; MODELING TECHNIQUES TO SUPPORT ABNORMAL SITUATION MANAGEMENT
Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out
Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

MODELING TECHNIQUES TO SUPPORT ABNORMAL SITUATION MANAGEMENT

VIEWS: 3 PAGES: 10

  • pg 1
									In Proceedings of the Symposium on Industrial Engineering and Management (pp. 249-256). Toronto: Canadian Society for Mechanical Engineering.

Modeling Techniques to Support Abnormal Situation Management in the Petrochemical Processing Industry
Greg A. Jamieson and Kim J. Vicente
Cognitive Engineering Laboratory Department of Mechanical and Industrial Engineering University of Toronto

ABSTRACT We describe the use of the Abstraction Hierarchy framework to model a petrochemical system. The framework provides a description of the physical and functional relationships within the plant which reveal opportunities for operator action in both normal and abnormal situations. We demonstrate that the Abstraction Hierarchy can be meaningfully applied to petrochemical systems as it has been in other domains and we discuss some implications for future research. 1. INTRODUCTION The inability to effectively manage abnormal situations exacts an annual $20 billion toll from the U.S. petrochemical industry [1]. This occurs despite continual technological developments in advanced control systems and training of operational personnel. Many of these developments have been based on an implicit assumption: that the range of representative events considered by designers are sufficient to cover all possible contingencies. However, experience with large industrial systems inevitably leads to the conclusion that unanticipated events will occur regardless of the extent of engineering analysis and planning. Further, it is that very unanticipated variability which represents the greatest threat to plant safety and productivity [2]. How does one model a plant so that the representation is both useful and meaningful to operators forced to contend with unanticipated events? The Abstraction Hierarchy Representation The Abstraction Hierarchy [3] is a multi-level representation of the structure of a plant. Although the number and nature of levels are not fixed, in the process control domain we have found it useful to include models of production and safety goals, first principles, general functions, plant equipment, and equipment location and appearance. Each level of the Abstraction Hierarchy constitutes a complete system model and is distinguished by a specific language employed at that level.

Individual levels of the Abstraction Hierarchy (AH) are related to adjacent levels by a means/ends relationship. When transitioning between levels, an operator can exploit these relationships to ask three crucial questions (see Figure 1). By entering any level of abstraction operators are implicitly asking themselves the question, “What?” More specifically, “What is the form, function, or purpose that I am interested in?” When traversing up the levels of abstraction, the operator can ask the question “Why?”. That is, “Why does the structural description of the plant include equipment X or function Y at this level?” At the adjacent level above, the operator should find the answer to that question in a more abstract function or purpose. Similarly, when stepping down to lower levels of abstraction, the operator can ask the question, “How?” For example, “How can function Y or purpose Z be realised?” The adjacent level below should denote the functions or equipment that are pertinent to answering such questions. The WHY:WHAT:HOW questions form a window that can be vertically translated through the levels of abstraction. That which constitutes a WHY in one window can serve as the WHAT if the frame is moved up one level of abstraction.
WHY?

Purpose

WHY? Z1

Purpose

WHAT? Z1

Function

WHAT? Y1

Y2

Function

HOW? Y1

Y2

Equipment

HOW? X1

X2

Equipment

X1

X2

Figure 1: The shifting WHY:WHAT:HOW window in the Abstraction Hierarchy. The AH can be complemented by a second dimension to describe the physical aggregation of components at various levels of resolution. However, the nature of the relationship described along the aggregation dimension is conceptually distinct from the means/ends relationship described along the abstraction dimension. The various levels of aggregation are ordered by a part/whole

