Nuclear

Sri Hartati Faculty of Computer Science University of New Brunswick P.O. Box 4400 Fredericton, N.B., Canada E3B 5A3

Process

Theory
Bradford G. Nickerson

Faculty of Computer Science University of New Brunswick P.O. Box 4400 Fredericton, N.B., Canada E3B 5A3

Abstract
Nuclear Process Theory (NPT) models nuclear physics processes using a formal grammar. This, in turn, allows one to write computer programs which simulate nuclear process inderactions. Knowledge of the nuclear physics processes is expressed in terms of a basic nuclear physics model and an aggregate eflects model. The basic nuclear physics model has the capability ojexpressing knowledge of nuclear physics processes for dijjerent types of reactions. The products of reactions can be expressed in terms of either deterministic or probabilistic values. The aggregate effects model is used to simulate nuclear physics processes.

Background

and Motivation

Several attempts and approaches have been made to enable symbolic computer representation of physical systems [l, 2, 3, 6, 7, 8, 9, lo]. Forbus and Woods [3,8,9, lo] incorporate the notions of physical processes for describing the behavior of a physical system. They provide a way of specifying processes and their effects in a way that allows both deduction on what processes occur and how they might change. Forbus approach is still a purely qualitative approach. By choosing a purely qualitative approach, many researchers have disqualified their methods from a vast number of application areas in engineering and science. Woods [lo] introduced a combined qualitative and quantitative approach that can have a quantitative interpretation. His approach introduces parametric state space models which can represent parameters that physicists and engineers normally use. Physical system reasoners usually require a representational apparatus that deals with the vast amount of physical knowledge that is used in reasoning tasks. There are two kinds of knowledge required. First is knowledge about properties of different types of com-

ponents of physical systems. Second is knowledge about physical interactions that follow the underlying physical laws. All of the methods have been developed to reason about different types of physical systems. None of these approaches has been developed to reason about nuclear physics processes [4]. This paper introduces a method, called Nuclear Process Theory, for reasoning about nuclear physics processes. NPT formalizes how to structure parts of the available knowledge on nuclear physics processes to enable a computer program, taking a description of the topology of a nuclear physics process as its input, to derive conceivable dynamic interactions in the system. Both a quantitative state space model and a probabilistic space model of the system are incorporated. NPT integrates a symbolic, qualitative description of structural aspects of the process with a quantitative state space and probabilistic space model. Part of NPT s formal description is contained in a context free grammar. Knowledge of nuclear physics processes is expressed in descriptions of basic nuclear physics and aggregate effects models. NPT uses a novel approach to formalize the relationships between mathematical and probabilistic models of nuclear physics processes. NPT can be used to reason about nuclear physics processes such as fission reactions with different types of nuclides.

Modeling

Assumptions

NPT extends the Hybrid Phenomena Theory (HPT), a framework for formalizing dynamic state space models for physical systems [lo]. Besides allowing us to formalize the subset of the assumptions underlying a mathematical model, it also allows us to formalize nondeterministic actions which can possibly occur in a pro cess being modeled, especially for nuclear physics reactions. NPT allows the explicit representation of knowledge of nuclear physics interactions, and formalizes how

340 1043-0989/95$4.0001995 IEEE

Proceedings of the 11th Conference on Artificial Intelligence for Applications (CAIA '95) 1043-0989/95 $10.00 1995 IEEE

