Professional Documents
Culture Documents
Generating Operating Procedures For Chemical Process Plants
Generating Operating Procedures For Chemical Process Plants
Access to this document was granted through an Emerald subscription provided by emerald-srm:123683 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for Authors service
information about how to choose which publication to write for and submission guidelines are available for all. Please
visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company manages a portfolio of
more than 290 journals and over 2,350 books and book series volumes, as well as providing an extensive range of online
products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee on Publication
Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive preservation.
Operations management,
Operating systems In this paper we discuss an approach to Figure 1 represents the main stages in the
operating procedure synthesis (OPS) for design, construction and running of a che-
Abstract chemical process plants using AI planning mical plant. It can be seen that operating
Operating procedure synthesis procedures are a consideration at several
(OPS) has been used to generate
technology, considering in detail a particular
case study, the double effect evaporator different points within this. At the design
plant operating procedures for
chemical plants. However, the (DEE) test rig. We argue that a successful stage, where the process is first described at a
application of AI planning to this
approach to OPS has the potential not only high level in process flow sheets, and then in
domain has been rarely consid- more details as engineering line diagrams
ered, and when it has the scope of to reduce the substantial time currently
devoted to developing plant operating proce- (ELD), they may form an implicit element of
the system used has limited it to
solving ``toy'' problems. This paper the design rationale ± that is, the reasoning
dures manually, but also to support the
describes the application of state- lying behind the specific design decisions
of-the-art AI planning techniques to
consideration of operability problems at a
made. Designers will clearly avoid design
the generation of operating proce- much earlier stage in the plant life-cycle,
decisions that they perceive will produce
dures for chemical plant as part of therefore avoiding late changes to the design.
the INT-OP project at the Univer- inoperable plant.
We demonstrate that the use of modern AI
sities of Salford and Loughbor- When the design is evaluated for safety and
planning techniques allows successful gen-
ough. The CEP planner is outlined operability at the HAZOP stage, some of this
and its application to a double eration of plant operating procedures for a rationale may be drawn out as part of the
effect evaporator test rig is dis- real-world plant. While concentrating on
cussed in detail. Particular atten- HAZOP process, but investigation conducted
tion is paid to the issues involved
chemical process plant in this work, we during the project suggests that it is unusual
in domain modelling, requiring the believe it is straightforward to extend our for operating procedures to be documented
description of the domain, devel- system to other continuous process plant and in any formal and systematic way at this
opment of AI planning operators,
that in principle it could also be extended to stage. An output of the HAZOP exercise may
the definition of safety restrictions,
and the definition of the problem. batch plant, giving it a wide and general consist of operating constraints as well
There is then a presentation of the applicability. as design changes, but in the case of
results, lessons learned and pro- the chemical process industry, a multi-
blems still remaining.
Operating procedures and plant life-cycle disciplinary commissioning team is nor-
All industrial plants require an extensive set mally responsible for defining the detailed
of operating procedures which define the sets of operating procedures for a plant
This work was made
possible through funding steps required; for example, to start the plant alongside their other commissioning work.
from the EPSRC grant: up, to shut the plant down, to isolate pieces of This is often a substantial exercise taking
``Integrating Operability into equipment for maintenance or to deal with around two man-years of effort, and usually
Plant Design'', the ESRC, emergency situations. In older plants, all of overlaps with the construction of the plant.
the SERC, BG plc, BP,
Cogsys, ICI and Subs-IAD. these steps are carried out manually by A consequence is that if operability pro-
P.W.H. Chung is grateful to human operators and are usually officially blems are uncovered during this process, as
BG plc and The Royal documented in a multi-volume set of manuals may happen, they may force late changes to
Academy of Engineering for the design of the plant. If this occurs while
financial support through a in the control room, while in highly auto-
Senior Research Fellowship. mated modern plants the lower-level and the plant is actually being constructed it is
more detailed steps are embodied in the plant clearly undesirable and costly. Thus there is
control system. It is clearly vital for reasons an advantage in considering operating pro-
both of safety and efficiency that procedures cedures in a more systematic way and earlier
Integrated Manufacturing are of a high quality and therefore much than at present. This is an important moti-
Systems vation for the development of computer-
10/6 [1999] 328±342 based tools to aid in the authoring of operat-
The current issue and full text archive of this journal is available at
# MCB University Press ing procedures at the pre-HAZOP stage of the
[ISSN 0957-6061] http://www.emerald-library.com
life-cycle.
