMENTATION ically, while cell identification is done manually. The operation
In this section, the implementation of parallel operation will designer defines (or uses) a set of keywords that are used to
be explained as part of the proposed automated operating pro- define operating procedures in all levels, i.e., procedure, unit
cedures synthesis solution, focusing on the enhancements to the procedure, operation, and phase. SOPFL statements are defined
control recipe generation process. using the predefined set of keywords. The SOPFL-based editor
and parser are embedded in the master recipe editor and con-
trol recipe generator where the master and control recipe are
A. CAPE-Oper
represented in the form of a SOPFL statement. The proposed
In this section, the implemenation of parallel operation syn- solution enables the visualization of the generated control recipe
thesis will be described as part of the developed automated op- as mapped to plant topology. The current design of the proposed
eration management solution (i.e., CAPE-Oper). CAPE-Oper is computer-aided batch control is used to synthesize the master
developed as shown in Fig. 4. It includes a subsystem to design recipe and generate the corresponding control recipe of the ac-
operation, i.e., synthesize the master recipe and generate the tual (newly established) batch plant in Japan.
corresponding control recipe [9], [10]. In the proposed solution, The operation synthesis (including parallel operation) process
the master and control recipe are represented using formal lan- starts with the defining and analyzing plant topology, including
GABBAR: SYNTHESIS OF PARALLEL OPERATION FOR ENHANCED CHEMICAL PLANT OPERATION 707

TABLE III
EXAMPLES OF PARALLEL OPERATION

need to be extracted using solvent. Solvent should also be fed

to the extraction tank, either in parallel to feeding material X or
in sequence.
In the earlier example, it is required to feed water from MT2
to main reactor A1 and feed emulsifier from MT4 to A1. The
decision of parallel operation can be decided from the concep-
Fig. 5. Process flow diagram of case study batch plant as shown in CAPE- tual stage where it is defined as per material processing that in
ModE. order to obtain output material “A,” water and emulsifier should
be fed in parallel. In case there is safety, quality, or cost restric-
process model, which is defined in three ways: structure, be- tions to allow feeding these two materials simultaneously, then
havior, and operation. CAPE-ModE is used to capture plant a control rule (i.e., constraint) will be defined at the concep-
topology and associate materials, products, functions, etc. Con- tual stage, which will be considered during the master recipe
trol rules and process constraints are identified as associated synthesis stage. In some cases, the decision of parallel oper-
with the different model elements and in different abstraction ation is not clear or cannot be considered during the material
levels. In order to synthesize operating procedures, operation design or conceptual stage. In such cases, parallel operation
tokens and recipe definition statements are defined and used might be decided in later stages such as during master recipe,
to synthesize the master recipe and generate the corresponding scheduling, control recipe generation, and/or execution via dis-
control recipe. Fig. 5 shows the used case model of the proposed tributed control system (DCS)/sequential function chart (SFC).
operating procedures synthesis solution, CAPE-Oper. If an operation designer decided to perform such operation (i.e.,
Section IV-B provides detailed examples using different par- feeding water and emulsifier) in parallel for efficiency or quality
allel operation scenarios in different abstraction levels. reasons, a suitable parallel operation representation can be used
as described earlier (i.e., parallel subroutine, meta-operation, or
B. Parallel Operation Examples control rule). A parallel operation subroutine can be used by
defining two tasks associated with the unit procedure to prepare
In this case study, the focus will be on the design of a par- materials for reaction in unit A1. The first task is to move water
allel operation starting from master recipe preparation until the from feed to tank A1, while the second task is to move emulsifier
generation of the corresponding control recipe. In addition, con- from feed to tank A1.
siderations will be made on group operation when it is required In the example shown in Fig. 5, some possible operations that
to perform a set of operations in sequence. The discussion of can run in parallel (or in sequence) are defined as in Table III,
how to decide parallel operation is kept outside the scope of this which shows material movement and area/line isolation.
research paper where it requires detailed explanation of process Section IV-C describes enhancements to the control recipe
design alternatives as linked to process operation. The example generation mechanism to manage parallel operation synthesis.
shown earlier is used to describe a mixing or extraction process
where two or more input material are fed to the processing unit
such as tank, and output materials are produced and discharged, C. Enhanced Master-to-Control Recipe
as shown in Fig. 5. Generation Mechanism
1) Mixing: In case of mixing, it is required to feed mate- Fig. 6 shows the proposed enhanced mechanism to generate
rial X and material Y to the tank, and in the same time, allow the control recipe with the considerations of parallel operation.
for releasing or discharging air from the tank. The feeding of This flowchart shows the synthesis of material movement,
material X and material Y can be done in sequence or in par- while isolation operation synthesis can be derived similarly. In
allel as per the control rules or safety restrictions. This can be step one, and for each material movement operation, identify
represented using any of the proposed representation methods the physical source and destination equipment. This will be
described earlier. limited to the equipment available in the underlying unit, which
2) Extraction: Similarly, an extraction operation can be syn- is already selected when generating the control recipe of unit
thesized. Extraction is to separate or absorb material X from procedure. This is derived using unit-level scheduling infor-
process fluid using solvent. Extraction can be done by feeding mation, which is kept outside the scope of this paper. For each
process fluid, which usually includes (undesired) material X that source and destination equipment, define all possible topology
708 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 4, JULY 2007