Knowledge Level Reaction components, and nuclear interactions. Symbolic Level Description of objects, and instances of Basic Nuclear Physics Model (BNPM) and Aggregates Effect Model (AEM). basicnuclearphysicsmodel reactants {
01 is NUCLIDE 02 is PARTICLE COLLIDE(o1, 02) }

bj is fission (01, 02) {

parametricreaction { def-par nu(v&e: !S!.#) defdbj Fir&e NUCLIDE (value: 01) defdonst threshold (value: 3.15elO) } intermediatestate { neutronabsorption camp-nut (01) } activityconditions { Fiasile.4 > 0 } products { Neutron.amoant is PROBABILISTIC
(Poisson(2./S, 2.43, 0 to 7), pdj-neutr[7] (index neutronproduced, value: decomposition(Poiason(2.@,2.~S,0,7)))) Fl.Al ia PROBABILISTIC ( mass,gieldl(Firsile), def-vector pdj-41 [a] (index Al, value: l&p-table(mas.qieldusyieldl(Fissile)))) F2.A) is PROBABILISTIC ( mass,yield2(Fissik), def-vector pdjJ2 [&] (index AI, value: looLup~able(massyield2(Fissile)))) deltaZ is PROBABILISTIC ( Gaussian(0, 7,-42 to J2), def,vector pdjdellaZ [84] (index d&Z, value: decomposition(Gausrian(O,l,-42J2)) ) EnergyBeutron.amount is PROBABILISTIC prompt-neutrqx&um(Fissik), def-vector pdjl3-neut[7](indes:fieutr, value:lo&+table(neutro~gmztrum(Fiwik)))) Energy-G-. amount is PROBABILISTIC ( gammaspectrum(Fisrile), defmxtor pdjJLgamma[7](indclrEgomma, value:lookup~abIe(gammaspeettarm(Fissile)))) EnergyJ l.amount is PROBABILISTIC ( energydistributionsl(Fisaile), def-vector pdjX-Fl[&j (index Al, value: loo~uptable(energydistributionsl(Fissik)))) EnergyJ 2.amount ir PROBABILISTIC ( (energydirtributionr2(Fihle), def,vector pdjJLF2[44] (indes: A%, u&e: looLup~table(ene~isgydistn butions2(Fi&k)))) 1

def,vector

Figure 1: Model Hierarchy in NPT

these affect the structure and parameters of the model. The effects on the model are encoded in a symbolic structure. NPT has three different levels. These are shown in Fig. 1. Each of them is described separately.

2.1

Knowledge

level

General descriptions of reaction components, materials and nuclear interactions are incorporated in the knowledge level. Knowledge describing the properties of various types of material substances is encoded in class descriptions. Knowledge on nuclear interactions is expressed by definitions in the basic nuclear physics model and the aggregate effects model. 2.1.1 Basic nuclear physics model

A basic nuclear physics model consists of descriptions of the underlying nuclear process being modelled. It contains conditions for when a process will occur, intermediate states in which a process will participate, and the products of a process. It is characterized by five fields: reactants, activity conditions, parametric-equation, intermediate&ate, and products. Examples of such definitions are shown in Fig. 2.

Reactants describe what reagents are being used in the reaction being modelled, and relations between reagents which are necessary for the reaction to occur. In our example, 01 and 02 are the

Figure 2: Basic nuclear physics model for fission.

341

Proceedings of the 11th Conference on Artificial Intelligence for Applications (CAIA '95) 1043-0989/95 $10.00 1995 IEEE

