DERIVING BEHAVIOR FROM GOAL STRUCTURE FOR THE

INTELLIGENT CONTROL OF PHYSICAL SYSTEMS

Richard Dapoigny, Patrick Barlatier, Eric Benoit, Laurent Foulloy

LISTIC/ESIA - University of Savoie (France)

BP 806 74016 Annecy cedex

Keywords:

Knowledge-based systems, teleological model, Formal Concept Analysis, Event Calculus.

Abstract:

Given a physical system described by a structural decomposition together with additional constraints, a major

task in Artiﬁcial Intelligence concerns the automatic identiﬁcation of the system behavior. We will show in

the present paper how concepts and techniques from different AI disciplines help solve this task in the case

of the intelligent control of engineering systems. Following generative approaches grounded in Qualitative

Physics, we derive behavioral speciﬁcations from structural and equational information input by the user in the

context of the intelligent control of physical systems. The behavioral speciﬁcations stem from a teleological

representation based on goal structures which are composed of three primitive concepts, i.e. physical entities,

physical roles and actions. An ontological representation of goals extracted from user inputs facilitates both

local and distributed reasoning. The causal reasoning process generates inferences of possible behaviors from

the ontological representation of intended goals. This process relies on an Event Calculus approach. An

application example focussing on the control of an irrigation channel illustrates the behavioral identiﬁcation

process.

1 INTRODUCTION

One of the most interesting and challenging tasks of

Artiﬁcial Intelligence is to derive the behavior of a

system from its components and additional informa-

tion or constraints. Reasoning about physical systems

constitutes an important and active area of research

of Artiﬁcial Intelligence, also known as Qualitative

Reasoning (QR). In QR, most works have focussed

on the representation and composition of models to

describe physical systems either with a component-

based approach (de Kleer and Brown, 1984) or with a

process-based approach (Forbus, 1984; Falkenhainer

and Forbus, 1991). Major areas of investigation are

i) the simulation of physical systems to predict their

behavior ii) given a domain theory, a structural de-

scription of the system and a query about the system’s

behavior, the composition of a model answering the

query.

Another major domain of reasoning which involves

structural modelling and behavioral analysis is the

Software Engineering (SE). A signiﬁcant part of work

in software engineering is dedicated to temporal log-

ics (McDermott, 1982; Manna and Pnueli, 1992; Ma

and Knight, 1996; Galton, 1987; Freksa, 1992) with

extensions for specifying concurrent systems (Bar-

ringer, 1986; Chen and de Giacomo, 1999). These

logics form the basis of behavior analysis relying on

concepts such as goals, actions and event structures.

In this paper, we are concerned with the control of

physical activity by means of software engineering

mechanisms. Let us consider Intelligent Systems in-

teracting with a physical system. It requires at least AI

domains such as QR, for the abstraction of physical

mechanisms and SE, for behavioral analysis of soft-

ware components. This analysis has a great impact

both on the processing of variables related to physi-

cal quantities and on their control. We introduce the

notion of Intelligent Control System (ICS) composed

of a computing unit (e.g., PC, workstation, micro-

controller card, DSP-based system, ...) and sensor(s)

and/or actuator(s) unit(s). Distributed ICS exchange

information through networks using appropriate pro-

tocols (e.g., TCP/IP/Ethernet or dedicated ﬁeld buses

such as CAN, LonWorks, ...). Inside an ICS, two in-

formation ﬂows co-exist, information from/to other

ICS via network ports and information from/to the

physical system via I/O ports. The control of phys-

ical systems with ICS requires reasoning capabilities

extracted both from QR techniques and SE concepts

Dapoigny R., Barlatier P., Benoit E. and Foulloy L. (2005).

DERIVING BEHAVIOR FROM GOAL STRUCTURE FOR THE INTELLIGENT CONTROL OF PHYSICAL SYSTEMS.

In Proceedings of the Second International Conference on Informatics in Control, Automation and Robotics, pages 11-18

DOI: 10.5220/0001162300110018

 SciTePress

such as events and actions. From the outside, an ICS

can be seen as an intelligent device offering a set

of services. Each of theses services are designed to

achieve a given goal, provided that some sequence of

atomic goals is achieved. More precisely, we tackle

the following problem.