relationship. For example, a casing, impeller and a motor might be aggregated to form a pump. The individual components can be treated separately or they can be treated at a lower level of physical resolution (i.e., higher aggregation) as a pump. These orthogonal dimensions of abstraction and aggregation define an area over which a number of different plant models are described (see Figure 2). Although each of these models reveals a unique set of information about the modeled system, we emphasize that each is a complete model within its particular cell in the full abstraction/aggregation area. Navigation through this area facilitates an understanding of both the physical and functional relationships between plant elements. Finally, it is often helpful to specify what an abstraction hierarchy is not. The AH is not a specification of events, situations, or plant states. These concepts are temporally restricted whereas the structural representation provided by the AH is relatively invariant over time. That is to say, the AH is event-independent [4]. The model content is not specified via a finite set of abnormal events which are anticipated by designers. Further, the AH does not specify operator tasks or goals. While such a specification can be useful, tasks and goals can vary while structure remains constant. The McFarlane et al. FCCU Model The Abstraction Hierarchy has been employed previously to model a simulated thermal-hydraulic process [5], a simulation based on an existing pasteurisation process [6], aircraft engineering systems [7], a power plant feedwater system [8], and conventional [9] and nuclear [10] power production. In order to test the feasibility of employing the AH in the petrochemical domain, we have employed a partial plant simulation described by McFarlane, Reineman, Bartee, and Georgakis [11]. Henceforth we will refer to the simulation as the McFarlane et al. model. This simulation focuses on the reactor/regenerator section of a Fluid Catalytic Cracking Unit (FCCU). Within a refinery, an FCCU breaks down high boiling point input feeds and separates valuable products from waste products. Within the FCCU, the reactor breaks up the hydrocarbon chains by combusting input feed with the help of a catalyst. The regenerator serves to clean the catalyst used in the reactor. The FCCU is the economic heart of a refinery. To a large extent, its successful operation determines whether or not the refinery will be profitable [12]. The critical nature of its employment makes the FCCU an ideal candidate for exploring the techniques described in this report.

The McFarlane et al. model is a highly simplified model of an FCCU. However, it is designed to capture major system dynamics in order to explore various control configurations [11]. The important criterion in this case is the degree to which the model is representative of the complexities of existing FCCUs. The McFarlane et al. model is multivariable, nonlinear, and features strong interactions between sub-systems. Further, it imposes both mechanical and operational constraints that would be found in actual applications [11]. Given these characteristics, the McFarlane et al. model seems to strike a good balance between representativeness and simplicity. A major advantage of the McFarlane et al. [11] model is that a full description of the model equations is provided. The modeling techniques described here can only yield models as precise as the engineering process models on which they are based. Our goal is to transform the algebraic and differential equations that engineers use to describe the plant into a model that is psychologically relevant to operators challenged with controlling these processes [2]. 2. MODELING THE STRUCTURE OF THE FCCU In this section we apply the AH modeling framework to the McFarlane et al. model of the FCCU. In the following section, we highlight some lessons learned from the application of the framework to this novel domain. The McFarlane et al. Abstraction Hierarchy Space Figure 2 provides an overview of the regions in the abstraction/aggregation space which have been defined for the McFarlane et al. FCCU. Eight cells in the space have been found to be useful in describing the plant. The arrow between the Physical Function and Generalized Function levels at the Component level of aggregation indicates that we have constructed a diagram detailing the transition between these cells.

Aggregation
System (S) Sub-system (SS) Unit (U) Component (C)

Functional Purpose (FP)

$$

Abstraction

Abstract Function (AF) Generalized Function (GF)

AF-SS

AF-U

AF-C (mass) AF-C (energy)

Figure 3 provides a representation of the levels of aggregation. It emphasizes the manner in which lower level nodes are aggregated to form higher level nodes. We give the levels of aggregation a cursory treatment in this paper because part/whole hierarchies are common in system descriptions. What is important at this stage is that the reader recognize that the four levels of aggregation cut across the five levels of abstraction (see Figure 2). Thus, at each level of abstraction there are four possible complete descriptions of the plant.
PART - WHOLE (LINKS)

GF-SS

GF-U

GF-C

Physical Function (PFn)

SYSTEM

FCCU

PFn-C

Physical Form (PFo)

SUB-SYSTEM

AIR INPUT SUBSYSTEM

CATALYST REGENERATION SUBSYSTEM

PRODUCT GENERATION SUBSYSTEM

PRODUCT SEPARATION AND OUTPUT SUBSYSTEM

HEAT AND FEED INPUT SUBSYSTEM

Figure 2: The abstraction/aggregation topography for the McFarlane et al. [11] model FCCU. In reviewing Figure 2, the reader might question why only 8 of 20 cells in the AH space have been described. We emphasize that all cells of the AH are valid and potentially useful representations of the plant. However, in constructing an AH we must evaluate which cells are likely to be most useful to operators. Each cell has been evaluated in terms of what information is added/lost when transitioning from one cell to another. When a representation is constructed for a given cell, we evaluate what information is contained therein which is not contained in other cells. If there is little or none, we sacrifice use of that cell. Through this evaluation process we can limit the presence of marginally useful information that might drain the limited cognitive resources of the operator. Experimental observations of problem solving behavior (from which the AH concept was initially formulated) have shown that such behavior typically falls along the diagonal from the Physical Function/Component cell to the Functional Purpose/System cell [3]. Not surprisingly, the evaluation process described in the preceding paragraphs frequently yields a set of representations that fall along the same diagonal. When operators think about the purposes and functions of a plant, they tend to adopt a coarse unit of analysis (e.g., system, subsystem); when operators think about physical properties of a plant they tend to adopt a fine unit of analysis (e.g., component). Levels of Aggregation