Fig. 7. P&ID of case study—two units.

Fig. 8. Improved topology knowledge model.

topology analyzer to support the synthesis of parallel operation,

which have been achieved in two major tracks by the following:
Fig. 6. Enhanced control recipe generation mechanism.
1) improving knowledge structure and representation of plant
topology and the associated domain knowledge;
nodes. In this step (i.e., 2), control rules will be validated to 2) improving knowledge mining and discovery including the
avoid topology nodes that are used by other operations, or not inference engine to select the suitable plant topology area
available for any reason (e.g., maintenance). This is done using for each operation, and for parallel operation.
active table, which keeps track of the status of each active op- 1) Improved Topology Knowledge Modeling: As part of the
eration, initiation trigger, activation trigger, and used topology development of computer-aided design environment (CAPE-
area. The optimum topology node will be identified in terms of ModE), which is used as an engineering user interface in the au-
cost, time, criticality, and other predefined criteria to select the tomated operating procedures synthesis (CAPE-Oper), the data
optimum topology node. The defined control rules for parallel model of the plant topology has been developed [10], [11]. In the
operations will be applied at this stage, i.e., steps 4, 5, and 6. developed data model, structure equipments are associated with
While searching for the topology line to execute one operation ports, function, material, topology nodes, and partitions, i.e.,
such as material movement, a topology line will be ignored when cell, OIA, and EIA. In addition, structure is linked to operation
one or more segments of that line are used by other operations. model elements where procedure is linked to cell, unit proce-
The phase-level operation will be generated such as open or dure is linked to unit, operation is linked to OIA, and phase-level
close valves in the selected topology line path. In case of close operation is linked to EIA and equipment. To support the syn-
(or isolate) line, the nearest control valve to the destination thesis of parallel operation, an improved topology knowledge
equipment will be closed, without closing other valves in the model is constructed, as shown in Fig. 8. The tracking of line
selected topology line. This will be explained in more detail segments and topology nodes along with their status, i.e., ac-
in Section IV-D as part of the proposed intelligent topology tive, under maintenance, in use, or idle, is essential to manage
analysis algorithm. parallel operation during scheduling, control recipe generation,
as well as during the execution of operating procedures. Topol-
ogy constraints are defined and associated with equipment, line
D. Enhanced Plant Topology Analyzer segment, topology line, operation task, and control equipment.
Fig. 7 shows another example of a two-unit process and in- In the master recipe design stage, when designer specifies par-
strumentation diagram (P&ID). A topology analyzer is devel- allel operation, the system validates the topology knowledge for
oped as part of the proposed CAPE-ModE [11] to analyze the the availability of such parallel topology lines. Table IV shows
underlying plant topology and to identify the different structure examples of topology knowledge from the above example.
hierarchical partitions, i.e., JIA, cell, unit, OIA, and EIA, as well The examples of the primitive and derived knowledge are
as topology nodes. Enhancements are developed as part of the represented using binary relationships in first-order predicate
GABBAR: SYNTHESIS OF PARALLEL OPERATION FOR ENHANCED CHEMICAL PLANT OPERATION 709