Given:

• A scenario description including a physical hier-

archical structure together with a set of physical

equations relating physical variables.

• A modelling theory (derived from the General Sys-

tem Theory) whose instantiation on the given do-

main together with a set of rules will produce a lo-

cal domain theory.

• A goal (i.e., a service) request concerning the local

domain.

Produce:

• during the design step, a goal hierarchy.

• during the design step, an action hierarchy which

traduces the way of achievement of each goal.

• at run-time, the most relevant behavior depending

upon constraints.

This problem concerns major applications such as the

control of industrial processes, automotive systems,

automatic planning for control of physical systems

and measurements, robotics, ... Notice that in the

present model, the structural description of the physi-

cal system may be replaced in unknown environments

by a learning phase based on classical techniques such

as neural networks, genetic algorithms, fuzzy logic,

etc.

2 FOUNDATIONS FOR THE

MODELLING OF CONTROL

SYSTEMS RELATED TO

ENGINEERING PROCESSES

2.1 The structural model

When designing or analyzing a system, the particular

model formalisms that are used depend on the objec-

tives of the modelling. In the engineering domain, the

formalisms commonly adopted are functional, behav-

ioral and structural (Dooley et al., 1998). The struc-

tural representation is an essential component of the

model involving physical systems. Most variables of

the control process are physical variables, that is, they

are an abstraction of the physical mechanism which

is related with each ICS. We consider the semantic

representation of control variables as a tuple includ-

ing the physical role and the physical (i.e., spatial)

entity in which the physical role is evaluated. These

tuples will be referred to as Physical Contexts in the

following. Physical variables are a subset of control

variables and their physical role is in fact the so-called

physical quantity deﬁned in standard ontologies (Gru-

ber and Olsen, 1994). In a ﬁrst step, a part-of hier-

archy of physical entities can be easily sketched. In

a second step, the physical behavior of physical en-

tities is described by expressing the way these enti-

ties interact. The physical interactions are the result

of energetic physical processes that occur in physical

entities. Whatever two entities are able to exchange

energy, they are said to be connected. Therefore, the

mereology is extended with a topology where con-

nections highlight the energy paths between physical

entities. This approach extracts in a local database,

energy paths stretching between ICS in the physical

environment.

2.2 The teleological model of

actions: the goal structure

In the teleological reasoning, the structure and be-

havior of a physical system are related to its goals.

In other words, purposes are ascribed to each com-

ponent of the system and to achieve a global goal,

one must describe how each function of the systems’

parts can be connected. Moreover, since diagno-

sis is an essential part of models describing physi-

cal processes, most works relative to functional rea-

soning in the last decade have incorporated teleolog-

ical knowledge in their model (Lind, 1994; Larsson,

1996; Chittaro et al., 1993). Finally, Qualitative Rea-

soning based on a teleological approach appears to

be a useful component for planning systems involv-

ing the physical world (de Coste, 1994). Therefore,

we adopt the teleological model where goals describe

the purposes of the system, and function(s)

represent

the way of achievement of an intended goal. This ap-

proach is similar to that of some authors (Kitamura

et al., 2002) which claim that base-functions repre-

sent function types from the view point of their goals’

achievement.

The concept of goal is central for behavior analysis

in the control of physical systems. For example, in

failure analysis, when a behavioral change affects one

of the system goals, it means that a failure occurred

(the effect is expressed in terms of the goals that have

not been achieved). Basically, the goal representation

must facilitate the construction of knowledge data-

bases and allows to classify goals and sub-goals rel-

atively to the designers’ intents. The goal modelling

requires i) to describe goal representation (i.e., data

structures), ii) to deﬁne how these concepts are re-

lated.

i.e., computing function

ICINCO 2005 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION

A goal structure must incorporate some possible

actions (at least one) in order to fulﬁll the in-

tended goal (Hertzberg and Thiebaux, 1994; Lif-

schitz, 1993). Representation of intended goals as

”to do action” has been proposed by several re-

searchers (Lind, 1994; et al., 1996; Kmenta et al.,

1999) but neither proposes a formal structure on

which reasoning can be based. Therefore, we extend

that textual deﬁnition by introducing a goal structure

with information relative to the physical system. Two

types of atomic goals are deﬁned, a goal type (uni-

versal) which relate an action verb, a physical quan-

tity

with its arity and an entity type, and a goal token

(particular) by particularizing the physical entity type

item of the goal type. In such a way, the goal mod-

elling deﬁnes the terms that correspond to actions ex-

pressing the intension with the terms that are objects

of the actions. As it incorporates an action verb, the

basic goal deﬁnition proposed here can be seen as a

generalization of the action concept closed to the ac-

tion (or event) types deﬁned in (Galton and Augusto,

2000).