aggregate-effects-model fission (bj, RC, H W, Hw){ material { 01 } equipment { RC is REACTOR, HW is MODERATOR, Hw is CLADDING ) preconditions { RC.power is ON 1 activityconditions{ RC.juelamount > 0 or Rc>O } parametric-reaction { def-const NA (uolue:6.02217eQ3,unit: ) defxonst o6T_deltoZ (value:-20,unit: ) defsonst o&Al (value: 72, unit: ) def_const o&.42 (volue: 117, unit: ) defxonat ei (value: 1, unit:MeV ) defpar f (volue: ol.sf * ol.N/ (ol.sj * o1.N + HW.s,o * Hw.N + HW.s-a * Hw.N) defpar k (value: RC.eta * RC.epsilon * RC.p * j * RC.PNLj * RC.PNLth ) def-var Rc ( unit: fissions ) def-var C (unit: gr / set) def-var pn (value: of.omount) def,vector distr-neutr[9] (index:neutronssroduced) def-vector distr-Ai[&](index:Af) defarray distr,Fl[169,109](xindex:N,yindex:Z) def,vector distr..A2[45] (index: A2) defarray distr-F2[169,109](xhdex:N, yindex:Z) def,vector distr,E,neutr[9](index:energyneut) def,vector distr-Ewgomma[9](index:energy~amma) def-vector diatr-E-Fl[89](index:energyJl) def-vector distr-E-F2[89](index:energy-F2) defarray nuclidesJi[l69,109](~ ndex: neutrons per nuclide,yindex:Z) } defarray nuclidesJ2[169,109](~ ndex: neutrons per nuclide, yindex:Z) } relations ( alginfl ((Rc) (k, nu, pn) (k / nu * pn)} alginfl ((B)(Rc, Fissile. A)(Rc * Fissile. A/NA)) alginfl { (C)(Fissile.s j, Fissile.s-g, B) (((Fisai1e.s.f + Fissile.s-g) /ol.s j) * B)J aggrinfl { (distr-neutr)(Rc) ((Rc * pdj-neutr[i]) i = 0..8 ) ) aggrinfl { (dish-4 l)(Rc) ((Rc * pdj-Al [i]) i = 0..44 ) ) aggrinfl { (distr-A2)(Rc) ((Rc * pdj-A2[i]) i = 0..44 ) } aggrinfl { (distr-E,neutr)(pn,pdj&neutr) ((pn * pdj-&neut [i]) i = 0..8) } cuminfl { (pn)(distrneutr) ( (sum(i * distr-neutr[i])) i = 0..8) ) aggrinfl { (distr-l-gamma) (Rc) ( (Rc * pdj-E-gamma [i/l i = 0..8 ) } distrinfl{ (nuclides-Fl)(distr-A i,pdj-deltoZ, comp,nuc,f,nu)(isotopesdistribution (distr-41, pdj-delta-Z, camp-nuc.Z, camp-nuc.A))

distrinfl { (nuclides_FZ)(distr42,pdj..deltoZ, camp,nut j,nu)(isotopesdistribution (distrJd,pdjdelta-Z, comp,nuc.Z, camp-nuc.A)} aggrinfl { (distr-E-Fl)(Rc) ((Rc * pdj-E-M [i]) i = 0..90) ) aggrinfl { (distr_GF2)(Rc) ((Rc * pdj-E-F2 [i]) i = 0..90) } products { defabj Neutron PARTICLE { (value (amount) : cuminfl { (Neutron.amount) (pn) (pn) })} defobj Energy-Neutron ENERGY { (volue(amount): cuminfl{ (Energy-Neutron.omount)(distrJbeutr) ((sum((i + ei/2) * distrLneutr[i]) i = 0..8) })} defabj Energy-Gamma ENERGY { (volue (omount): cuminfl{ (Energy~Gomma.omount)(distr-E,gommo) ((sum((i + ei/2) * distr,E-gammo[i]) i = 0..8) })} def_obj Fl PRODUCT-NUCLIDES ( (value: cuminfl { (Fl) (nuclidesJ1) ((nuclidesJl[i,j]) i = O..169,j = 0..108) })} defnbj F2 PRODUCT-NUCLIDES ( (value: cuminfl ( (F2) (nuclidesJ1) ((nuclides-F2[i,j]) i = O..iSS,j = 0..108) })} def-obj Energy-F1 ENERGY { (volue(amount): cuminfl { (Energy-Fl.amount) (distr-GFl) ((sum((i + ei/2) * distrLFl[i])) i = 0..90)} defabj Energy-F2 ENERGY ((volue(omount): cuminfl { (Energy-F2.omount) (distr_E_F2) ((sum((i + ei/2) * disk-E-F2[i])) i = 0..90))

