close

Вход

Забыли?

вход по аккаунту

?

Generating Operating Command Sequences for the Regulation of Complex Chemical Processes Using Artificial Intelligence Techniques.

код для вставкиСкачать
Generating Operating Command
Sequences for the Regulation of
Complex Chemical Processes Using
Artificial Intelligence Techniques.
J.R. Roach', B.K. O'Neill and P.J. Strimaitis
Department of Chemical Engineering,
University of Adelaide
Adelaide 5005
South Australia.
Complex pmessing plants are operated (and for that matter designed) by individuals
who cannot always envisage in sufficient detail the longer-rangeconsequences of their
current actions. This is particularly the case where interactions can occur between
sequential operating modes of a plant. Given that the desired outcome is the
achievement of the 0vera.U process goals safely and efficiently, it is worthwhile
considering the factors which can mitigate against the achievement of these aims and
how they can be avoided. This study describes methods which can be used to help
develop safer and more efficient plant operation by the use of the computer to devise
operating command sequenceswhich achieve consecutivesets of process goals subject
to a variety of constraints (e.g. the avoidance of unsafe states). The operating
command sequences are required to be generated in the form of a sequence of valve
operations. A catalyst-regenerationplant example originally devised by Rivas and
Rudd (1974) is aesented.
Introduction
The complexity of modem process plants means that the operator cannot always
envisage in detail the longer-rangeconsequencesof their current decisions. Rivas and
Rudd (1975) have demonstrated that current operating decisions may lead inexorably
to a system mode from which extraction is impossible without catastrophic
consequences. Even in comparatively simple situations, operators in difficult
situations may face a combhatorial number of choices of action. I h e result is that
they may have no real chance of effectively evaluating the ultimate consequences of
their proposed actions before their operating decisions must be implemented. In
practice, only the normal procedures for start-up, running and shut-down modes
together with perhaps a few major emergency situations may be considered and
examined in detail a-priori. Often this engenders a false notion of security,
particularly in partially disabled plants.
Other early papers which addressed this general problem are those of Rivas and
Rudd (1974) and Rivas, Rudd and Kelly (1974). followed by OShima (1978). They
*
Author for correspondence
85
J.R. Roach, B.K. O'Neill and P.J. Strimaitis
all recognized that mass flow conditions provide the dominant constraints.
Subsequent developments have been reported by Tomita et J (1986). Fusillo and
Powers (1987, 1988) , Lakshmanan and Stephanopoulos (1988 a & b, 1990).
Foulkes et al. (1988), and Bunn and Lees (1988). The various works of Rims and
Rudd contain the essential ideas for a huts and bolts' representation of the flow paths
and inventory states of a process system. Representation at this very basic level is
consided neceSSary for the gemation of accurate and unequivocal results that can be
readily followed and understoodby pxucess engineers and opemtom who may have to
implement them in actual operational situations. OShima attempted to develop
these ideas further but was unable to report any significant advances although some
conceptual progress was made in the form of goal description and path search
technique.
However, Tomita et al. (1978) and Fusillo and Powers (1987,1988) attempted a
more abstract representation with a view to developing a more general "topdown"
approach. This involved starting with a broad concept like "start-up plant" and
attempting to expand this goal downwards in a hierarchal Sense through a series of
less abstract concepts with a view to identifying a concrete sequence of primitive
commands such as 'open valve x', 'close valve y' and the like. In a sense, this
attempt is analogous to trying to automate a process design methodology in a highly
constrained environment, viz. trying to infer valid processing sequence with the
equipment specified in advance (albeit appropriately)rather than following the nomal
design procedure of determining equipment requirements subsequent to the
specificationof the process conditions.
Such a task is considered to be significantly mure difficult than the determination
of suitable valve sequencesonce the process goal sequencehas been fixed. This latter
task is the subject of this paper. However, Once the method for the determination of
the basic operating sequences for a given set of process goals has been unraveUed, it
is a logical next step to attempt the more general 'topdown' approach and try to use
the computer to determinethe appropriateprior process goal sequenceas well.