Unlike general framework where goals cannot be for-

malized and relationships among them cannot be se-

mantically captured, the present framework restricted

to engineering physical entities makes it possible to

describe a hierarchical structure (i.e., a mereology) of

goals where the bottom level is composed of atomic

goals. Complex goals can be expressed as a mereo-

logical fusion of atomic sub-goals. One of the major

beneﬁts of mereological framework is that it allows

for different abstraction levels to appear in the same

model.

2.3 The behavioral model

Behavior models play a central role in systems spec-

iﬁcations and control. In order to specify the be-

havior of a system, different approaches are possible.

From the software engineering point of view, a ref-

erence model specifying the behavior of distributed

systems (ISO, 1996) introduces a minimal set of ba-

sic modelling concepts which are behavior, action,

time constraints and states. The popular approaches

to behavioral speciﬁcation languages are based on ei-

ther states or actions (Abadi and Lamport, 1993). In

the state-based approach, the behavior of a system is

viewed as a sequence of states, where each state is an

assignment of values to some set of components. Al-

ternatively, an action-based approach views a behav-

ior as a sequence of actions. Selecting behavior, goals

(i.e., extended actions) time constraints and time-

variant properties (i.e., ﬂuents), we adopt the logical

formalism of the Event Calculus (EC). This formal-

we generalize this deﬁnition with ”physical role” con-

cerning sorts which are not physical

ism presents some advantages well-suited to our pur-

poses, in particular its ability to represent actions with

duration (which is required to describe compound ac-

tions) and to assimilate narrative description of events

with an adjustment of the actions’effects in a dynamic

way.

Deﬁnition 1 A behavior is deﬁned as a collection of

extended actions occurring according to a set of con-

straints, known as pre-conditions.

Therefore, two concepts are highlighted, action types

(goals) and constraints. As the physical system in-

clude artifacts, interaction with the physical system

involves actions with these artifacts. This assertion

reinforces the choice of a goal structure relating an

action with its physical entity. As a consequence, be-

haviors are the result of teleological interpretation of

causal relations among atomic goals.

As suggested in (Barwise and Seligman, 1997), dis-

Figure 1: The local model for Intelligent control.

tribution of knowledge presupposes a system of clas-

siﬁcation. Alternatively, conception and analysis of

knowledge relies on powerful techniques such as For-

mal Concept Analysis. The uniﬁcation of these two

related theories seems a promising candidate to build

the foundations of a theory of distributed conceptual

structures (Kent, 2003). As a consequence each of

the previous sub-models can be related to classiﬁca-

tions through formal contexts as described in ﬁgure 1.

The ﬁrst part produces a goal hierarchy according to

a spatial classiﬁcation where types are goal types and

tokens are spatial instances of goals, i.e., goals related

to a spatial localization. Constraints on goal types

DERIVING BEHAVIOR FROM GOAL STRUCTURE FOR THE INTELLIGENT CONTROL OF PHYSICAL

SYSTEMS

are given through physical equations or/and control

laws relating physical roles. Then, the goal hierarchy

is mapped onto programming functions through ﬂu-

ents constraints. These constraints correspond to the

well-known pre-conditions in STRIPS-like planning.

If several preconditions are deﬁned, then several ways

of achievement exist for a given goal, each of them

corresponding to an event (or action) type. Therefore,

a second classiﬁcation is deﬁned with events as types

and events occurrences as tokens. This classiﬁcation

is a temporal one where the constraints are given by

the Event Calculus formalism through ﬂuents, axioms

and a set of rules. The whole design process begins

with the introduction of a set of universal goals and

produces through a reﬁnement process, a planning for

a given control system in a given environment, i.e.

spatial and temporal instances of general information.