[ 328 ]
Ruth Aylett, Gary Petley, Figure 1
P.W.H. Chung, James Soutter The plant life-cycle
and Andrew Rushton
Generating operating
procedures for chemical
process plants
Integrated Manufacturing
Systems
10/6 [1999] 328±342
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
Once the plant is running, operating proce- The post-conditions, or effects of carrying out
dures may need modification in the light a GRASP action would then be:
of actual operations experience. If mechan- 1 The object location is now in the gripper
isms for doing this in a controlled way do not of the robot.
exist, there is a risk of operations practice 2 The robot arm is in the same location as
diverging noticeably from the documented before.
procedures. A further source of change is 3 The robot gripper is full.
modification of the plant whether as a result 4 The robot gripper is closed.
of maintenance and repair or of continuous
Note that this general action can be used for
improvement methodologies. The application
any robot in any location grasping any object
of computer-based tools offers the means of
by instantiating its pre- and post-conditions
reconciling the need for flexibility in the
construction and modification of plant oper- to particular values.
ating procedures with the need for accuracy, A particular planning problem consists of a
consistency and accountability for changes. description of the initial state of the world (in
the robot example, the objects in the world
Previous work and the location of each, together with the
OPS is a field of research that has largely initial state of the robot) together with a
been carried out in the chemical engineering description of the goals that should be true in
domain (Rivas and Rudd 1974; Ivanov et al., the desired end-state of the world. The
1980; Fusillo and Powers, 1987, 1988a, 1988b; planner will then build a sequence of actions
Foulkes et al., 1988; Lakshmann and Stepha- from those in its repertoire which when
nopolous, 1988a, b, 1990; Kinoshita et al., 1981; executed will produce the desired end-state
Aelion and Powers, 1991; Crookes and Mac- from the given initial state. In the robot
chietto, 1992), rather than by AI researchers. example a sequence of GRASP, RELEASE,
Yet there is an intuitively obvious relation- and TRANSPORT actions might be con-
ship between an operating procedure and the structed to solve a specific bin-packing pro-
output of an AI planning system. blem.
AI planning constructs a plan automati- Now the steps in an operating procedure
cally using a model of the domain and a are actions to be carried out; the procedure is
knowledge base of relevant actions. Each designed to take a plant from a start state to
action has a set of logical pre-conditions, an end state; each step in the procedure must
defining the situation in which it must be be carried out in the appropriate state and
applied, and a set of post-conditions, defining will result in a new state. This clearly
the effects of carrying it out. For example, a matches the description of AI planning just
general GRASP action for a robot arm might given. However, only Aelion and Powers
have the following pre-conditions: (1991) of the works referenced above have
1 The object to be grasped is in a given seriously considered AI planning technology,
location. and in this case a linear STRIPS type engine
2 The robot arm is in the same location. was used, dating back to the mid-1970s (Fikes
3 The robot gripper is empty. and Nilsson, 1971). Much work has been
4 The robot gripper is open. carried out in AI planning since then, and in
[ 329 ]
Ruth Aylett, Gary Petley, particular planners which use a hierarchy of plant correctly, not only must flows of
P.W.H. Chung, James Soutter actions and which apply ``least-commitment'' chemical be created through paths which
and Andrew Rushton
Generating operating (Weld, 1994) techniques have been developed. meet the safety conditions, and not through
procedures for chemical The failure to draw on these techniques has others, but also ``purging'' ± passing a neutral
process plants limited the scope of the systems developed so chemical down a flow path to clean it ± is
Integrated Manufacturing far to ``toy'' plant domains and has led to a required at a specific stage.