Figure 3: Aggregate effects model for fission. reagents, and their types are NUCLIDE and PARTICLE. These types are encoded in classes NUCLIDE and PARTICLE. The reaction occurs when 01 and 02 collide.
l

Activityconditions describe conditions which are necessary for processes to be active. Whenever the conditions are violated, the model is no longer an adequate description of the process. These conditions depend on the state on the system. This field limits the values of variables. It specifies either absolute limits on one variable, or limits relative to two variables. Parametric-reactions

field describes a set of pa-

342

Proceedings of the 11th Conference on Artificial Intelligence for Applications (CAIA '95) 1043-0989/95 $10.00 1995 IEEE

rameters which are necessary for a system being modelled.

l

Intermediatestate field describes actions which occur before the reaction generates a set of products. For example, the intermediatestate of a fission reaction is a state where the projectile particle and the target nucleus perform neutron absorption to create a compound nucleus. Products describes a set of products generated as a consequence of an active reaction. The do scriptions of products are characterized by types of particles or nuclidea being produced along with the descriptions of the quantities which characterize them.

Products field specifies the creation of objects which come into being as a consequence of an active aggregate effects model. The aggregate effects model becomes active if the associated basic nuclear physics model is active. This causes a set of new objects to be created as a set of products. These created objects are products which are stated in the basic nuclear physica model.

The values of the quantities may be probabilistic or de terministic depending on the type of reaction itself. If the value of quantity is PROBABILISTIC, the distribution along with its parameters must be defined in the model. 2.1.2 Aggregate effects model

The products field in the basic nuclear physics model gives the amount specifications of products being generated in a reaction. If the basic nuclear physics model of a specified reaction is active, the basic nuclear physics model generates a set of products symbolically, and their amounts are either PROBABILISTIC or DETERMINISTIC values. The PROBABILISTIC values are represented in a vector, while the deterministic values are represented by either integer or real values. A statement in the products field in the aggregate effects model does creation of objects which come into being as a consequence of the instance becoming active over period of time. Each attribute of the object is obtained by any influence specified in the field. 2.2 The Symbolic level

An aggregate effects model is used to simulate nuclear physics processes. It consists of conditions for when a simulation occurs, and the effects of this on both a state space and a probabilistic space model describing the process. In addition, the aggregate model can aggregate a set of products as a consequence of an active process. The aggregate effects model describes nuclear process interactions which incorporate dynamics, including any descriptions of influences which affect parameters of the system over time (e.g once per second). It is characterized by fields material, equipments, preconditions, activityconditions, relations, and products such as shown in Fig.3.
l

Material field describes material being used in a reaction being modelled; it may be the same as one of the reactants in the basic model. Equipment field specifies a place where a reaction is to be considered, along with its components. Preconditions describe conditions which may depend on external actions. Turning a power switch on or off is an example. No reaction occurs unless the power is turned on, but, once on, the power is toggled depending the state on the system. Relations field specifies relationships to be associated with each instance of the definitions. The relationships are expressed as many influences, such aa aggregate influences, cumulative influences, distribution influences, and parametric influences.

The symbolic level consists of objects, relations and properties for these objects which are defined in the input description such as shown in Fig. 6. The symbolic level also consists of objects representing instances of basic nuclear physics definitions and aggregate effects definitions. The system identifies every set of objects which when assigned to reactants of the basic nuclear physics model, satisfy the reactant conditions. The system also identifies every basic nuclear physics model which, when assigned to the aggregate effects model, satisfies the aggregate effects model conditions. For each set of objects which is bound to the reactants of the basic nuclear physics model, instances of the basic nuclear physics model are created. Some of these instance are then bound to the aggregate effects model which in turn instantiates the aggregate effects model. For example, consider a set of objects defined in the input descriptions which are shown in Fig. 6. These objects are considered to be active. The subset of instantiations of these objects, U2.95 and Nedron are bound to reactants of the basic nuclear physics model of fission and radioactive-capture. The relation COLLIDE (U235, Neutron) matches the reactant s condition apecified at the reactant fields of the basic nuclear physics model fission and radioactive caplure. This causes instances of the basic nuclear physics model, bnpm fission (3235, Neutron) and bnpm radioactive caplure (U235,