3 THE TARGET APPLICATION

Figure 2: The hydraulic control system with two Intelligent

control nodes.

The real-world example concerns an open-channel

hydraulic system which is controlled with (at least)

two ICS, as shown in ﬁgure 2. The control nodes are

connected with a ﬁeldbus (CAN network). Each ac-

tive ICS

, in the open-channel irrigation channel is

located near a water gate and performs two pressure

measurements from a Pitot tube (resp. in SF Area

and DF Area

). In addition, it is able to react accord-

ingly and to modify the gate position with the help

of a brushless motor. Pairs of goal-program func-

tions are the basic elements on which knowledge rep-

resentation is built. While the basic functions are ex-

tracted from libraries, the goal/subgoal representation

requires a particular attention. To each subgoal, one

or several dedicated software functions can be either

extracted from libraries or deﬁned by the user. Goals

and functioning modes are user-deﬁned. All functions

handle variables whose semantic contents is extracted

from the structural mereology.

4 THE CONCEPTUAL GOAL

HIERARCHY

The Formal Concept Analysis produces a conceptual

hierarchy of the domain by exploring all possible for-

mal concepts for which relationships between prop-

erties and objects hold. The resulting concept lattice,

also known as Galois Lattice, can be considered as a

semantic net providing both a conceptual hierarchy of

objects and a representation of possible implications

between properties. A formal context C is described

by the triple C = (O, A, I), where O is a nonempty

ﬁnite set of objects, A is a nonempty ﬁnite set of at-

tributes and I ⊆ O × A is a binary relation which

holds between objects and attributes. A formal con-

cept (X, Y ) is a pair which belongs to the formal con-

text C if X ⊆ O, Y ⊆ A, X = Y

and Y = X

X and Y are respectively called the extent and the in-

tent of the formal concept (X, Y ). The ordered set

(B(C), ≤) is a complete lattice called the concept lat-

tice of the formal context (C).

Deﬁnition 2 Given R, a ﬁnite set of physical

roles and φ, the ﬁnite set of physical entities,

a Physical Context (PC) is a tuple: θ =

(r, µ(r), ϕ

, ϕ

, ...ϕ

µ(r)

), where r ∈ R, denotes its

physical role (e.g., a physical quantity), µ : R →

Nat, a function assigning to each role its arity (i.e.,

the number of physical entities related to a given role)

and {ϕ

, ...ϕ

µ(r)

} ⊆ φ, a set of entities describing the

spatial locations where the role has to be taken.

Deﬁnition 3 Given Φ the ﬁnite set of physical enti-

ties types, a goal type is a pair (A, Ξ), where A is an

action symbol and Ξ a non-empty set of tuples ξ =

(r, µ(r), φ

, φ

, ...φ

µ(r)

) where {φ

, ...φ

µ(r)

} ⊆ Φ,

a set of entities types describing the spatial locations

where the role r has to be taken.

def

= (a, Ξ) (1)

Deﬁnition 4 A goal token is a pair (A, Θ), where A

is an action symbol and Θ a non-empty set of tuples

θ = (r, µ(r), ϕ

, ϕ

, ...ϕ

µ(r)

def

= (a, Θ) (2)

The hydraulic control system requires the following

list of basic goals

= (to

acquire, {(pressure, 1, SF Area1)})

= (to

acquire, {(pressure, 1, DF Area1)})

= (to

compute, {(velocity, 1, W aterArea1)})

= (to

compute, {(level, 1, W aterArea1)})

these goals are not concepts

ICINCO 2005 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION

= (to send, {(velocity, 1, W aterArea1),

(level, 1, W aterArea1)})

= (to

receive, {(velocity, 1, ExtEntity),

(level, 1, ExtEntity)})

= (to

compute, {(level, 2, W aterArea1,

ExtEntity)

= (to

compute, {(of f set, 1, Gate1)})

= (to

receive, {(of f set, 1, Gate1)})

= (to

move, {(position, 1, Gate1)})

A close connection between FCA and mereology

can be established by focusing on their basic topics,

i.e., concept decomposition-aggregation and concept

relationships. FCA helps to build ontologies as a

learning technique (Cimiano et al., 2004) and we

extend this work by specifying the ontology with

a part-of hierarchy. The goal hierarchy is derived

from the subsumption hierarchy of conceptual scales

where the many-level architecture of conceptual

scales (Stumme, 1999) is extended taking into

consideration the mereological nature of the extents.