Systems
10/6 [1999] 328±342 number of specific omissions: for example, Then the system described in this paper
the inability to deal with the creation of flows generates the procedure shown at the bottom
of chemicals by the opening and shutting of of Figure 2 in one second on a Sparc IPX
valves, while at the same time reasoning station. Instructions 1-3 result in vessel 1
about the filling of a tank or the starting up of filling with acid; instructions 4-5 purge the
a heater. pipework with water in order to meet the
first safety criterion; finally, instructions 7
and 8 establish the required flows in the
A small-scale example correct order. Note that if Step 7 had pro-
duced the flow of acid through the top route
Having outlined why AI planning is seen as
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
Figure 2
A simple OPS example
[ 330 ]
Ruth Aylett, Gary Petley, INTergrating OPerability (INT-OP) project First it was applied to all the ``toy'' problems
P.W.H. Chung, James Soutter being carried out jointly between Loughbor- discussed in the literature by previous
and Andrew Rushton
Generating operating ough and Salford Universities in the UK. CEP workers in this area (except for those
procedures for chemical is being developed incrementally through requiring the handling of numerical quanti-
process plants case studies of increasing scope and com- ties which were excluded from CEP's scope).
Integrated Manufacturing plexity, and as the case study discussed in Having successfully produced correct oper-
Systems
10/6 [1999] 328±342 this paper shows, is already more capable ating procedures from these examples, it was
than any of the systems referenced earlier. decided that CEP should be applied to a real-
We will only summarise the structure of CEP world plant. The jump from ``toy'' problems to
in this paper. the whole of an ammonia plant, say, was felt
CEP divides the tasks involved in OPS into to be too large, and so a small plant equiva-
three areas: planning using operators, the lent in complexity to a section of a commer-
handling of safety considerations and valve cial plant was sought.
sequencing. The first two of these three areas The case study discussed in this paper used
are handled by a state-of-the-art least-com- a DEE test rig constructed in the chemical
mitment planner (Penberthy and Weld, 1992), engineering department at Loughborough
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
which uses the concept of ``goals of preven- University. Although no longer used, the test
tion'' (Soutter and Chung, 1996) to prevent rig was designed for teaching basic princi-
actions being incorporated into the operating ples of plant operation to chemical engineer-
procedure that will take a plant through any ing students. The layout of the test rig is
unsafe states. Safety is clearly a particular shown in the plant diagram (see Figure 3).
concern in a chemical plant domain: a plan This Figure shows the complexity of the
which moves the plant to a desired end-state domain for this case study, which is much
is unacceptable if, for example, along the way nearer to a real-world chemical plant than
explosive gases have been mixed together. the domains used in the previous work
``Goals of prevention'' are defined as safety referenced above. Not only does the DEE set-
restrictions as part of the overall description up contain a larger number of components
of the plant and those required for the DEE than in most previous domains, but the
domain are discussed below. number of different types of equipment is
CEP deals with valve sequencing as a also large, with valves, controllers, pumps,
special case. A characteristic of the opening heaters, coolers, evaporators, feed tank,
and closing of valves in a chemical plant ± mixing tank and a barometric condenser.
actions required in order to produce flows of The purpose of the DEE is to remove water
chemicals to specified vessels or other com- from a salt-water solution (known as brine).
ponents ± is that the effect of the action at a It is called ``double effect'' because the steam
particular valve is dependent on associated that is evaporated off from the brine solution
actions at other valves. However, an as- in the first evaporation is used to supply the
sumption of the standard AI planner repre- energy for the second evaporation. Because
sentation of actions is that the effect of an the test rig was used for teaching, the
action should be represented through its concentrated brine is returned to the starting
post- point and mixed with water to return the
conditions ± and should therefore always be brine solution to its original concentration of
the same. Valve operations violate this salt, thus allowing the process to continue
assumption (Aylett et al., 1998). indefinitely. A diagram of the basic process is
Therefore, valve sequencing is handled by shown in Figure 4.