It is difficult to determine the degree of success achieved by Tomita et al as their
results are not reported in sufficient detail. In the case of Fusillo and Powers (1987).
in an observationwhich tends to confii the above comments, they report that
"Thelow-level procedural stepsfor dfferent units may cause interactions which
are not possible to predict using the abstractfunctional modelling of POPS."
The 'primitive' representation of earlier workers for mass flow regulation is
furtherdeveloped and expanded in this paper and a successful method for the synthesis
of operating command Sequences from a given set of process goal is described.
Problems involving heat and momentum transfer are not addressed at this stage but
such constraints could be considered. Further, it is considered that the work of
Sacerdoti (1977) is also likely to be similarly relevant to future developments,
particularly the notion of 'criticism'.
A Description of the General Problem: The general problem has been well
described by Rivas and Rudd (1974). They focus on the process industries where
potentially dangerous material moves in a complex fashion through piping systems,
vessels, reactors and the like. A variety of dangerous events are possible. Operating
policies must then be generated to achieve complex process objectives while avoiding
these dangerous events, especially when the system is already partiaUy disabled from
previous events. In terms of 'primitive' operations, operating plans can include in
86
Generating Operating Command Sequencesfor Regulation of Complex Chemical Processes
the sequences, operationswith electric circuits as well as valve operations (OShima).
Statistics confirm that more than 80% of all accidents in chemical plants can be
attributed directly to mistakes in valve operations. Hence, the need for the
development of improved methods for sequencing such operations.
The Objectives: It is the aim of this work to describe methods which can be
used to develop safer and more efficient plant operation by the use of the computer to
devise operating command sequences which achieve consecutivesets of process goals
subject to constraints of the type previously outlined, In this case, operating
command sequences are required to be generated in the form of sequences of valve
operations. Checking for the repetition of patterns of valve state and component
disposition is a necessary extension for the analysis of cyclic operations. However,
the software described in this work only partially addresses this problem. No
extensions have been made at this stage to include systems requiring the starting of
drives for pumps, compressors, conveyors and the like.
Speciiic objectives are:
1. The determination of a method for representing in the computer the overall
process operating strategy as a sequence of independent process goals for
input to the operating command sequence generator.
2. The determinationof a methodologyfor the generation of operating command
(valve action) sequences for the individual independent process goals cited
above on a one to one basis and which meet imposed safety and efficiency
criteria.
3. The determination of a methodology for 'criticism' of the above sequential set
of independent operating command sequences so that they may be considered
as an integrated whole with redundancies removed.
4. The production of interactive computer code that can be used to apply the
above techniques to realistic process design and operation problems.
The Computer Language Used Although there is no explicit identification of
the computer language used by the earlier workers, there are certain indications that it
may have been FORTRAN. On the other hand, Tomita et al, and Fusillo and
Powers both cite the use of the list-processing, object-directed language Lisp.
However, Tomita et al refer to the construction of Prolog-likeelements in their Lisp
code.
In the present case,Prolog was chosen because it is a list-processing ,objectdirected language like Lisp but with the additional built-in fmt-order, W c a t e calculus features. Prolog has proved to be a most useful language for representing
the process system states and the knowledge base (i.e. facts plus rules) required for
planning the operating command sequences encountered in process plant operation.
In addition, it is considered that Prolog may provide a sound basis for the extension
of the work to the real-time control domain.
Details of Example and Representation of the Physical Plant
The conventional engineering flowsheet (sometimes known as an engineering
l i e diagram) remains the prime basis for the schematic representation of the vessels
and associated pipework in a chemical processing system. The example to be
consideredis from Rivas and Rudd (1974).
87
J.R. Roach, B.K. 0'Neil1and P.J. Strimairis
Within the context of such flowsheets, a key concept is the notion of a
"connector" for a process zone bounded by material flows. " ... a connector is
defined as a region between nodes (i.e. branches) where connectors join and may
include large amount of equipment or merely short sections of pipe. Connectors
which are capable of being closed have vabe symbols &ed."