Higher level scales which relates scales on a higher

level of abstraction provide information about hier-

archy. Considering the atomic goals, the compound

goals corresponding to the user intents, the ontolog-

ical nature of the extents (i.e., the physical entities)

and some basic assumptions, one can automatically

produce the relevant instrument functional context.

This context is required to produce the ﬁnal concept

lattice from which the functional mereology will be

extracted.

As suggested in (Stumme, 1999), the set of sub-goals

is extended with hierarchical conceptual scales such

as the intent includes atomic and compound goals

and the ICS scale (highest level). Higher level

scales deﬁne a partially ordered set. The formal

context is ﬁlled in a two-stages process. Then, we

derive some rules from the structural mereology S

which concerns the physical entities. To overcome

difﬁculties about the conceptual equivalence between

sets and mereological individuals, we make the

assumption that a mereological structure can be

reproduced within sets provided we exclude the

empty set. Therefore, a set can be seen as an abstract

individual which represents a class

. The part-of

relation can be described as a conceptual scale which

holds between the objects (i.e., extensions) related to

the mereological individuals. The context considers

goal achievement predicates as formal objects, goals

and compound goals as formal attributes. First, the

sub-context between basic goals is derived from

qualitative reasoning on PC tuples within each

physical (or control) equation. Then this sub-context

is extended with conceptual scales corresponding to

goal requests speciﬁed by the user. For the hydraulic

system for example, we plan three services with the

A class is simply one or more individuals

respective goals:

= (to

measure, {(speed, 1, W aterArea1),

(level, 1, W aterArea1)})

= (to

control, {(speed, 1, W aterArea1)})

= (to

manuallyM ove, {(position, 1, Gate1)})

Then, the concept lattice is transformed in a partial

order by some elementary rules. First, for each node

the concept is labelled with the intent of the lattice

node (Cimiano et al., 2003). In a second step, over-

laps are highlighted and the previous ordering is re-

duced based on simpliﬁcation rules (Dapoigny et al.,

2005). In a third step, we reduce the labelling (Ganter

and Wille, 1999), providing that each intent is entered

once in the lattice. Finally, the bottom element is re-

moved. These rules applied on the raw lattice (ﬁg3)

result in the goal hierarchy of ﬁg4.

Figure 3: The goal lattice.

5 BEHAVIOR REPRESENTATION

While the EC supports deductive, inductive and ab-

ductive reasoning, the latter is of particular interest for

our purpose. Given an ontological description based

on possible causal behaviors, dynamical constraints

can be translated in EC axioms. The EC axioms pro-

vide a partial temporal order from ontologies inferred

with mereological logic from user-deﬁned goals and

SP pairs. Moreover, the abductive implementation of

the EC is proved to be sound and complete (Russo

et al., 2001). To solve the frame problem, formulae

are derived from the circumscription of the EC repre-

sentation.

DERIVING BEHAVIOR FROM GOAL STRUCTURE FOR THE INTELLIGENT CONTROL OF PHYSICAL

SYSTEMS

Table 1: Functional context for the open-channel irrigation canal

F g

Achieved(g

) x x x x x x x x x

Achieved(g

) x x x x x x x x

Achieved(g

) x x x x x x x

Achieved(g

) x x x x x x x

Achieved(g

) x x

Achieved(g

) x x x x x

Achieved(g

) x x x x

Achieved(g

) x x x

Achieved(g

) x x x

Achieved(g

) x x x

Figure 4: The goal part-of hierarchy.