343

Proceedings of the 11th Conference on Artificial Intelligence for Applications (CAIA '95) 1043-0989/95 $10.00 1995 IEEE

11295 Neutron

RC HW
HW

BNPM fission (U235,Neutmn)

BNPM mdioactive,capture (UZSS, Neutron)

AEM (BNPMjission (lJ23S,Neutron),RC,HW,Hw) AEM (BNPM mdioactive-capture(U235, Neutron),RC, HW,Hw) Figure 4: Symbolic instances of basic nuclear and aggregate effects model. Neutron), to be created. These instantiations along with some instantiations of basic objects RC, H W, Hw, are bound to the aggregate effects models of fission and radioaciiue caplure. This causes the instanstiation of the aggregate effects models,AEM (BNPM fission (17.995, Neuiron),RC,HW,Hw) and AEM ( BNPM radioadiue,capiure(lS, Neutron), RC, H W, Hw). Fig.4 shows a subset of objects which are bound to the reactants of the basic nuclear physics model which causes them to be bound to the aggregate effects models for a set of reactions. This figure also shows a subset of instances of the basic nuclear physics model and aggregate effects model. 2.3

Quantitative

level

An instance of the basic nuclear physics model and aggregate effects model does not cause any analysis of the system. Only active instances can cause the analysis. The active instances are obtained when the objects are bound to the reactants, the basic nuclear model is bound to the aggregate effects model, and objects satisfy the preconditions and quantity conditions for both basic nuclear physics and aggregate effects models. After instantiation of basic nuclear and aggregate effects models, the next step of reasoning is to identify the active instances. The basic objects are always considered active, e.g. U2.95, Neutron, RC, H W,Hw. The set of active basic nuclear physics and aggregate effects models are set to follow phenomena reaction in the current state. For example, we have specified the fuel amount of reactor (RCJuelamount) is greater than 0 and U235 cro8s section fission (ol.sf) is greater than 0. This causes only BNPM fission and AEM fission to be active. Since the aggregate effects model describes nuclear interactions every neutron cycle, then by combining all the influences defined by these instances, the system generates objects describing the system state model ev-

ery period of time. Fig. 5 shows a state space model and probabilistic space model which are generated as a consequence of the basic nuclear physics and aggregate effects models being active. The state space model consists of equations which are considered to consist of terms. Each instantiation of the basic nuclear physics model and the aggregate effects model comprises a set of equations which describes how a nuclear process relates to the underlying equations. The probabilistic space model consists of different types of distributions which govern the process. Each instantiation of the basic nuclear physics model and the aggregate effects model comprises a set of parameters which is considered to have non-deterministic values, along with their probabilistic distributions. These distributions constitute a sufficient conceptual basis for obtaining non-deterministic values from the process being modeled. For example, the number of neutrons produced for each fission is non-deterministic; it ranges between 0 to 8. The probability of producing 0 - 8 neutrons for each fission is controlled by the Poisson distribution. This distribution is represented as pa,neutr (see Fig. 5). The distribution of neutrons produced for fissions is dependent on the rate of reaction Rc and the pdf,neutr. The distribution of neutron products is represented by distraeutr. If the state of the system changes Rc, the distribution of product neutrons, dish-neutr, will change as well. The state space model described above are formalized by all influences of all instantiated of the active processes. In general, influences express how one or a set of variables influence another variable or another set of variables. There are several types of influences in NPT. Two of them, dynamic and algebraic influences, are similar to HPT influences. Their syntax specifications are described in [8,9, lo]. Other influences, unique to NPT, are described in [5].