TABLE IV TABLE V
PLANT TOPOLOGY KNOWLEDGE PARALLEL OPERATION EXAMPLE

eration solution [9], [10] is based on the synthesis of a master

recipe using a formal language called SOPFL [8], which en-
ables the operation designer to capture the master recipe and
define the detailed operating procedure tasks using a predefined
metalanguage. The editor and parser are proposed to synthe-
size operation tasks for each operating procedure level, i.e.,
procedure, unit procedure, operation, and phase, in the form of
initiation and termination triggers and actions.
For the example shown in Fig. 5, there are two actions needed
to be performed in parallel: 1) action 1—move water from water
logic. In addition to topology knowledge, constraints are defined feed (MT2) to main tank (A1); and 2) action 2—move emulsifier
and associated with plant topology element/area/path. In the from emulsifier feed (MT4) to main tank (A1).
earlier example, the topology node of TNK1 and TNK2 cannot The implementation in the master recipe can be realized using
be used simultaneously, while the paths of TNK3 and TNK4 any of the following ways.
should be used simultaneously. In the case study shown in Fig. 5, 1) Using Similar Initiation Triggers: The simplest way to
one can define a topology constraint to avoid the parallel use implement parallel operation is to define similar initiation trig-
of two topology lines: 1) from MT2 to A1; and 2) from MT2 gers for action 1 and action 2. This can be realized using a popup
to AJ2. Such a topology constraint can also be realized using a list of all defined initiation triggers and link them to any opera-
special control valve instead of the splitter T-OUT3, where the tion task that need to be executed in parallel. The first operation
output line can only be opened (or closed) in one way. task will be defined normally as explained in [9] and [10]. When
2) Topology Knowledge: Machine learning is the study of defining second or next operation action(s), which are defined
computational methods for improving performance by mecha- as parallel, a user can simply select an initiation trigger from the
nizing the acquisition of knowledge from experience. Machine list of all defined initiation triggers, previously used. This will
learning aims to provide increasing levels of automation in the minimize the time required to reenter a similar initiation trigger.
knowledge engineering process. In the initial design of the topol- 2) Subroutines for Parallel Unit-Procedure/Operation: In
ogy analyzer [8]–[11], plant topology data and knowledge mod- this implementation option, a subroutine can be defined, which
els and instances are constructed and used to generate the control includes a set of operation tasks to be executed in parallel. The
recipe. In this research, additional topology knowledge is dis- header of the defined subroutine is marked with a “Parallel” flag,
covered using rules inductive and learning techniques, which are which could be in the unit procedure or operation level. Within
used to answer queries related to finding or validating the use of the defined parallel operation or unit procedure, all operation
topology nodes. As shown in Table II, knowledge such as up- tasks will be executed in parallel where they will have similar
stream or downstream equipment, and parallel topology nodes initiation triggers.
are validated and obtained. For example, from the following When the recipe header record is marked as “Parallel,” then
knowledge statements “CV12 is connected to T-IO5,” “T-IO5 the system will understand that all defined operation tasks will
is connected to T-IO7,” “T-IO5 is connected to T-IO6,” and be executed in parallel. The system will accept only one set of
“T-IO7 is connected to CV17,” the system detects that “CV12 initiation triggers for the whole set of operation tasks in the un-
is connected to CV17,” which is learned using a transitivity derlying operation level (i.e., unit procedure, operation, and/or
relation. The enhanced topology analyzer validates the associ- phase). Table V shows one example of parallel operation as im-
ated rules and constraints while analyzing the underlying plant plemented within AOPS, where water and emulsifier feeding is
topology. The description of the proposed knowledge mining done in parallel. The detailed operation tasks associated with
and discovery is discussed separately in more details. both operations will also be performed simultaneously, accord-
ing to the initiation and termination triggers.
3) Definition of Control Rules: Parallel operation can be de-
E. Implementation in Master Recipe
fined using control rules. For example, operation-action 1 and
The master recipe is a recipe type that accounts for equipment operation-action 2, described earlier, can be forced to run in
capabilities and may include process cell-specific knowledge to parallel using a control rule which explains that: moving wa-
produce one unit of the required product for the underlying plant ter from water feed to main tank should be done in parallel to
(ANSI/ISA-S88). The proposed automated control recipe gen- moving emulsifier from the emulsifier feed to the main tank for
710 IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 4, JULY 2007