Deﬁnition 5 Let G be a goal, let Σ be a domain de-

scription, let ∆

be an initial situation, let Ω be a

conjunction of a pair of uniqueness-of-names axioms

for the actions (goal) and ﬂuents (Achieved(goal))

mentioned in Σ, and let Ψ be a ﬁnite conjunction of

state constraints. A plan for G is a narrative ∆ such

that,

CIRC[Σ; Initiates, T erminates, Releases] ∧

CIRC[∆

∧ ∆; Happens] ∧ Ψ ∧ EC ∧ Ω |= G (3)

In this work, we use the version presented in (Shana-

han, 1997) which consists in a set of time points, a

set of time-variant properties (i.e., ﬂuents) and a set

of event types. Each event type is in fact an action

type which at least requires an action verb, therefore

we associate the extended actions (EA) to each

operational goals (i.e., the triple action verb, physical

role and physical entity). Under the assumption

where a unique computing function is related to a

single goal, domain equations are simple. For more

complex situations, several computing functions

(the actions of EC) can be related to a single goal

provided that a set of ﬂuents (pre-conditions) selects

the relevant association in a given situation. The

circumscriptive condition is consistent if the domain

description does not allow a ﬂuent to be both initiated

and terminated at the same time. EA in the event

calculus are considered as ﬁrst class objects which

can appear as predicates arguments. The conjunction

of Initiate, T erminates and Releases formulae

describe the effects of EA and correspond to the

domain description. The ﬁnite conjunction of state

constraints Ψ expresses indirect effects of potential

actions. These constraints are available implicitly

through the goal mereology since its description is

deduced from qualitative equations where a complex

goal achievement requires physical constraints to

be satisﬁed. Mereological individuals from a given

level and their adjacent lower ones give rise to a

morphism between P art − of relations and state

constraints. From this assumption, state equations

expressing physical and computational causality can

be derived in event calculus (constraints on what

combinations of ﬂuents may hold in the same time).

The uniqueness of EA names, i.e. to

Achieve(goal)

and ﬂuents names, i.e. Achieved(goal) is a logical

consequence of the uniqueness of goal description

in the mereological model. Taking the example of

the complex goal G

, state equations are deﬁned as

follows:

HoldsAt(g10, T ) ← HoldsAt(g8, T ).

HoldsAt(g8, T ) ← HoldsAt(g7, T ).

HoldsAt(g7, T ) ← HoldsAt(g3, T ),

HoldsAt(g4, T ), HoldsAt(g6, T ).

HoldsAt(g4, T ) ← HoldsAt(g1, T ).

ICINCO 2005 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION

HoldsAt(g3, T ) ← HoldsAt(g2, T ),

HoldsAt(g1, T ).

together with domain equations:

Initiates(achieved(g8), g8, T ) ← HoldsAt(g7, T ).

Initiates(achieved(g7), g7, T ) ← HoldsAt(g3, T ),

HoldsAt(g4, T ), HoldsAt(g6, T ).

Initiates(achieved(g4), g4, T ) ← HoldsAt(g1, T )].

Initiates(achieved(g3), g3, T ) ← HoldsAt(g2, T ),

HoldsAt(g1, T )].

Initiates(achieved(g2), g2, T ).

Initiates(achieved(g1), g1, T ).

Initiates(achieved(g6), g6, T ).

Applying abductive logic for theorem proving, we

get the plan ∆:

Happens(achieved(g6), t7, t7),

Happens(achieved(g4), t6, t6),

Happens(achieved(g1), t5, t5),

Happens(achieved(g2), t4, t4),

Happens(achieved(g3), t3, t3),

Happens(achieved(g7), t2, t2),

Happens(achieved(g8), t1, t1).

before(t7, t2), bef ore(t5, t6), before(t6, t2),

before(t5, t3), bef ore(t4, t3), before(t3, t2),

before(t2, t1), bef ore(t1, t).

Figure 5: The temporal hierarchy.

This plan is sketched at ﬁgure 5. Initially an empty

plan is presented with a goal G in the form of a

HoldAt formulae. The resulting plan must respect

the causal hierarchy obtained in section 2.

6 RELATED WORK

Goal modelling is obviously investigated in require-

ments engineering. Modelling goals for engineering

processes is a complex task. In (Rolland et al., 1998)

goals are represented by verbs with parameters, each

of them playing a special role such as target entities

affected by the goal, resources needed for the goal

achievement, etc. Centered on the KAOS method,

(El-Maddah and Maibaum, 2003) used conditional

assignments based on the application’s variables in

goal-oriented process control systems design with the

B method. A tool translates the goal model into B

speciﬁcations where the behavior is state-based. In

this method no reasoning is performed at the system

level due to the lack of semantic content for variables.