Example

This section illustrates how NPT works. Nuclear reactions under consideration are fission reactions. The basic nuclear physics model and the aggregate effects model definitions are given in Fig. 2 and Fig. 3. The input descriptions for NPT program are shown in Fig.6. Assuming that all relevant basic nuclear physics and aggregate effects models are included in the system, NPT will identify all possible instantiations of the basic nuclear physics and aggregate effects models. This task is accomplished when a set of existing objects match

344

Proceedings of the 11th Conference on Artificial Intelligence for Applications (CAIA '95) 1043-0989/95 $10.00 1995 IEEE

Neutron.amount is PROBABILISTIC pdj-neutr Fl.Al is PROBABILISTIC pdjJ1 delta.2 is PROBABILISTIC pdf,deltaZ Energy-Gamma.amount is PROBABILISTIC pdjJLgamm0 Energyneutr.omount is PROBABILISTIC p&!Lneut EnergQl is PROBABILISTIC pdj-Fl Energy-F2 is PROBABILISTIC pdjJ2 01 is U235 Fissile is 01 02 is Neutron k = RC.eta * RC.epsilon * RC.p * f * RC.PNLj * RC * PNLth Rc = (k /nu) * pn B = (Rc /NA ) * Fissi1e.A C = (Fissile.8 j + Fissi1e.s-g) / Fissi1e.s j)

U235 NUCLIDE Z A N omount *J *-9 1

( 92 295 lOe21/* otdms per cm3 */ 100 /* gr */ 585 /* born */ 99 /* km */

Neutron PARTICLE { Z 0 A 1 omount Se20 1 HW MODERATOR ( 0.883 /* barn */ s-0 N 3.679e22 /* atoms per cm3 */ 1 Hw CLADDING

* B

RC.jueLomount = RC.jueLomount - C
disk,neutr dish-A 1 distrA2 distr-E,neutr d&r-E-F1 distr-gommo dish-F1 dish-F2 distr&gommo d&--E-F2

LO
Ncl 1 RC REACTOR corn epsilon (

( 0.66 /* ban */ 2.013e23 /* atoms per cm3 */

products PARTICLE Neutron Neutron.omount = (sum(i

* distr,neutrfij ))i

= 0..8

eta
Fuel PNLth PNLf 1

PRODUCT,NUCLIDES Fl (distr-AI, Fl.omount = isotopesdistribution pdj-deltoZ, camp,nut. Z, comp,nuc.A) PRODUCT,NUCLIDES F2 Fl.omount = isotopesdistribution (distr42, pdj-deltoZ, comp,nuc.Z, wmp,nuc.A) ENERGY EnergyNeutron EnerggJeutron.omount = (sum((i+ei/2) * dirtr-E,neutr[i]))i=0..8 ENERGY Energy,Gommo EnergkGommo.omount = (sum((i+ei/2) * diskE-gammo[i]))i=0..8 ENERGY Energy&

homogeneous 2.22 /* thermal jkion factor */ 1.115 /*fast fission factor */ 0.667 /* nsononce erwpe probability U235 0.833 0.865

*/

COLLIDE

(U235, Neutron)

Figure 6: An example of input descriptions for NPT.

the requirements for the reactants and the material as stated in the basic nuclear physics and aggregate effects of a number of reaction definitions are considered as conceivable nuclear reactions. These may cause the generation of a set of products described symbolically. The instances of objects, basic nuclear physics and aggregate effects models of a fission reaction are shown symbolically similar to Fig. 4. When all possible nuclear reactions have been created, the program proceeds to test the validity of the activity conditions for each instance. This requires that the initial values for at least a subset of parameters and variables have been specified. The result of this process is a set of active processes, which is a subset of the
models. The instantiation

EnergrQi.omount
(aum((i+ei/2)