UNIT

LIFT AIR COMBUSTION REGENERATOR REACTOR UNIT SUPPLY AIR SUPPLY UNIT UNIT UNIT

MAIN WET GAS FRACTIONATOR OUTPUT UNIT UNIT

FEED HEAT FEED SUPPLY SUPPLY UNIT UNIT

C O M P O N E N T

Figure 3: Levels of Aggregation with emphasis on relationships between levels. The Level of Physical Form The reader will likely notice that no cells at the Physical Form level are defined in Figure 2. This is because there is no physical instantiation of the McFarlane et al. FCCU. The plant is exclusively simulated and thus has no Physical Form to model. In an operational plant this level would be filled with representations of the location and appearance of the physical components. The Level of Physical Function The Physical Function (PFn) representation resembles a traditional piping and instrumentation diagram except that instrumentation is not specified by the AH (see Figure 4). Also absent from this representation are the various controllers employed in the FCCU model. In constructing an AH, we pay particular attention to restricting our descriptions to the means/ends structure of the plant elements. While control systems are crucial to the successful operation of a modern petrochemical plant, they do not lend themselves to characterisation by means/ends descriptions. In our work, we employ a different framework (not described in this paper) to model the behaviour of control systems. The lines connecting the nodes at the PFn level represent physical relationships between the

structural components. For example, the two Ubend lines connecting the reactor and regenerator imply that there is a direct physical connection between these units. While this statement may strike the reader as being obvious, higher levels of abstraction do not follow this rule. The reader should be careful to understand the meaning of connections between nodes at each level. The Level of Generalized Function The level of Generalized Function (GF) reveals information about heat transfers and flows of commodities (e.g., catalyst, feed, products). We have found it useful to discuss chemical reactions at this level also. The primary reason for including reactions at the GF level is that the language typically employed here lends itself well to discussing reacting commodities. Further, chemical reactions are equally subject to the mass and energy first principles that are described at the Abstract Function level (see below). In other words, we can talk about chemical reactions at the GF level and emphasize which commodities are reacting, in what proportions, with certain products and heat transfers. At the Abstract Function level we can talk about relevant mass and energy relations. Thus, the same chemical reaction can be treated at two levels of

abstraction, each complete yet unique in the type of information it provides. Figure 5 details the transition between the cells of the PFn and GF levels at the Component level of aggregation. It is the manifestation of the arrow between the cells visible in Figure 2. Such a representation is not typical of AHs and was initially constructed as a memory aid for the authors. In review, however, we realised that it was a valuable explanatory tool and we have continued to find it useful. This representation provides a very detailed explanation of how each node at the PFn level is related to its associated end(s) at the GF level. Conversely, each node at the GF level is connected to its mean(s) at the PFn level. Thus, a given node can have a single or multiple means and ends, emphasizing the homomorphic nature of the AH representation. There are a couple of noteworthy features in Figure 5. One is the empty node at the PFn level which is connected to the Gas Oil Flow node at the GF level. This flow is different from other flows in this simulation in that it has no regulating valve. Our conversations with process engineers indicate that it is not uncommon for flowrates in FCCUs to be determined by upstream processes over which the FCCU operators have no control. This lack of
V13

V12 Disengaging Cyclones V11 Main Fractionator V9 Lif t Pipe Wet Gas Compressor

Cyclones V8

V14

Lif t air blow er

Standpipe

Steam Stripper

Riser Furnace V6 V7 V3 V1

Combustion air blo w er

U-bends

Fuel Gas

V5

V2 V4

Figure 4: The Physical Function level at the Component level of aggregation.