For more general frameworks, (Giorgini et al., 2002)

describes a logic of goals based on their relationship

types, but goals are only represented with a label, and

the reasoning is elicited from their relations only.

7 CONCLUSION

This work is part of an ongoing attempt to develop

an automaton able to derive executable program for

the intelligent control of physical systems. From a

user-deﬁned description of the context (i.e., the struc-

tural description of the physical system together with

the control system which operates on it) and an ini-

tial goal request, the system architecture consists of

two layered control modules to provide a response

to this request. The ﬁrst layer extracts a hierarchical

functional model centered on the goal concept. This

model is mapped through ﬂuents on the hierarchy of

action types. The second layer constructs a partial-

order temporal hierarchy relating grounded actions.

Unlike classical AND/OR goal decomposition which

does not clearly distinguish dependency types be-

tween different abstraction levels, the part-of hierar-

chy is able to extract potential relevant dependencies

(such as goal g

in 4). The major beneﬁts of Arti-

ﬁcial Intelligence in this context turns out to reduce

the design process, to generate automatic sound plan-

ning and to allow dynamic control at runtime through

a reasoning process between distributed intelligent

nodes. This last topic (under development) is cen-

tered on information ﬂow methodology and concep-

tual structures.

REFERENCES

(1996). Open Distributed Processing - Basic Reference

model - Part 2: Foundations. ISO/IEC and ITU-T.

Abadi, M. and Lamport, L. (1993). Composing speciﬁca-

tions. ACM Transactions on Programming Languages

and Systems, 15(1):73–132.

Barringer, H. (1986). Using temporal logic in the compo-

sitional speciﬁcation of concurrent systems. Techni-

cal Report UMCS-86-10-1, Dpt of Computer Science

Univ. of Manchester.

Barwise, J. and Seligman, J. (1997). Information ﬂow - The

logic of distributed systems, volume 44 of Cambridge

DERIVING BEHAVIOR FROM GOAL STRUCTURE FOR THE INTELLIGENT CONTROL OF PHYSICAL

SYSTEMS

Tracts in Theoretical Computer Science. Cambridge

University Press.

Chen, X. and de Giacomo, G. (1999). Reasoning about non-

deterministic and concurrent actions: a process alge-

bra approach. Artiﬁcial Intelligence, 107:63–98.

Chittaro, L., Guida, G., Tasso, C., and Toppano, E. (1993).

Functional and teleological knowledge in the mul-

timodelling approach for reasoning about physical

sytems: a case study in diagnosis. IEEE Transactions

on Systems Man and Cybernetics, 23(6):1718–1751.

Cimiano, P., Hotho, A., Stumme, G., and Tane, J. (2004).

Conceptual knowledge processing with formal con-

cept analysis and ontologies. In ICFCA, number 2961

in LNAI, pages 189–207. Springer.

Cimiano, P., Staab, S., and Tane, J. (2003). Deriving con-

cept hierarchies from text by smooth formal concept

analysis. In Procs. of the GI Workshop LLWA.

Dapoigny, R., Barlatier, P., Benoit, E., and Foulloy, L.

(2005). Formal goal generation for intelligent con-

trol systems. In 18th International Conference on In-

dustrial & Engineering Applications of Artiﬁcial In-

telligence & Expert Systems, number 3533 in LNCS.

Springer.

de Coste, D. (1994). Goal-directed qualitative reasoning

with partial states. Technical Report 57, Institute

for the Learning Sciences, Northwestern University

(Evanston).

de Kleer, J. and Brown, J. (1984). A qualitative physics

based on conﬂuences. Artiﬁcial Intelligence, 24:7–83.

Dooley, K., Skilton, P., and Anderson, J. (1998). Process

knowledge bases: Facilitating reasoning through

cause and effect thinking. Human Systems Manage-

ment, 17(4):281–298.

El-Maddah, I. and Maibaum, T. (2003). Goal-oriented re-

quirements analysis for process control systems de-

sign. In Procs. of the International Conference on for-

mal methods and models for co-design, pages 45–46.