which can be associated with any operating procedure. Two or

more parallel operation actions, with a generic initiation and
termination trigger, as well as a set of parameters can be defined
using the meta-operation editor on the basis of SOPFL.
In the example shown earlier to move water and emulsifier
from their source tanks to the reaction tank A1, meta-operation
can be defined as in S1 as “Parallel: move material X from its
source to structure-unit S and move material Y from its source to
structure-unit S” using the SOPFL-based editor and embedded
parser explained by Gabbar et al., [8], [9].
Parallel operation can provide several benefits where multiple
cells or topological areas can be used simultaneously for better
productivity and simple operation. This has positive impact on
the simplicity of the scheduling process.
Fig. 9. Rule editor as part of CAPE-Oper.

the reaction operation. The control layer will force these two F. Implementation in Control Recipe Generation
operation actions to run in parallel during execution. Also, the The control recipe is the masterpiece that includes the detailed
control layer will validate other operating procedures, i.e., ac- operation steps to produce the desired product on a specific plant,
tions, to avoid any conflict with two such operations. The control including quality and safety considerations (ANSI/ISA-S88).
rule editor is used to define such generic control rules, including The generation of the control recipe from the corresponding
those for parallel operation. master recipe is based on the mapping between plant structure
Another case is where control rules can be used to represent hierarchy and operation hierarchy: procedure to cell, unit pro-
parallel operation restrictions, such as “Operation-action 1 and cedure to unit, operation to OIA, and phase to EIA. The control
operation-action 2 should NOT run in the same time.” This can recipe generation process, which is a part of CAPE-Oper, au-
be represented using control rules that will be evaluated during tomatically generates all operating procedures in all hierarchies
the synthesis of the master recipe, during the generation of the by identifying the actual physical plant structure for each op-
control recipe, and during the execution of the batch control eration. This requires performing the scheduling process in all
recipe. It is important to specify a practical way to represent structure hierarchical levels in view of the predefined plans.
such generic control rules so that they can be synthesized, Operating procedures of the control recipe of parallel oper-
evaluated, and executed. ations are generated from the corresponding master recipe as
SOPFL [8] is used as a basis to define formal control rules. part of the control recipe generation practice. In case of parallel
First, metarules are defined and used as a metalanguage to define operation, the control recipe is generated from the master recipe
control rules as composed of keywords called tokens. A rule edi- in any of the three forms: similar initiation triggers, use of
tor and embedded parser is developed as part of the proposed op- parallel subroutines, use of control rules, or meta-operation. The
erating procedure synthesis solution (CAPE-Oper) [8], as shown control recipe generation mechanism described by Gabbar et al.
in Fig. 9. Three abstraction levels of control rules can be defined: [8]–[10] can handle all the four implementation alternatives
a) metarule, which describes the syntax or language used to de- of parallel operation from the master recipe. The modified
scribe control and process rules; b) generic rules, which are algorithm described in Fig. 6 shows how to generate the control
defined using material and structure classes, i.e., independent recipe of parallel operations using CAPE-Oper. One major
from the underlying plant-specific information; and c) plant- modification is applied to the topology analyzer so that it can
specific rules. The proposed rule editor can be used to define all validate topology segments against any active operation to
the three levels of abstractions of control rules as well as process avoid the duplicate use of topology lines. In addition, it should
constraints. The control recipe generator program will validate validate all topology segments (i.e., parts of topology lines
the defined control rules, including those for parallel operation between two adjacent structure units/elements) to be sure that
control, and generate the detailed control recipe accordingly. they are not used by any other active operation. For example,
4) Meta-Operation: Another alternative to realize parallel when moving water from MT2 to A1 and moving emulsifier
operation is to define the meta-operation in generic form as from MT4 to A1, all segments should be available between MT2
part of the master recipe stage. Meta-operation is a generic and A1 as well as between MT4 and A1. This is essential to
form of operation library, which is used to define a generic execute both operations in parallel. Once operation is activated,
form of operation libraries that can be used to define more these two topology lines along with their segments are marked
complex operating procedures [10]. Meta-operations are used as in use, until the termination of these two operations. In case
to define operations such as cooling, initialization, heating, or of using control rules, the control recipe generator will validate
isolation operation. It enables the structuring and modularization all available (i.e., defined) control rules in each step. In case
of operating procedures using hierarchical definitions. In some control rules are defined for parallel operation, the generated
cases, parallel operation can be generalized so that it can be control recipe will include operation tasks marked as “Parallel,”
defined using the meta-operation of generic parallel operation, with a similar initiation trigger, or none.
GABBAR: SYNTHESIS OF PARALLEL OPERATION FOR ENHANCED CHEMICAL PLANT OPERATION 711