Wash Oi l Fl ow, F1

Di esel Fl ow, F2

Fresh Fee d Fl ow, F3

Gas Oil Fl ow, Fgo

Sl urry Re cycle Fl ow, F4

He at Transfer 3

Fu el g as Fl ow, F5

Co mb usti on 1 , He at Transfer 1 , F5Hfu/l m UA fT l m/l m

Exhau st, Ql oss

Ai r Su ncti on Fl ow, FV6

Su ctio n Pressu re , P1

Co mb . Ai r Th rou ghpu t, F6

Di sch arge Press., P2

V1 , C1

V2 , C2

V3 , C3

V4 , C4

V5 , C5

Fu rn ace , T3

V6 , C6

Co mb usti on Ai r Bl ower

Co mb .V ent Fl ow, FV7

Atmosphe ri c Press, P atm

Li ft Ai r Th rou ghpu t, F8

Di sch arge Press., P3

Li ft Ven t Fl ow, FV8

Sp il l Ai r Fl ow, F10

Sl urry/WG He at Transfer Se paratio n

Fracti on ator Press., P 5

Fl are Gas Fl ow, FV1 2

Wet Gas Fl ow, FV1 1

Su ctio n Presure, P7

Wet Gas Co mp re ssor Fl ow, F11

Di sch arge Press., Pvru

V7 , C7

Li ft Ai r Bl owe r, sa

V8 , C8

V9 , C9

Fracti on ator

V1 2, C12

V1 1, C11

Wet Gas Co mp re ssor

Wet Gas An ti -S urge Fl ow, FV1 3

Sp ent Ca tal yst Fl ow, Fsc, Qsc

Re gene ra te d Ca tal yst Fl ow, Frgc, Qrgc

Ca tal yst/WG Se paratio n, F wg

He at Cracki ng Transfer, Re acti on, Qff, Qslu rry , Qcracki ng Qsr, Qfr

Co ke Produ cti on, F co ke

Pressu re a t Ri se r Bo ttom, Prb

Ca tal yst Inve ntory, Wr

Re actor Press., P4

Ca tal yst/ He at Hydroca rb on Transfer 3 Se paratio n

V1 3, C13

Lo wer U-ben d (sl i de va lve)

Up per U-ben d (sl i de va lve)

Di se nga gi ng Cycl one s

Ri se r, Tr

Re actor

Stripp er

Re gene ra te d Ca tal yst Re servo ir, W sp

Ca tal yst Outflo w, Fsp

Re gene ra to r Pressu re , P 6

Co mb usti on 2, QC

Ca tal yst Re servo ir, W reg

Re gene ra to r Bo ttom Pressu re , P rgb

Fl ui di za ti on

Stack gas Fl ow, Fsg, Qfg, Qe

Sp ent Ca tal yst Fl ow, ca t,l ift

Li ft Ai r Fl ow, ai r,l i ft, ca t,l ift, Qai r, QH

Pressu re at bottom of li ft pi pe , Pbl p

Ca tal yst / Stack Gas Sepa ra ti on

Stand Pi pe, l sp

Di se nga gi ng Se ctio n

Fl ui di ze d Be d, Treg

V1 4, C14

Li ft Pi pe

Cycl one s, Tcyc

Figure 5: Detail of transition from the Physical Function to Generalized Function level.

opportunity for action is clearly reflected in the AH. In a fault management situation involving
this flow, a well designed interface should make it clear to the operator that he has no capability to affect this flow. A second point of interest is that many nodes at the PFn level have multiple ends.
Ca tal yst/ Hydroca rb on Se paratio n

Changes in the state of the equipment will lead to multiple changes in flows and heat transfers. The GF-Component cell representation is shown in Figure 6. The nodes at this level represent general functions of the plant, e.g. flows, heat transfers. The connections between these nodes represent causal relationships. Note that
Fl are Fl ow, FV1 2 Fracti on ator Press., P 5 WG Anti -surge Fl ow, FV1 3 Su ctio n Press., P 7 Di scharge Press., P vru

He at Tran sfer 3

Re actor Ca tal yst Inve ntory, Wr

Ca tal yst/WG Se paratio n, F wg

Sl urry/WG Se paratio n

Wet Gas, Fl ow, FV1 1

Wet Gas Co mp re ssor Fl ow, F11

Co ke Fl ow, Fco ke

Press. at Bo ttom of Li ft Pi pe, P p bl Li ft Ai r Bl owe r Di scharge Press., P 3 Li ft Ai r Fl ow, F9 , ai r,l i ft, ca t,l ift Qai r, , QH Re actor Press., P 4