IEEE Comp. Society Press.

et al., Y. U. (1996). Supporting conceptual design based on

the function-behavior-state modeler. Artiﬁcial Intel-

ligence for Engineering Design, Analysis and Manu-

facturing, 10(4):275–288.

Falkenhainer, B. and Forbus, K. (1991). Compositional

modeling: ﬁnding the right model for the job. Arti-

ﬁcial Intelligence, 51:95–143.

Forbus, K. (1984). Qualitative process theory. Artiﬁcial

Intelligence, 24:85–168.

Freksa, C. (1992). Temporal reasoning based on semi-

intervals. Artiﬁcial Intelligence, 54:199–227.

Galton, A. (1987). Temporal logics and their applications.

Academic Press.

Galton, A. and Augusto, J. (2000). Two approaches to event

deﬁnition. In Springer, number 2453 in LNCS, pages

547–556.

Ganter, B. and Wille, R. (1999). Formal concept analysis -

mathematical foundations. Springer.

Giorgini, P., Nicchiarelli, E., Mylopoulos, J., and Sebas-

tiani, R. (2002). Reasoning with goal models. In

Procs. of the int. conf. on Conceptual Modeling, num-

ber 2503 in LNCS, pages 167–181. Springer.

Gruber, G. and Olsen, G. (1994). An ontology for engi-

neering mathematics. In Doyle, J., Torasso, P., and

Sandewall, E., editors, Fourth International Confer-

ence on Principles of Knowledge Representation and

Reasoning, pages 258–269. Morgan Kaufmann.

Hertzberg, J. and Thiebaux, S. (1994). Turning an action

formalism into a planner: a case study. Journal of

Logic and Computation, 4:617–654.

Kent, R. (2003). Distributed conceptual structures. In in

Procs. of the 6th Int. Workshop on Relational Meth-

ods in Computer Science, number 2561 in LNCS.

Springer.

Kitamura, Y., Sano, T., Namba, K., and Mizoguchi, R.

(2002). A functional concept ontology and its ap-

plication to automatic identiﬁcation of functional

structures. Artiﬁcial Intelligence in Engineering,

16(2):145–163.

Kmenta, S., Fitch, P., and Ishii, K. (1999). Advanced fail-

ure modes and effects anlysis of complex processes.

In Procs. of the ASME Design Engineering Technical

Conferences, Las Vegas (NE).

Larsson, J. (1996). Diagnosis based on explicit means-end

models. Artiﬁcial Intelligence, 80:29–93.

Lifschitz, V. (1993). A theory of actions. In Procs. of the

tenth International Joint Conference on Artiﬁcial In-

telligence, pages 432–437. Morgan Kaufmann.

Lind, M. (1994). Modeling goals and functions of complex

industrial plant. Journal of Applied Artiﬁcial Intelli-

gence, 8:259–283.

Ma, J. and Knight, B. (1996). A reiﬁed temporal logic. The

Computer Journal, 39(9):800–807.

Manna, Z. and Pnueli, A. (1992). The temporal logic

of reactive and concurrent systems (Speciﬁcation).

Springer-Verlag.

McDermott, D. (1982). A temporal logic for reasoning

about processes and plans. Cognitive Science, 6:101–

155.

Rolland, C., Souveyet, C., and Achour, C. B. (1998). Guid-

ing goal modelling using scenarios. IEEE Transac-

tions on software engineering, pages 1055–1071.

Russo, A., Miller, R., Nuseibeh, B., and Kramer, J. (2001).

An abductive approach for analysing event-based re-

quirements speciﬁcations. Technical Report DoC-

2001/7, Dpt of Computing Imperial College, London.

Shanahan, M. (1997). Solving the frame problem: A math-

ematical investigation of the common sense law of in-

ertia. MIT Press.

Stumme, G. (1999). Hierarchies of conceptual scales.

In Procs. of the Workshop on Knowledge Acquisi-

tion, Modeling and Managenent (KAW’99), volume 2,

pages 78–95.

ICINCO 2005 - INTELLIGENT CONTROL SYSTEMS AND OPTIMIZATION