TABLE VI REFERENCES
EXAMPLE OF THE GENERATED CONTROL RECIPE
[1] M. Pinedo, Scheduling: Theory, Algorithms and Systems. Englewood
Cliffs, NJ: Prentice-Hall, 2001.
[2] P. M. Castro and I. E. Grossmann, “An efficient MILP model for the
short-term scheduling of single stage batch plants,” Comput. Chem. Eng.,
vol. 30, no. 6–7, pp. 1003–1018, May 15, 2006.
[3] L. Mockus and G. V. Reklaitis, “Mathematical programming formulation
for scheduling of batch operations based on nonuniform time discretiza-
tion,” Comput. Chem. Eng., vol. 21, p. 1147, 1997.
[4] E. Kondili, C. C. Pantelides, and R. W. H. Sargent, “A general algorithm
for scheduling batch operations,” Comput. Chem. Eng., vol. 17, p. 211,
1993.
[5] R. E. Knox and R. J. Daty, “Concurrent engineering: New technologies
for concurrent engineering,” CALS J., vol. 3, no. 1, pp. 63–67, 1994.
[6] P. M. Herder and M. P. C. Weijnen, “A concurrent engineering approach
to chemical process design,” Int. J. Prod. Econ., vol. 64, no. 1–3, pp. 311–
318, Mar. 1, 2000.
The major changes will be made in the scheduling process [7] S. Y. Han, Y. S. Kim, and T. Y. Lee, “A framework of concurrent process
where it is required to consider the requirements of parallel engineering with agent-based collaborative design strategies and its appli-
cation on plant layout problem,” Comput. Chem. Eng., vol. 24, no. 2–7,
operation while allocating and scheduling the topology (i.e., pp. 1673–1679, Jul. 15, 2000.
physical) resources in all the hierarchic levels: cell, unit, and [8] H. A. Gabbar, Modern Formal Methods and Applications. New York:
OIA/EIA. When parallel operation is defined, the scheduler Springer-Verlag, 2006.
[9] H. A. Gabbar, A. Aoyama, and Y. Naka, “Recipe formal definition lan-
will ensure that the two (or more) topology lines are available guage for operating procedures synthesis,” Comput. Chem. Eng., vol. 28,
for each parallel operation, in view of domain knowledge and no. 9, pp. 1809–1822, Aug. 15, 2005.
required resources. [10] H. A. Gabbar and Y. Naka, “Computer-aided operation design environ-
ment for chemical production plants,” in Proc. IEEE 12th Int. Conf. Com-
In case of parallel operation, the generated corresponding put. Theory Appl. (ICCTA 2003), Alexandria, Egypt, Aug. 26, 2003.
control recipe will also be marked with the “Parallel” flag, and [11] H. A. Gabbar, A. Aoyama, and Y. Naka, “Model-based computer-aided
will have the same initiation trigger for all parallel operation design environment for operational design,” J. Comput. Ind. Eng., vol. 46,
no. 3, pp. 413–430, Jun. 2004.
tasks. Table VI shows the generated control recipe of the exam- [12] H. A. Gabbar, “Computer-aided enterprise safety management (CAPE-
ple shown in Table V above. SAFE) in plant enterprise engineering environment (PEEE)” Ph.D. dis-
sertation, Okayama Univ., Okayama, Japan, 2001.
[13] Batch Control Part 1: Models and Terminology, ANSI/ISA-S88.01, 1995.
V. CONCLUSION AND FUTURE WORK [14] A. Aoyama, I. Yamada, R. Batres, and Y. Naka, “Development of batch
process operation management platform,” in Proc. 10th Eur. Symp.
In order to overcome time and cost challenges in chemical Comput.-Aided Process Eng., 2000, vol. 24, pp. 519–524.
production plants, it is required to perform operation tasks in [15] K. E. Arzen, “Grafcet for intelligent supervisory control applications,”
Automatica, vol. 30, no. 10, pp. 1513–1525.