a specialist module in CEP that uses an
approach that we call ``action synergy'' and is
based on work by O'Shima (1978) and Foulkes Methodology
et al. (1988). A maze-searching algorithm is
used to find a route for a flow between given Two closely-coupled steps are involved in
start and end points. All the valves around applying a planner to a new domain: knowl-
this route are then closed and finally those edge acquisition and domain modelling. In
actually on the route are opened. Thus CEP the DEE case-study, knowledge was acquired
can be seen as a general-purpose AI planner by reading the documentation on the test rig,
with domain-related specialist additions. visiting the DEE installation, and by inter-
viewing a Loughborough University collea-
gue with an understanding of the working of
the test rig. While there are important issues
Double effect evaporator (DEE)
here, we will not touch on them in this paper.
An incremental approach was taken to the We will, however, discuss domain model-
development of CEP through its application ling in more detail, as the amount of time and
to case studies of increasing complexity. effort required to construct a particular
[ 331 ]
Ruth Aylett, Gary Petley, domain model is a major obstacle to the use of a hierarchy of components and the con-
P.W.H. Chung, James Soutter of AI planning in the solution of real-world nections between them. The complete taxon-
and Andrew Rushton
Generating operating problems (Chien et al., 1996). If CEP is to be omy for the components in the DEE domain
procedures for chemical used by design engineers for real industrial is shown in Figure 5. CEP uses an imple-
process plants plant, it must be straightforward to construct mentation of a hierarchical frame-based
Integrated Manufacturing the domain model. We therefore report the description developed during earlier work at
Systems
10/6 [1999] 328±342 lessons learned from the DEE case study. Loughborough (Chung, 1993) to model indi-
vidual components in a particular plant.
Domain modelling Figure 6 shows the frame definitions for a
After knowledge acquisition, the information component type within the framework of the
acquired must be transformed into a form component hierarchy and then provides an
that the planner CEP can understand and example of the description of one such
use. CEP requires the following: component.
. Domain description (plant model). As the complexity of Figure 3 demon-
. Pairs (knowledge association). strates, the manual entry of these descrip-
. Planning operators (actions that can be tions for every component in a particular
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
Figure 3
Plant diagram for double effect evaporator test rig
[ 332 ]
Ruth Aylett, Gary Petley, Figure 4
P.W.H. Chung, James Soutter Basic process for double effect evaporator test rig
and Andrew Rushton
Generating operating
procedures for chemical
process plants
Integrated Manufacturing
Systems
10/6 [1999] 328±342
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
Planning operators
The next stage is to develop the planning
operators used to produce the plan (the
operating procedure). Where the domain
description gives the static content and
layout of a plant, planning operators define
its behaviour. A CEP operator consists of a
goal(s) that can be achieved when the pre-
condition(s) for the operator are true ±
essentially the STRIPS (Fikes and Nilsson,
1971) representation still widely used in AI
planning systems in spite of all the other
changes in the field since then. The CEP
operator for operating a control valve is
shown in Figure 8.
In this operator, the identifiers starting
with ``?'' represent variables which will be
instantiated with actual components when
the operator is used: ?c with the particular
control valve, ?state1 with the initial state of
the valve, ?state2 with the final state of the
valve. Thus one action can be instantiated to
operate any of the many control valves in
Figure 3. The primitive operator that is
shown in Figure 8 was supplemented with
macro operators (Figure 9), which allow
[ 333 ]
Ruth Aylett, Gary Petley, information about the order in which the pre- Figure 7
P.W.H. Chung, James Soutter conditions must be satisfied to be entered Example pair descriptions
and Andrew Rushton (Aylett et al., 1998).