Produ ct Outflo w Sl urry Fl ow, F4 & He at Tran sfer 3 Wash Oi l Fl ow, F1

Cracki ng Re acti on, Qcracki ng

Li ft Ai r Th rou ghpu t, F8

Sp ent Cat. Fl ow, F sc, Qsc

Li ft Ven t Fl ow, FV8

Co mb usti on 2, QC

Press. at bo ttom of reacto r riser, Prb

He at Tran sfer 2 (la te nt , Q , Qfr, sr an d sen si bl e, Qff, Qslu rry )

Fresh Fee d Fl ow, F3 He at Tran sfer 1 , UA fT l m/l m

Di esel Flo w, F2

Gas Oil Fl ow, Fgo

Re gen. ca t. Fl ow, Fsp Frgc , Qrgc Ca tal yst Storage 2 , Wsp , Wreg

Sp il l Ai r Fl ow, F10

Co mb usti on 1 , F5Hfu/l m

Exhau st, Ql oss

Fl ui di za ti on Press. at Bo ttom of Re gen., Prgb Cycl one Fl ow

Atmosphe ri c Pressu re , P atm

Ai r Su ctio n Fl ow, FV6

Re gene ra to r Press., P 6

Fu el Ga s Fl ow, F5

Co mb . Ai r Bl owe r Su cn. Press., P 1

Co mb usti on Ai r Th rou ghp ut, F6

Co mb usti on Ai r Fl ow, F7

Exhau st Fl ow, Fsg, Qfg, Qe

Co mb usti on Ve nt Fl ow, F V7

Co mb . Ai r Bl owe r Di sch. Press., P 2

Figure 6: The Generalized Function level at the Component Level of Aggregation.

causal relationships need not be coupled with physical relationships such as those described by the connections between nodes at the PFn level. The GF-Component cell is quite complicated. The extensive connections between the nodes reflects the high degree of interaction between plant functions. The influence of pressure propagations is a major driver of this inter-dependency. Note that the complexity increases around those nodes related to the reactor and regenerator units. Whereas the functionality around the Feed Input and Heat Transfer Units is essentially sequential (which is typical of AHs we have dealt with previously), the functionality in the reactor and regenerator is circular. The GF level at the Unit and Subsystem levels of aggregation are shown in Figure 7. Note that we employ the same language to describe the functions at these levels. Thus, units and subsystems are discussed in terms of their functions as flows, reactions, and heat transfers. Note how much simpler these representations are compared to the GF-Component level. Moving up a level of aggregation allows the operator to think about the same system in fewer terms, exploiting hierarchy to reduce memory demands. The Level of Abstract Function The Abstract Function (AF) level reveals information about mass and energy relationships in the plant. Connections between nodes at this level again reflect causality. Note that at this level of abstraction the various commodities are no longer distinguishable, they are all represented as masses. Further, different types of heat transfer are treated as energy exchanges.

Generalized Function - Unit Level Catalyst Regeneration, Transport, & Storage Heat Transfer Feed Input Generalized Function - Subsystem Level Air Input Catalyst Regeneration Product Generation Heat and Feed Input Product Separation and Output Product Generating Reaction & Storage Product and Slurry Separation Wet Gas Compression and Transport

Lift Air Flow

Combustion Air Flow

Figure 7: The Generalized Function Level at the Unit and Subsystem levels of Aggregation The representations employed in the AF level are adapted from Multi-level Flow Modeling (MFM) [14]. We should clearly note, however, that we do not adopt all of the MFM rules of syntax. Thus, we do not claim that these representations are examples of MFM. MFM prescribes six types of functions; source, sink, store, balance, transport, and barrier (see Figure 8). A source occurs when mass or energy crosses a system boundary into the system. Similarly, a sink occurs when mass or energy crosses a system boundary away from the system. A store represents a point in the system at which mass or energy can accumulate. A balance describes a conservation of mass or energy without
Source Sink Balance Tran sport Sto re Barrier

Figure 8: The functions of MFM employed in the Abstract Function Level of the AH.