parallel. This paper showed a practical approach to achieve such [16] P. Bourseau, L. Valadou, and G. Muratet, “Coupling intervals constraint
a target where hierarchical operation levels are mapped to plant propagation and assumptions-based reasoning for flowsheet analysis,”
hierarchical levels. A concept of job isolation area is introduced Comput. Chem. Eng., vol. 18, pp. S289–S293, Dec. 1994.
[17] B. Carre, “Reliable programming in standard languages,” in High-
to isolate the topological area required to carry out a given job. Integrity Software Computer Systems Series, C. Sennett, Ed. London,
Similarly, EIA, OIA, unit, and cell topological areas are defined U.K.: Pitman, 1989, ch. 5.
to carry out operation in different hierarchical levels, i.e., unit
procedure, procedure, task, and phase.
Parallel operation can be defined for any operation hierarchi-
cal level, i.e., job, unit procedure, procedure, task, and phase, Hossam A. Gabbar (SM’04) received the B.Sc.
and Master’s degrees from Alexandria University,
where the associated operation tasks (i.e., at each level) are exe- Alexandria, Egypt, in 1988 and 1990, respectively,
cuted in parallel or in an overlapped time period, using a shared and the Ph.D. degree in computer-aided process
topological and other resources. Overlapped operation tasks are safety from Okayama University, Okayama, Japan,
in 2001.
also proposed where the initiation and termination conditions He is currently an Associate Professor in the Di-
might be matched. The proposed mechanism is realized and vision of Industrial Innovative Sciences, Graduate
implemented as integrated with the developed automated oper- School of Natural Science and Technology, Okayama
University. He has worked as a Software Engineer, IT
ating procedures synthesis solution of chemical batch control Project Manager, and Consultant in several industrial
solution (CAPE-Oper). projects in different disciplines such as oil and gas, manufacturing, investment,
To achieve the complete solution for parallel operation, still telecommunications, marine, and the chemical/pharmaceutical industry. In the
academic side, he worked in research centers in the areas of marine supply au-
there are further research and development tasks and work to tomation and coast protection. He joined the Tokyo Institute of Technology and
be continued. For example, a batch scheduler can be used to Japan Chemical Innovative Institute, where he participated in national projects
investigate more parallel operation opportunities by analyzing related to batch plant control, oil and gas offsite systems, biomass production
systems, and plastic production chain with recycling. He is specialized in the de-
dependent and independent operations tasks as mapped to plant sign of intelligent systems and computational intelligence methods for process
topology and other required resources, as well as production engineering. He is the author of more than 80 publications, including books,
schedules. In addition, process simulation can be used to eval- book chapters, patents, and papers in international journals and conference pro-
ceedings. His recent achievements include design of innovative solutions for oil
uate the different possible concurrent operation scenarios for and gas offsite plant operation, intelligent fault diagnostic systems, and intelli-
more efficient production (i.e., agile manufacturing). gent systems for future energy production chain planning.