OShima (1978) redefined "COM~C~OIS"
as the smallest isolatable sections of
plant (i.e. isolatable by valves). Whilst retaining this concept in this work, it is
renamed "a containment zone" with the boundaries fmed by the notional centres of
the operable valves. The concept provides a natural method for Visualizing the
movement and containment of materials in process plants, particularly when
considering the constraints as previously described as they affect the movement of
materials. Furthermore, an additional property of the 'connector' is retained and all
'containment zones' are conservatively assumed as "perfectly mixed". Figure 1
illustrates the concept.
The reactor example first provided by Rivas and Rudd is an example of a
comprehensive keal-world' problem. It is used again for this work. A detailed
expansion of the reactor section is shown Figure 2. Two further abstractions of the
flowsheet are also included. First, Figure 3 is a representation in the form of a
digraph with the containment zones shown as nodes. Finally, a special form of
digraph which is a type of Petri-net reprmntation (Brauer et al. (1987). Malcolm
(1986)) is included as Figure 4. The bidirectional nature of some links indicates that
material flow is permissible in both directions.
Rivas and Rudd provide an excellent description of the process from which the
following is r e p r o d d
"Hydrocarbon enters the reactors to be upgraded by chemical reacrions, which also
degrade the catalyst. over a 24-hour period, the catalyst in a given reactor is
regenerated by a complex st of operations involving hydrogen, oxygen, natural gas,
inert gas, and carbon tetrachloride. A catalyst -regeneration cycle can involve thirty
or forty changes in valve position, the sequence of which must be designed to skirt
these dangerous conditions:
I . The hydrocarbon inlet must not be connected by an open path to either the
hydrogen or natural-gas inlets.
2. The hydrogen inlet must not be connected by an open path to the natural-gas
inlet.
3. The hydrocarbons, hyhogen and natural gas w t not mix with inert gas, oxygen
or carbon tetrachloride.
4. Theflow of hydrocarbonsand inert gas mut never be blocked.
5. 0.xygen and carbon tetrachloride must neverfiw to the hydrocarbon outlets.
6. Hydrocarbons must never appear in the upper and lower reactor headers.
7. Hydrogen and natural gas must never go into the lower reactor headers."
88
Generating Operating Command Sequencesfor Regulation of Complex Chemical Processes
I
Figure 1: Schematic illustrating the "containmentzone" concept.
Inlet
Vl?
A
J
Cubw TS,
27
Inert Gaa
Inla
Exit
Figure 2: The detail of the reactorjlowsheet
89
J.R. Roach, B.K. O'Neill and P.J. Strimaitis
Representation of Goals and Constraints
Goals: Broadly speaking, sequences of goals are represented as sequences of paths
of containment zones ' f d through the Petri net. (As will be explained below,
'firing' results €tom a valve being permitted to pass composition tokens between the
adjacent zones it separates in the target path, and results in the valve being flagged for
subsequent opening). Flow goals are paths from source to sink zones, evacuation or
draining goals are paths from internal zones to sink zones, and pressurization or
filling goals are paths from source zones to internal zones. As an example, the left
hand reactor is place on-line by the following path:
[hydrocarbonflow source (21) 3reactor (22) 3reactor exhaust (213)l
To purge hydrocarbon from the reactor with hydrogen would requirethe path:
[hydrogenflow source (Z10) reactor(Z2) =+reactorexhaust (213)J
With reference to Figure 2, the actual path for the second of the above illustmtive
goals is:
[hydrogenflow source (210) 3auxiliary header (29) +upper reactor header
(Z4)
=+reactor (Z2) reactorexhaust (213)l
which is an expansion of the antecedent pppl to the
expanded path available.