Fe ed Prehe at Uni t Fu el Ga s V5 Fu el Feed Un it Wash Oi l Di esel Oi l V1 Fu rn ace

V2

V3

Gas Oil Fu rn ace Exhau st

To VRU

V1 3

Wet Gas Co mpressor

V4

Cracki ng

WGC

V1 1

Fl are Gas

V1 2

Fracti on ator

Di se nga ger (Cyclo nes)

Stripp in g Steam

Stripp er Re actor

Atmosphe re

Co mbusti on A ir Blo wer

Atmosphe re

Fl ue Ga s V7

Cycl one s

Ca tal yst Re gene ra ti on

Co mb usti on Ai r Bl ower

V6

Re gene ra to r Be d

V9

Steam Exhau st

Tu rbi ne Dri ve Li ft Ai r Bl owe r

Li ft Pi pe Re gene ra to r

Li ft Ai r Bl owe r

Steam

V8

Atmosphe re Atmosphe re

Figure 9: The Abstract Function level for mass relationships at the Component level of aggregation (figure split due to size and resolution limitations).

the use of a store, usually in the form of an exchange. A transport function indicates that mass or energy has been moved from one physical location to another. A barrier is used to indicate a prevention of mass or energy transport. The functions at the AF level most likely to be confused are barrier and balance. A heat exchanger is an example of a common piece of process equipment that serves to exemplify and distinguish between these two functions. In a typical shell and tube heat exchanger, energy is transferred from the hot side flow to the cold side flow. As such, the heat exchanger acts as a balance to energy because the energy is conserved without being stored. In contrast, the two flows never come into contact with each other (a major advantage in nuclear systems). Thus, the heat exchanger acts as a barrier to mass because it prevents physical contact between the commodities. The AF level at the Component level of aggregation for mass representations is shown in Figure 9. The parallel representation for energy relationships has not been included due to space limitations. Figure 10 shows the separate mass and energy portions of the AF level at the Unit level of aggregation. Corresponding representations for the AF level at the Sub-system level of aggregation are not provided.
Mass

installation as a whole. This may appear to be a simplistic statement, but the rest of the AH demonstrates that, in order to achieve this purpose, an extensive range of functions must be properly arrayed. We have experimented with including a couple of other purposes for the FCCU. The most prevalent among these were safety-related. However, out conversations with process engineers have convinced us that these concerns are either coincident with production interests or not worthy of mention. 3. DISCUSSION Prior to this exercise, the Abstraction Hierarchy had not been employed to model petrochemical processes. The results of our efforts shown here indicate that it is indeed feasible to extend the AH to this domain. While the transition has necessitated some modifications and extensions of the AH concepts, it has not posed any insurmountable obstacles or demanded any changes in philosophy. In the following paragraphs, we will discuss some of the peculiarities associated with employing the AH in the petrochemical domain. Dealing with advances in technology. In reviewing the AH described here, two process engineers noted that advances in FCCU technology are manifested at the Physical Function level of Abstraction. In other words, the higher level functions which comprise an FCCU are seldom modified by new technology. This observation has strong implications for employing an AH throughout the life cycle of a plant. Modifications of plant equipment are to be expected, although their actual form cannot be anticipated far in advance. If the immediate effects of those changes are manifested in a single level of abstraction, modifications could be restricted to that level. Information systems and displays could be designed flexibly in areas where changes are likely. Such an approach would alleviate the need to overhaul the AH when slight (but influential) process modifications are introduced. This observation can also be extended to creating AHs for other FCCUs. If differences between multiple plants also lie primarily at the PFn level of abstraction then it is likely that higher levels of abstraction will be relatively consistent between plants. This suggests that once an AH has been created for a full scale FCCU, it can be adapted to other FCCUs with modifications to low levels of abstraction only. If this extension holds then it has strong implications for the flexible application of a particular AH modeling effort.

Energy

Figure 10: The Abstract Function Level at the Unit Level of Aggregation (mass and energy treated separately). The Level of Functional Purpose The overview of the AH space (Figure 2) has a pair of dollar signs in the Functional PurposeSystem cell. The overall purpose of an FCCU is to contribute to the financial viability of the