Generating operating
procedures for chemical Operator development is a time-consuming
process plants and difficult part of domain development and
Integrated Manufacturing one on which there is minimal guidance in
Systems the literature. Yet the correctness and effi-
10/6 [1999] 328±342
ciency of the planning process in a domain
depend very heavily on operator definition.
We therefore summarise the lessons of the Figure 8
DEE case study for operator development in Operator-control valve
the sections below.
Required operators
The initial step is to establish the task
requirements by producing a list of all the
tasks to be carried out in the domain. For
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
Quality of results
The procedures produced were evaluated by
the same domain expert who had been used
for knowledge acquisition. He found the
generated procedures adequate ± in the sense
that the start-up procedure would
[ 337 ]
Ruth Aylett, Gary Petley, through it at any time before there is a flow of HeatExchanger-4 are partially ordered and
P.W.H. Chung, James Soutter brine into the cooler. Now, the smaller the thus can be taken in any order.
and Andrew Rushton time gap between turning on the cooler and However, it may be the case that some
Generating operating
procedures for chemical the brine arriving, within the constraints of actions which are in a particular order in the
process plants safety, the smaller the amount of cooling plan really should be carried out one im-
Integrated Manufacturing water wasted and the more economical the mediately after another, without interpola-
Systems tion of other actions partially ordered with
10/6 [1999] 328±342 start-up of the operating procedure.This kind
of optimisation is not, we feel, well handled respect to them. For example, changing the
by a planner and suggests the need for an operation of the plant from using one eva-
optimising back-end such as that in the work porator to using both requires the opening of
of Crookes and Macchietto (1992). one valve and the closing of another at
virtually the same time to create a flow of
Linking and ordering actions steam through the second heat exchanger.
The output of a partial-order planner such as One would not wish other valve operations
CEP is a plan-network in which only those elsewhere in the plant along a different
actions which must follow each other are branch of the plan graph to be inserted into
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
ordered with respect to each other. Actions this sequence even though this linearisation
which may be taken in any order appear in is formally possible. As a step towards
parallel in such a representation. For exam- capturing this type of dependency, CEP's
ple, a plan to make a cup of instant coffee original linearisation mechanism was
might be represented at a high level as: fill rewritten so that all the actions down one
and boil the kettle; get cup, coffee, sugar and parallel branch of the plan net are placed
milk; put coffee and sugar in cup; add boiling together ± in other words, depth- first ex-
water; add milk. The plan will still work pansion was applied to parallel branches
whether the kettle is boiled first or the cup rather than any other ordering. We are also
and other materials are assembled first, so considering the use of short links between
that these actions may be partially ordered. actions that allow the links in the plan graph
On the other hand, it is not possible to put to carry the information that actions should
boiling water into the cup until after the be kept close together at linearisation
kettle has boiled, so these actions must occur (Soutter and Chung, 1997).
in that order. The plan net supports user interaction in
It is usual for the last step in an AI planner the linearisation process since it shows what
to be the transformation of the plan net into a actions can be moved in a particular linear-
linear sequence, since usually only one-step- isation and which cannot. The re-ordering of
at-a-time execution is planned for. This actions may be desirable for two reasons in
process is known as linearisation. As the addition to the one just discussed. First,
example above demonstrates, there may be grouping actions together for operating a
reasons apart from ``any order that works'' certain piece of equipment makes the plan
for linearising in a particular way: most of us clearer. Second, a ``better'' plan may be
would prefer to put the kettle on before produced by taking the actions in a certain
assembling the other materials since we order. For example, if two valves have to be
know that it takes time to boil the kettle and opened manually, then these actions should
the overall time for the task will be shorter if be together if they are geographically next to
we assemble the materials while this hap- each other in the plant. CEP cannot currently
pens. In addition, while it is still possible to take geographical proximity into account,
make a cup of instant coffee if the milk is however, since the ELD from which it works
added before the boiling water, many people contains only the topological relationship of
would not do this since they feel the coffee equipment.