Each of the =$ symbols in this expanded path represents the required opening of the
intervening valve, viz.the following valve action would be 'fired':
[IVZ open]. IV25 open], W13,open]. Iv7, open11
All the mixing and flow constraints must of course be met for this valve
sequence to be legal. The legality of such paths and the required valve action
sequence are readily determined from the Petri-net representation. Often, only one
single-legal shortest path exists between source and sink and the middle or 'through'
zone in the goal is redundant and can be omitted. Internal pathways can also be
established. (Note: should 'back-tracking' to seek an alternate path be likely to occur
in the Prolog code, then it is advisable to retain the through nodes in the goal).
*
Goal Paths and the Designer's Intent: The function of the operating
command sequence generation is to make the plant carry out the process
specification, not to design the process. Hence, the goals must implicitly express a
predetermined process specification. However, if the designer's intent is inconsistent
with the pre-specified safety constraints as imbedded in the Petri net, then an
automatic command generator should detect this emor, report it and stop. This
overall approach is consistent with that adopted by Rivas et al (1974,1975).
The normal way to summarise the basic process goals is in some form of
process flowsheet. These flowsheets normally contain information such as the
connectivity of process units, and mass and energy balance data involving flow paths
between inlets, outlets and intermediate process vessels. In the case of batch
processes, the description is normally in the form of a sequence of paths, including
paths from sources to holding vessels, from holding vessels to sinks, or from
holding or reaction vessels through functional units (e.g. separators, heat exchangers
and the like), and back to the holding vessel or reactor. Start-up and shutdown
transients, and cyclic or other mode changes for continuous systems may be similarly
represented. This method of goal definition is used in this paper with all closures of
Ifmished' paths handed by default actions.
The representation of flow paths in Petri-net form allows additional process
knowledge about pre-conditions for the establishment of these process paths to be
90
Generating Operating Command Sequences for Regulation of Complex Chemical Processes
included in the model. An example of such pre-condition might be the prohibited
mixing of certain species during the change from one cofligmtion to another.
The sequence of goals may be regarded as a set of consecutive independent paths
to be achieved one by one. However, in some instances, neighbouring goal paths
may have to be held 'open' simultaneously. An automatic generator must be able to
identify such situationsand make appropriate provision. To assist with this task, the
'must flow' paths in the system that must always exist between some sources and
sinks are treated in a special manner. This will be discussed in the following section.
Table 1 lists the process goals from the Rivas et al. (1974,1975) example used
in this study. 'Ihe goals are descn'bed in words and in the 'path' format.
Figure 3: Containment Zone Representation of Flowsheel
The Default Valve State and the 'Must-flow' Constraints: The default
valve state is a most important concept in the method. It embodies the 'must flow'
constraints cited earlier, viz. that the blocking of the flow of certain materials must
never occur. In this sense, it bears some relation to the 'stationay' state concept of
Fusillo and Powers: however, it has other purposes in this work. The default state
for a valve is 'closed' except for certain special valves associated with the 'must flow'
paths which are also designated a priori as 'special' valves. The default state of these
specialvalves is 'opn' and they can only be closed if the current goal creates alternate
flow paths for the materials whose flow is controlled by these special valves. The
flow paths formed by these special valves may be regarded as means of establishing
goals that are simultaneously valid with any other system goal or set of goals. It is
the use of the default closures that enables stopping of flows through valves to be
91
J.R. Roach, B.K. O'Neill and P.J. Strimaitis
handled automatically when the particular flow through those valves is no longer
As a result, there is no need for goal
statementsfor stopping flows unless pressurization or filling is intended.
requid by a goal path or a 'must flow' path.
Table 1: Sequential Goal Seriesfor Operation of the Flowsheet Example
~~
Species Location and Species Mixing Constraints: These are handled by
a token passing system in the Petri-net. For a valve 'transition' to 'fire' compatible
species must exist on each side. Other constraints affecting valve %Iring' are such
that certain species may not be permitted iri certain nodes and that certain nodes may
not contain m m than a certain number of different species.