New developments in AH methodology. Extending the AH to the petrochemical domain has challenged our understanding and appreciation for the modeling technique itself. Two particular principles have evolved from this application. First, we have concluded that not every node in the AH needs to have an associated quantitative variable. Previously we had assumed that each node could be quantified in some manner. Our experience with the FCCU has convinced us that this is not necessary. Qualitative labels can be employed as place holders as long as the distinction is clearly drawn. For example, the cyclones in the reactor and regenerator clearly serve the function of separating commodities (see Figure 6). However, there is no model variable that characterizes this function. Despite our inability to attach a quantitative parameter to this function it is still important that it be spelled out in the plant model. Operators can still use the qualitative concept (i.e., the object) in their reasoning processes, even though there is no quantitative value for it. The second principle is that not all higher level functions must be connected to lower level nodes. In other words, not all ends have means which can be efficiently described at the adjacent lower level. The introduction of pressures to the GF level provides a good case in point (see earlier discussion). In this case, there is frequently no node at the PFn which acts and a means for the pressure. In previous applications of the AH this problem was not encountered. We elected to allow nodes at functional levels to remain unconnected because it would be misleading to suggest to operators that a node at a lower level could be employed to affect a higher level function. In other words, we opted for no information over misleading information. 4. CONCLUSIONS We stated earlier than the greatest threat to plant productivity and safety is the reality of unanticipated variability. Because all events cannot be foreseen by plant designers, event-based responses to abnormal situations must eventually fail to encompass the range of situations operators will face. This statistical fact need not lead to the conclusion that there is nothing that we, as engineers, can do to support operators faced with abnormal plant situations. The Abstraction Hierarchy is a framework which allows a description of opportunities for acting on the plant which is independent of transient states. The model formed by the AH is applicable in all situations, anticipated and unanticipated. Because the

Abstraction Hierarchy is event-independent, it accommodates unanticipated variability by describing the invariant relations [15] of the plant that constrain both operator and control system behaviour.
ACKNOWLEDGEMENTS This research was sponsored by the Honeywell Technology Center (Peter Bullemer, Grant Monitor). REFERENCES [1] P. T. Bullemer, and I. Nimmo, 1994, “Understanding and Supporting Abnormal Situation Management in Industrial Process Control Environments: A New Approach to Training”, In Proceedings of the 1994 IEEE International Conference on Systems, Man, and Cybernetics (pp. 391-396). Piscataway, NJ: IEEE. [2] K .J. Vicente, and J. Rasmussen, 1992, “Ecological Interface Design: Theoretical Foundations”, IEEE Transactions on Systems, Man, and Cybernetics, 22, pp. 589-606. [3] J. Rasmussen, 1985, “The Role of Hierarchical Knowledge Representation in Decision Making and System Management”, IEEE Transactions on Systems, Man, and Cybernetics, SMC-15, pp. 234243. [4] K. J. Vicente, and F. Tanabe, 1993, “EventIndependent Assessment of Operator Information Requirements: Providing Support for Unanticipated Events”, In Proceedings of the Topical Meeting on Nuclear Plant Instrumentation, Control and Manmachine Interface Technologies (pp. 389-393). La Grange park, Illinois: American Nuclear Society. [5] A. M. Bisantz and K. J. Vicente, 1994, “Making the Abstraction Hierarchy Concrete”, International Journal of Human-Computer Studies, 40, pp. 83117. [6] D. V. C. Reising and P. M. Sanderson, 1996, “Work Domain Analysis of a Pasteurization Plant: Using Abstraction Hierarchies To Analyze Sensor Needs”, In Proceedings of the Human Factors and Ergonomics Society 40th Annual Meeting (pp. 293297). Santa Monica: Human Factors Society. [7] N. Dinadis and K. J. Vicente, in press, “Designing Functional Information for Aircraft System Status Displays”, International Journal of Aviation Psychology. [8] N. Dinadis and K. J. Vicente, 1996, “Ecological Interface Design for a Power Plant Feedwater Subsystem”, IEEE Transactions on Systems, Man, and Cybernetics, 43, pp. 266-277. [9] C. M. Burns and K. J. Vicente, 1995, “Physical and Functional Displays in Process Supervision and

Control (CEL 95-11)”. Toronto: Cognitive Engineering Laboratory. [10] J. Itoh, A. Sakuma, and K. Monta, 1995, “Ecological Interface for Supervisory Control of BWR Nuclear Power Plants”, Control Engineering Practice, 3, pp. 231-239. [11] R. C. McFarlane, R. C. Reineman, J. F. Bartee, and C. Georgakis, 1993, “Dynamic Simulator for a Model IV Fluid Catalytic Cracking Unit”, Computers in Chemical Engineering, 17, pp. 275300. [12] N. P. Lieberman, 1991, “Troubleshooting Process Operations”, Tulsa: Penwell. [14] M. Lind, 1994, “Modeling Goals and Functions of Complex Plants”, Applied Artificial Intelligence, 8, pp. 259-283. [15] S. N. Kavuri and V. Venkatasubramanian, 1992, “Combining Pattern Classification and Assumption-based Techniques for Process Fault Diagnosis”, Computers in Chemical Engineering, 16, pp. 299-312.


								
To top