tastes better if the water really is boiling
when it is poured on rather than being Generic operators
instantly cooled by the presence of milk. We argue that the extent to which operators
Thus questions of efficiency and quality may are generic is a touchstone for the extent to
arise at linearisation. which CEP has principled solutions to the
A number of comparable issues arose from problems of OPS. We have seen above that
considering the linearisation process for CEP both the use of safety restrictions and of
in which its partially ordered output is a specialist valve sequencing component
turned into a sequence of operating instruc- removes the need to solve particular pro-
tions. An extract from a plan net generated blems by manipulating the operators into a
by CEP can be seen in Figure 14 and plant-specific form. An earlier version of the
represents a plan fragment for the operation DEE case study required a number of ad hoc
of an evaporator. Figure 14 shows that fixes, implemented in plant-specific planning
actions OperateGlassPreheater-3, Operate- operators, but careful analysis of the issues
[ 338 ]
Ruth Aylett, Gary Petley, involved produced improvements to CEP ± in functional units, such as the ``back-end loop''
P.W.H. Chung, James Soutter particular its valve-sequencing component ± referred to in connection with the ICI plant.
and Andrew Rushton which have solved these problems in a more They then conceptualised the operation of
Generating operating
procedures for chemical principled way. the plant at an abstract level in terms of these
process plants In the DEE case study, every component is functional units. As discussed in the next
Integrated Manufacturing now operated by a generic operator, all of section, this ``chunking'' process originates
Systems which are found in a component operator
10/6 [1999] 328±342 in design. Each of the functional units we
library. Moreover, 13 of these operators were examined was of the same order of complex-
reused for a subsequent case study using ity in terms of numbers of components and
the back-end loop section of an ICI ammonia connections between them as the DEE. We
plant, and six for the corrosion metal therefore argue that a large plant may be
removal system of a BP acetic acid plant, hierarchically decomposed and that CEP is
demonstrating the value of a library of capable of dealing with its functional units in
planning operators. the same sort of time as the DEE.
A more important issue concerns the
extent to which real-world components in Usability issues
process plant are generic. For example, of the
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
the procedure has been generated within though general-purpose hierarchical least-
CEP through various outputs. Planning can commitment planning algorithms are
be thought of as a search problem for the powerful, real domains also require domain-
correct sequence of actions in a very large specific problem solving along with a great
space containing all possible combinations of deal of domain-specific knowledge. Hope-
actions with all possible instantiations of the fully, in the same way as work in knowledge-
variables in them. CEP allows the user to engineering methodologies has identified
view that section of the search space it specific approaches to different types of
traversed in building its plan, giving an diagnosis, work in a wider range of real-
insight into the choices it made and the world planning domains will begin to estab-
difficulties it encountered. lish abstract categories in such domains
CEP also allows varying amounts of details which will support a more principled
on the planning process to be displayed and approach to the choice of planning technolo-
gives the user some ability to interact with gies for particular problems (Valente, 1995;
the process of plan generation. This is Aylett and Jones, 1996; Soutter and Chung,
achieved by allowing the user to select the 1997).
type of goal for CEP to solve first. Investiga- A number of knowledge engineering issues
tion suggests that this is essential since arose from the DEE case study. First, the
professional engineers would not accept a time and effort required to develop the
system in which planning was wholly auto- domain was substantial, with about 20 per-
matic and they were unable to apply their son-days of effort involved in developing the
own expertise. operators alone (though a proportion of this
The form in which the generated procedure was due to a learning curve which would be
is displayed is also important. CEP can show climbed quicker next time as a result of this
the procedure as a text list of actions, which experience). Development of tools to assist in
are in a similar format to the manually domain development is, we believe, vital to
generated procedures with which the user the use of AI planning to solve real-world
will be familiar. In addition, the procedure problems. The automatic generation of the
can be shown as a plan-net of the type domain description, as for CEP from Auto-
discussed above, showing the ordering rela- Cad, is a step along this path, and the library
tionships between actions in the plan and of generic operators derived from the case
thus the possible orders allowable after study is another even more important one.