The Elements of the Working Method
Procedure
- The General
Petri-net Paths: The path goals are taken seriutim and treated as fully
independent. For each goal in turn, the general procedure is as follows: The shortest
legal pathway of valve openings satisfying the goal is 'fired' through the Petri net in
Figure 4. All valves in the system are classified as either:
(a) 'fired' or 'unfired',and
a standard valve (default closed), or a special must-flow valve which must
@)
remain open (default open), or as an exception, a must-flow valve which can
be closed as a consequence of an alternate flow path created by the goal.
(Note: Any 'fired'special must-flow valve is also classed as being in an
' be
alternate path and hence as an exception, and as a result of the ' f ~ n g will
requiredto be open).
Valve sequences for individual goals: The goal paths can now be converted
to valve action lists depending on the current state of the valve position database (i.e.
an action is only specified if the particular valve is not already in that state in the
database, which is a cumulative list of valve actions).
92
Generating Operating Command Sequences for Regulation of Complex Chemical Processes
28
I
a transition
Figure 4: Petri-net Representation of Containment Zones
Criticism of the sequence of valve actions for individual goals: A
preliminary solution is now available and consists of a list of the valve actions for
every goal in the total goal list. The initial solution generated by this procedure
contains redundancies (including unneceswy default closures followed by openings in
the next goal) which must be removed as the goals have been achieved independently
apart h m the state of the valve position database. In addition, certain modifications
may be required in cases where adjacent sets of goals should hold simultaneously.
These are detected and joined into a single list and the redundanciesrechecked.
The Firing of a Valve Path through the Petri Net
The breadth-first search technique: Various techniques have been developed
for searching for feasible paths in graphs. The one of particular interest, the breadthfirst search is well described by Clocksin and Mellish (1987). and Bratko (1986). To
find the shortest path through a graph like Figure 4, a legal set of arcs must be
traversed such that the search extends the shorter paths before the longer paths are
considered. The search terminates when the goal is reached and hence the first path
found must be one of the set of shortest legal paths (i.e. shortest in terms of 'node
93
J.R. Roach, B.K. O'Neill and P.J. Strimaitis
length'). In the context of the sequence of valve actions, this shortest path has the
least number of valve actions and results in the least 'zone contamination'.
Normally, the generation of cycles is totally blocked in a breadth-first search.
However, in hopefully rare special cases, involving evacuation or draining, it is
useful to permit cycling in this method to a Limited extent. The essential use of
these cycles occurs during 'backtracking'in. the Prolog if the scheduled evacuation of
draining goals have been insufficiently specified. The cycles are formed when arcs in
the Petri-net permit flow in two directions through a valve. When the cycles are
'unravelled' they form a more deeply penetrating evacuation or drainage path on each
'backtrack'. Hence, an additional contiguous node is included in the evacuation or
drainage goal each time that 'backtracking'is triggered
Searching the Petri-net for the shortest path: The path goals for the
breadth-first search are supplied in the form of a start node and a goal node and in
addition a 'go-through' node if necessary. Table 1 illushates the path goals for the
example problem and several include 'go-through' nodes, e.g. path goal 6,Z7 22
=$a.
In this case, the breadth-first search technique of Figure 4 finds the shortest
legal node path from 22 to 26. Far a path to be legal, each associated 'transition'
must be able to 'lire'. For a 'transition' to 'fire',all the relevant constraints @reconditions) must be satisfied. Hence, when a shortest legal node path is found, the
search is terminated and the node path specification is returned (e.g. for path goal 6,
the shortest legal node path was [Z7,zA,Z2,Z5, 261). This would require the
following valve or 'transition' 'frring' sequence W20, V13, V17, V221. The more
direct route via V23 is not admissible as it does not pass through node 22.
*
Converting a 'Fired' Path to a Valve Action Sequence:
This is done according to the folowing heuristic rule set. The heuristic rules for
generation of the actual output action are (in execution order):
open by default the must-flow valves which are not exceptions.