* disk-E-Fl[i]))i=0..90

ENERGY Energy-F2 EnergQ 2,omount = (sum((i+ei/2) * distr-GF2[i])i=0..90 Figure 5: State model of NPT. space model and probabilistic space

345

Proceedings of the 11th Conference on Artificial Intelligence for Applications (CAIA '95) 1043-0989/95 $10.00 1995 IEEE

conceivable nuclear reactions. The instantiations of active processes will generate a set of products which is expressed either in deterministic values or probabilistic values. The state space and probabilistic models always follow directly from the set of active processes. These are the consequences of a number of influences defined by these instances. If the complete initial state is known, then both state space and probabilistic models can be used to determine the behavior of the system. An example with input shown in Fig. 6, along with the results of running a simulation after 10 seconds is discussed in [5]. These results include distribution of neutron products, distribution of energy of neutron products, energy of gamma products, distribution of mass number of fission fragments for light nuclei (Al), and heavy nuclei (A2), the distribution of fission fragments for light nuclei (Fl), and heavy nuclei (F2) along with the distribution of isotopes, the energy of fission fragments, and total energy produced by fissions.

Using Dimensional Analysis, gence vol.45, 1990, pp.73-111.

Artificial

Intelli-

PI deKleer,

J.D and Brown, J.S. A qualitative physics based on confluences, Artificial Intelligence, ~01.24, 1984, pp.7-83.
Process The-

[31 Forbus, Kenneth D. The Qualitative

ory, Ph.D Thesis, Department of Computer Science, Massachusetts Institute of Technology, 1984.

PI Hartati,

S. Reasoning about Physical Systems in Artificial Intelligence, Technical Report TR93080, Faculty of Computer Science, University of New Brunswick, 1993.

PI Hartati,

S. Nuclear Process Theory: A Symbolic Model for Representing Nuclear Physics Proceases, Technical Report TR94-087, Faculty of Computer Science, University of New Brunswick, October, 1994. causality:

El Kuipers,
4 Concluding remarks
This paper describes a formal knowledge representation for nuclear physics processes, called NPT (Nuclear Processs Theory). A simple nuclear process simulator based on NPT has been developed. The introduction of cummulative, aggregate and parametric influences should be useful in other domains where reasoning about complex physical processes is required. Probabilistic representation of processes is also unique to NPT, and may prove useful in reasoning about other processes requiring non-deterministic models (e.g quality control). It remains to implement a complete set of computer-based tools which can perform a simulation based on NPT models. The real challenges ahead lies in developing a set of basic nuclear physics and aggregate effects models which are as universal as possible.

B. Commonsense reasoning about deriving behavior structure, Arlificial Intelligence, ~01.24, 1984, pp.169-203. Simulation, vol.29, pp.289-388, 1986.
Artificial In-

171Kuipers, B. Qualitative
telligence,

PI Woods, E. The Hybrid Phenomena Theory, Proceedings of the Twelfth International Conference on Artificial Intelligence, Sydney, Australia, Aug 24-

30, 1991, pp.1138-1142.

PI Woods,

E. and Balchen, J.G. Structural Estimation with the hybrid phenomena theory, Proceedings of the 3rd IFAC Workshop, California, USA, September 23-25, 1991, pp.127-132.

[lo] Woods,

E.The Hybrid Phenomena Theory A Framework Integrating Structural Descriptions with State Space Modeling and Simulation, Ph.D

Acknowledgements
The authors would like to thank Dr. George DeMille of the UNB Department of Physics for his suggestions. This research is partly funded by the Higher Education Development Project of the Republic of Indonesia, and by the UNB Faculty of Computer Science.

Thesis, Division of Engineering Cybernetics, wegian Institute of Technology, Trondheim, way, 1993.

NorNor-

References
[l] Bhaskar, R. and Nigam, A. Qualitative Physics

346

Proceedings of the 11th Conference on Artificial Intelligence for Applications (CAIA '95) 1043-0989/95 $10.00 1995 IEEE