linearisation. This allows the user to alter Validation and verification tools such as
the ordering of some of the actions in the those described in Chien (1996) are also
original text list, while respecting the essen- important.
tial ordering constraints. The production of a library of generic
operators in the DEE case study illuminated
a particular problem. It is often possible to
solve problems in a particular domain by
Conclusions
ad hoc fixes, frequently in the planning
A number of conclusions can be drawn from operators. We discussed a number of these
the DEE case study, some specific to the above and remarked that a measure of CEP's
domain of process plant operating procedure ability in this domain lay in how many or few
synthesis, and others of more general rele- such fixes were required. As CEP's capabil-
vance to industrial applications of AI ities were increased, so it became possible
planning. to solve each problem in a more general
[ 340 ]
Ruth Aylett, Gary Petley, and principled way. Thus not only is the AI and Expert Systems, Gordon and
P.W.H. Chung, James Soutter ability to produce a library of generic opera- Breach, Cooper Station, New York, NY,
and Andrew Rushton tors an indispensable tool for domain devel- pp. 277-83.
Generating operating
procedures for chemical opment in the future, but also it is, we argue, Crooks, C.A. and Macchietto, S. (1992), ``A com-
process plants indirectly a measure of the adequacy of a bined MILP and logic-based approach to the
Integrated Manufacturing planner. synthesis of operating procedures for batch
Systems We argue that AI planning technology plants'', Chemical Engineering
10/6 [1999] 328±342
has now reached a level of maturity where Communications, Vol. 114, pp. 117-44.
it can be successfully applied to difficult Fikes, R.E. and Nilsson, N.J. (1971), ``Strips: a new
real-world problems. Just as KBS technology approach to the application of theorem-prov-
in general has made a powerful contribution ing to problem solving'', Artificial Intelligence,
to the management of manufacturing Vol. 2, pp. 189-208.
systems, so AI planning has the potential to Foulkes, N.R., Walton, M.J., Andow, P.K. and
solve problems in this area previously Galluzo, M. (1988), ``Computer aided synthesis
seen as too complex to be tackled success- of complex pump and valve operations'',
fully. Computers and Chemical Engineering, Vol. 12,
pp. 1035-44.
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
[ 341 ]
Ruth Aylett, Gary Petley, Conference on Knowledge Representation and Soutter, J. and Chung, P.W.H. (1996), ``Partial
P.W.H. Chung, James Soutter Reasoning, October. order planning with goals of prevention'',
and Andrew Rushton Petley, G., Aylett, R.S., Chung, P.W.H. and Proceedings of the 15th Workshop of the UK
Generating operating
procedures for chemical Rushton, A. (1998), ``Development of a reusa- Planning and Scheduling SIG, Vol. 2, John
process plants ble operator library for chemical plant Moores University, Liverpool, pp. 300-11.
Integrated Manufacturing domains'', Proceedings of the 17th UK Soutter, J. and Chung, P.W.H. (1997), ``Utilising
Systems hybrid problem solving to solve operating
10/6 [1999] 328±342 Planning and Scheduling SIG, University of
procedure synthesis problems'', Proceedings
Huddersfield, September.
of IChemE Research Event 97, Vol. 2, Notting-
Rivas, J.R. and Rudd, D.F. (1974), ``Synthesis of
ham, pp. 793-6.
failure-safe operations'', AIChE Journal, Valente, A. (1995), ``Knowledge-level analysis of
Vol. 20 No. 2, March, pp. 320-5. planning systems'', ACM SIGART Bulletin,
Soutter, J. (1996), ``An integrated architecture special issue, Vol. 6 No. 1, January.
for operating procedure synthesis'', PhD Weld, D. (1994), ``An introduction to least-
thesis, Loughborough University, commitment planning'', AI Magazine,
Loughborough. Vol. 15, pp. 27-61.
Downloaded by University of South Carolina Libraries At 14:42 12 July 2016 (PT)
[ 342 ]