(note: 'fired'special valves are classed as exceptions).
2. open special must-flow valves which are exceptions but which are 'fired' (and open
all 'fzed'standard valves since the goal has generated exceptions*).
3 . close by default 'unfired' standard valves.
4. open the 'fired'standard valves (if the goal has not generated exceptions*).
5 . close by default the 'unfired' special valve exceptions.
1.
NOIE: 'Ibis option &a not occur in the cxuaplc dven but would have done so had them been three or mote
radars.
The underlying philosophy of these heuristics is that the valve operations will
not intempt the 'must-flow' paths and that unscheduled flowpaths will not be created.
Following the determination of the valve sequence for a goal, the component
inventories of the containment zones are updated in the database and the individual
goal is complete.
As can be seen from Table 2, the resulting action sequence in this case is "17,
01, [22,0], 123, c]]. Fired valves V13 and V17 do not require opening as they are
already open and V23 must be closed last to maintain the 'must-flow' status of the
inert gas. (Note: that V23 has been classified as an 'exception' as the path goal
creates a new sourcesink path for the inert gas).
94
Generating Operaring Command Sequences for Regulation of Complex Chemical Processes
Table 2: Uncriticized Goal Achievement List of Valve States and Actionsfor
95
J. R. Roach, B.K. 0‘Neil1and P.J. Strimaitis
Criticism of the Raw Sequence of Valve Actions
Removal of redundant valve actions: As foreshadowed earlier, the raw
sequenceof valve actions contains a number of redundancies in the form of
unnecessary close/open occurrences associated with standard valves in the
achievement of consecutive goals. Similar redundancies occur with the reveme
situation, viz.opedclose occurrences of ‘special‘valves.
Fusion of adjacent goal operations that can occur simultaneously:
Aftex the initialremoval of redundancies, the solution of adjacent goals with common
sinks and compatible sources are ‘fused‘. In the example presented, this applies to
goals 8a,8b and to goals 9% 9b. Consider goal solutions 8a and 8b after the initial
removal of the redundant standard valve operations [ZO, c] in 8a followed by [20,01
in 8b, and the redundant ‘special‘valve operations [23,0] in 8a followed by [23,c] in
8b:
*
8a 212
22 3 zd
“61 011
8 b 2 8 + 2 2 = ~ 2 6
“6, cl, [5,01l
Since the two goals have common sink nodes and compatible source nodes, the
valve sequences joined to fonn a single sequence goal f& goal 8:
8 “6.01. [6. cl, [5,01l
A sim&~ocesS forgod 9yielcitx.
9 “23.01, [22,cl, U7, cl, 120%cl, [5. cl, [a, 011
Now fused goal solutions 8 and 9 are adjacent and a re-run of the redundancy
elimination process eliminates [6, c] from goal 8 and [6,0] from goal 9 yielding:
8 “6,oI. [5,01l
9 “23,01, 122, cl, [17, cl, PO, cl, [5, cll
These are the only goal solution ‘fusions‘ in this example that have any
consequencesfor redundancy elimination. Goal solutions 16a and 16b may be ‘fused‘
but contain no redundancies. Different systems may require different or additional
criteria for goal ‘fusion’. The overall results of the criticism process are shown in
Table 3.
Results Compared
The process synthesizer (coded in Prolog) described herein achieves the required
process goals with a legal sequence of 34 valve operations as set out in Table 3.
This may be contrasted with the results of Rivas et al who present two lists of valve
action sequences to meet the specified process goals. The fmt is by an experienced
engineer and the second the result of their computer pogram.
The engineer’s schedule at 35 operations is one more than the Prolog synthesizer
requixed and in fact has one missing operation and should be 36 actions. ’Ihe missing
action is ‘closevalve 17’after ‘open valve 23’ and before ‘close valve 22’ in the goal
to pressurize the reactor with inert gas, chlorine and oxygen (unless valve 17 is
closed the final and initial valve and inventory states are not the same as for the
cyclic process). However, as mentioned below, the Prolog synthesizer had one
containmentzone (zone 5 ) with a different composition to its initial evacuated state.
The computer solution proposed by Rivas et al. is not satisfactory. They
suggest a schedule of 43 valve actions with some glaring faults. Valves 5 , 8 and 22
are opened, and never closed again, and the system is not returned to its initial valve
state. Although leaving valves 8 and 22 open would not affect the final species state,
96
Generating Operating Command Sequencesfor Regulation of Complex Chemical Processes
leaving valve 5 open definitely would. By not closing valve 5, inert gas and oxygen
flow continuously, not just inert gas as required.
Table 3: Production of the Final Action List
PllsQ NWe pain
redundancies)
naw namn uo(
(Posr-criticism: to eliminate
(redundant elements flagged < > a [
I. I after goal fusion))
a
PIMI Acllon U8i
(redundancies removed.
Goals 3.5, Q, 15 are
Treatment of Cyclic Systems
Like any other system, cyclic systems go through transients on startup and
shutdown. In such transient situations, a process simulation as described in this
work may q u i r e several cycles to achieve a 'steady-state sequence' from a given set
of zone inventory conditions and valve states. This 'steady-state sequence' is defined
by a continuously reproducible set of valve sequences which achieve the stated goals
and produce a continuously reproducible set of zone inventories associated with and
following each goal. Many factors are involved, and the extent of the various
purging operations carried out during the cycle with respect to the containment zones
PWgd
Hence when dealing with cyclic systems, the ultimate computer program must
keep track of these inventory changes so that the system inventory changes as well as
valve sequences can be tracked until an acceptable true 'steady-state sequence ' is
reached. The problem is even more complex when attempts are made to produce
'optimum' cycles based on the minimum number of valve actions and perhaps
including some other appropriate criteria sich as a predefined set of zone inventory
states at the beginning and end of each cycle. The computer program described in
this work only partially addresses this problem by means of the cyclic variation of
the breadth-first search algorithm. When backtracking is requested to find a new
sequence, the 'depth' of a purge is automaticallyincreased to modify the containment
zone inventories. However, it is inefficient to rely too heavily on backtracking,
particularly over a m g e of many goals.
97
J.R. Roach, B.K. 0 'Neil1and P.J. Strimaitis
In this work, the final action list Shawn in Table 3 achieves the same valve
sequence as the initial state but there is a differencein the contents of zone 5, viz. the
initialcontents of zone 5 was specifred as evacuated while the final contents axe inert
gas, oxygen and carbon tetrachloride. If the initial contents were assumed to be the
latter combination,a perfect cyclic action sequence would have been achieved in 34
valve actions. more work is required in the area of cycle amlysis.
While the state achieved is perfectly legal' according to therules used,there may
have been other reasons for requiring the empty header. Currently, this can only be
achieved by 'manual' inmention, and specifidly modifying particular goals to
achieve this end, and re-rUnning the pmgram. Clearly, m m work is required in the
area of cycle analysis.
Conclusions
The overall process operating strategy can be broken down into a Series of individual
process goals as exemplified by Table 1. Each of these individual goals can be
communicated to the computer algorithm in the form of a partially specified process
path with begin and end nodes identified. In some instances, 'go--ugh'
nodes must
also be specified. By means of the 'breadth-fmt'search technique, the above goals
can be achieved independently, if solutions exist, by completion of the path
specificationsin tum in such a way that the resultant independent paths am safe and
efficient. The pathways identifed can be converted to appropriatesequences of valve
openings and closures such that individual process goals can be met independently.
Because of the independentstatus accorded the individual goals, redundant operations
are generated. These redundancies are removed by a process of 'criticism' which
includes amalgamation of adjacent goals where this is possible. The final outcome is
an efficient integrated sequence of valve actions which meets the overall process
objective subject to the imposed constraints.
The Prolog computer code produced was capable of solving the example problem in
an interactive way with no input required from the user other than the initial partial
node paths. However, continued development has revealed the need to m w the
method used in the handling of displacement flows of certain species. Although no
errors were generated in the example studied, the changes axe being made before
applying the program to a different type of problem. The work is continuing,
including the development of a graphics interface. In addition, some further
refinementof the system to handle cyclic systems more precisely is required.
References
Bratko, I (1986)" Prolog Programming for ArrifIcial Intelligence" Addison-Wesley,
Chapter 11.
Brauer, W. et al (Eds) (1987) "Petri Nets: Applications and Relationship to Other Models
of Concurrency" Lecture Notes in Computer Science, Springer-Verlag.
Bunn, AX.and Lees,F9. (1988) "ExpertDesign of Plant Handling Hazardous Materials
Design Expertise and Computer Aided Methods with Illustrative Examples" Chem. Eng.
Res. Des. 66, 419-445.
Clocksin W.F. and MeUish, C.S. (1987)" Programming in Probg" SpringerVerlag.Chapters 7-9.
Foukes, N, M.J. Walton ,P.K.Andow and Galluuo M. (1988) "Computer-Aided
Synthesis of Complex Pump and Valve 0perations"Compur. Chem. Eng..l2. 1035-1048.
Fusillo, R.H. and Powers, G.J. (1987) "A Synthesis Method for Chemical Plant Operating
ProcedurestTomput.Chem. Eng., 11, 369-382.
-
98
Generating Operating Command Sequencesfor Regulation of Complex Chemical Processes
Fusillo, R.H. and Powers, GJ. (1988) "Operating Procedure Synthesis using Local Models
and Distributed Goals"Comput. Chem. Eng.. 12, 1023-1034.
Lakshmanan, R and Stephanopoulos G. (1988a) "Synthesis of Operating Procedures for
Complete Chemical Plants - I. Hierarchical. Structured Modelling for Nonlinear
PlanningTomput. Chem. Eng., 12. 985-1002.
Lakshmanan, R and Stephanopoulos G. (1988b) "Synthesis of Operating Procedures for
Complete Chemical Plants - XI. A Nonlinear Planning Methodo1ogy"Compur. Chem.
Eng.. 12, 1003-1021.
Lakshmanan, R and Stephanopoulos G. (1990) "Synthesis of Operating Procedures for
Complete Chemical Plants ID.Planning in the Presence of Qualitative Mixing
Constraints"Compvt. Chem. Eng.. 14, 301-321.
Malcolm, C.A. (1986) ''Petri Nets for Representing Assembly Plans" (SERC/ACME Grant
GRJD/2700.6), DAI Working Paper No. 187. Department of Artificial Intelligence.
OShima, E. (1978) "Safety Supervision of Valve 0perations"J. Chem. Jqm, 11(5), 390395.
Rivas J.R. and Rudd, D.F.(1975) "Man-Machine Synthesis of Disaster-Resistant
0perations"Operations Res., 23-29, 2.
Rivas, JR. and Rudd,D.F.(1974) "Synthesis of Failure-Safe Operations'lAIChE J., 20,
320-325.
Rivas, J.R., Rudd, DF.and Kelly, L.R.(1974) "Computer-Aided Safety interlock
Systems"AIChE J.. 20, 311-319.
Sacerdoti, ED. (1977) "A Structure for Plans and Behaviour", Artificial Intelligence
Series, Elsevier.
Tomita, S, Nagata, M..O'Shima E. and McGreavy, C. (1986) "On the Development of an
Automatic Synthesizer of Operating Procedures for Chemical Plants'' IFAC Conference,
Kyoto/Japan Sept/Oct, pp. 66.
-
Received: 13 August 1995: Accepted afrer revision: 13 February 1996.
99
Документ
Категория
Без категории
Просмотров
1
Размер файла
927 Кб
Теги
using, complex, chemical, sequence, generation, operating, commands, processes, regulation, intelligent, techniques, artificial
1/--страниц
Пожаловаться на содержимое документа